index
int64
0
0
repo_id
stringlengths
26
205
file_path
stringlengths
51
246
content
stringlengths
8
433k
__index_level_0__
int64
0
10k
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/tree/CutTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static org.hamcrest.MatcherAssert.assertThat; import static org.hamcrest.Matchers.is; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertTrue; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class CutTest { private int splitDimension; private double splitValue; private Cut cut; @BeforeEach public void setUp() { splitDimension = 2; splitValue = 3.4; cut = new Cut(splitDimension, splitValue); } @Test public void testNew() { assertThat(cut.getDimension(), is(splitDimension)); assertThat(cut.getValue(), is(splitValue)); } @Test public void testIsLeftOf() { double[] point = new double[] { 1.0, 2.0, 3.0, 4.0 }; assertTrue(Cut.isLeftOf(point, cut)); point[2] = 99.9; assertFalse(Cut.isLeftOf(point, cut)); } }
400
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/tree/RandomCutTreeTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.toDoubleArray; import static com.amazon.randomcutforest.CommonUtils.toFloatArray; import static com.amazon.randomcutforest.tree.AbstractNodeStore.Null; import static org.hamcrest.MatcherAssert.assertThat; import static org.hamcrest.Matchers.closeTo; import static org.hamcrest.Matchers.is; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertNotEquals; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.ArgumentMatchers.any; import static org.mockito.ArgumentMatchers.anyInt; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.when; import java.util.ArrayList; import java.util.List; import java.util.Random; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.ValueSource; import com.amazon.randomcutforest.config.Config; import com.amazon.randomcutforest.sampler.Weighted; import com.amazon.randomcutforest.store.PointStore; public class RandomCutTreeTest { private static final double EPSILON = 1e-8; private Random rng; private RandomCutTree tree; @BeforeEach public void setUp() { rng = mock(Random.class); PointStore pointStoreFloat = new PointStore.Builder().indexCapacity(100).capacity(100).initialSize(100) .dimensions(2).build(); tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStoreFloat) .storeSequenceIndexesEnabled(true).storeParent(true).dimension(2).build(); // Create the following tree structure (in the second diagram., backticks denote // cuts) // The leaf point 0,1 has mass 2, all other nodes have mass 1. // // /\ // / \ // -1,-1 / \ // / \ // /\ 1,1 // / \ // -1,0 0,1 // // // 0,1 1,1 // ----------*---------* // | ` | ` | // | ` | ` | // | ` | ` | // -1,0 *-------------------| // | | // |```````````````````| // | | // -1,-1 *-------------------- // // We choose the insertion order and random draws carefully so that each split // divides its parent in half. // The random values are used to set the cut dimensions and values. assertEquals(pointStoreFloat.add(new float[] { -1, -1 }, 1), 0); assertEquals(pointStoreFloat.add(new float[] { 1, 1 }, 2), 1); assertEquals(pointStoreFloat.add(new float[] { -1, 0 }, 3), 2); assertEquals(pointStoreFloat.add(new float[] { 0, 1 }, 4), 3); assertEquals(pointStoreFloat.add(new float[] { 0, 1 }, 5), 4); assertEquals(pointStoreFloat.add(new float[] { 0, 0 }, 6), 5); assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(0, 1)); tree.addPoint(0, 1); tree.deletePoint(0, 1); assertTrue(tree.root == Null); tree.addPoint(0, 1); when(rng.nextDouble()).thenReturn(0.625); tree.addPoint(1, 2); when(rng.nextDouble()).thenReturn(0.5); tree.addPoint(2, 3); when(rng.nextDouble()).thenReturn(0.25); tree.addPoint(3, 4); // add mass to 0,1 tree.addPoint(4, 5); assertArrayEquals(tree.liftFromTree(new float[] { 17, 18 }), new float[] { 17, 18 }); } @Test public void testConfig() { assertThrows(IllegalArgumentException.class, () -> tree.setBoundingBoxCacheFraction(-0.5)); assertThrows(IllegalArgumentException.class, () -> tree.setBoundingBoxCacheFraction(2.0)); assertThrows(IllegalArgumentException.class, () -> tree.setConfig("foo", 0)); assertThrows(IllegalArgumentException.class, () -> tree.getConfig("bar")); assertEquals(tree.getConfig(Config.BOUNDING_BOX_CACHE_FRACTION), 1.0); assertThrows(IllegalArgumentException.class, () -> tree.setConfig(Config.BOUNDING_BOX_CACHE_FRACTION, true)); assertThrows(IllegalArgumentException.class, () -> tree.getConfig(Config.BOUNDING_BOX_CACHE_FRACTION, boolean.class)); tree.setConfig(Config.BOUNDING_BOX_CACHE_FRACTION, 0.2); } @Test public void testConfigStore() { assertEquals(tree.nodeStore.isLeaf(-1), tree.isLeaf(-1)); assertEquals(tree.nodeStore.isLeaf(256), tree.isLeaf(256)); assertEquals(tree.nodeStore.isInternal(-1), tree.isInternal(-1)); assertEquals(tree.nodeStore.isInternal(0), tree.isInternal(0)); assertEquals(tree.nodeStore.isInternal(255), tree.isInternal(255)); assertEquals(tree.nodeStore.isInternal(256), tree.isInternal(256)); } @Test public void testParent() { PointStore pointStore = mock(PointStore.class); tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStore) .storeSequenceIndexesEnabled(true).storeParent(false).dimension(3).build(); assertThrows(IllegalArgumentException.class, () -> tree.nodeStore.getParentIndex(tree.root)); } @Test public void testConfigDelete() { PointStore pointStore = mock(PointStore.class); tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStore) .storeSequenceIndexesEnabled(true).storeParent(true).dimension(3).build(); when(pointStore.getNumericVector(any(Integer.class))).thenReturn(new float[] { 0 }).thenReturn(new float[3]) .thenReturn(new float[3]); tree.addPoint(0, 1); // fails vor dimension assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(0, 1)); assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(2, 1)); // wrong sequence index assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(0, 2)); // state is corrupted assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(0, 1)); } @Test public void testConfigAdd() { PointStore pointStore = mock(PointStore.class); float[] test = new float[] { 1.119f, 0f, -3.11f, 100f }; float[] copies = new float[] { 0, 17, 0, 0 }; tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStore) .centerOfMassEnabled(true).storeSequenceIndexesEnabled(true).storeParent(true).dimension(4).build(); when(pointStore.getNumericVector(any(Integer.class))).thenReturn(new float[0]).thenReturn(test) .thenReturn(new float[62]).thenReturn(new float[4]).thenReturn(new float[17]).thenReturn(new float[4]) .thenReturn(new float[4]).thenReturn(new float[5]).thenReturn(copies).thenReturn(test) .thenReturn(copies).thenReturn(copies).thenReturn(test); // cannot have partial addition to empty tree assertThrows(IllegalArgumentException.class, () -> tree.addPointToPartialTree(0, 1)); // the following does not consume any points tree.addPoint(0, 1); // consumes from pointstore but gets 0 length vector assertThrows(IllegalArgumentException.class, () -> tree.getPointSum(tree.getRoot())); // passes, consumes pointstore assertArrayEquals(tree.getPointSum(tree.getRoot()), test); // sequel fails because dimension is 62 assertThrows(IllegalArgumentException.class, () -> tree.getBox(tree.root)); // in the sequel point is [0,0,0,0] fails because old point appears to have 17 // dimensions assertThrows(IllegalArgumentException.class, () -> tree.addPoint(1, 1)); // this invocation succeeds, but points are same tree.addPoint(1, 1); assertTrue(tree.isLeaf(tree.getRoot())); // dimension = 5 assertThrows(IllegalArgumentException.class, () -> tree.addPoint(2, 1)); // switch the vector assertArrayEquals(tree.getPointSum(tree.getRoot()), new float[] { 0, 34, 0, 0 }); // adding test, consumes the copy tree.addPoint(2, 1); assertEquals(tree.getMass(), 3); assertArrayEquals(tree.getPointSum(tree.getRoot()), new float[] { 1.119f, 34, -3.11f, 100 }, 1e-3f); // bounding boxes are incorrect they are minvalues = test, maxvalues = test assertThrows(IllegalStateException.class, () -> tree.validateAndReconstruct(tree.root)); assertTrue(tree.getCutDimension(tree.root) == 3); // cut cannot be the same as right minvalue tree.nodeStore.cutValue[tree.root] = 100; assertThrows(IllegalStateException.class, () -> tree.validateAndReconstruct(tree.root)); } @Test public void testConfigPartialAdd() { PointStore pointStore = mock(PointStore.class); float[] test = new float[] { 1.119f, 0f, -3.11f, 100f }; float[] copies = new float[] { 0, 17, 0, 0 }; tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStore) .centerOfMassEnabled(true).storeSequenceIndexesEnabled(true).storeParent(true).dimension(4).build(); when(pointStore.getNumericVector(any(Integer.class))).thenReturn(new float[0]).thenReturn(test) .thenReturn(new float[0]).thenReturn(test).thenReturn(new float[4]).thenReturn(new float[5]) .thenReturn(copies).thenReturn(test).thenReturn(copies).thenReturn(copies).thenReturn(test); // the following does not consume any points tree.addPoint(0, 1); assertThrows(IllegalArgumentException.class, () -> tree.addPointToPartialTree(1, 1)); // fails at check of dimension of retrieved point assertThrows(IllegalArgumentException.class, () -> tree.addPointToPartialTree(1, 1)); // fails at equality check assertThrows(IllegalArgumentException.class, () -> tree.addPointToPartialTree(1, 1)); } @Test public void testCut() { PointStore pointStore = mock(PointStore.class); Random random = mock(Random.class); tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStore).random(random) .storeSequenceIndexesEnabled(true).storeParent(true).dimension(1).build(); when(pointStore.getNumericVector(any(Integer.class))).thenReturn(new float[1]).thenReturn(new float[] { 1 }) .thenReturn(new float[] { 0 }).thenReturn(new float[] { 0 }).thenReturn(new float[] { 2 }) .thenReturn(new float[] { 1 }).thenReturn(new float[0]).thenReturn(new float[] { 2 }) .thenReturn(new float[] { 1 }).thenReturn(new float[1]); // testing the cut assumptions -- the values should not be 1 or larger, but is // useful for testing when(random.nextDouble()).thenReturn(1.2).thenReturn(1.5).thenReturn(1.5).thenReturn(0.0); // following does not query pointstore tree.addPoint(0, 1); // following tries to add [0.0], and discovers point index 0 is [1.0] tree.addPoint(1, 1); assertTrue(tree.getCutValue(tree.getRoot()) == (double) Math.nextAfter(1.0f, 0.0)); assertThrows(IllegalArgumentException.class, () -> tree.addPoint(1, 2)); // copy tree.addPoint(1, 2); // passes assertTrue(tree.getRoot() == 0); assertTrue(tree.getCutValue(0) == (double) Math.nextAfter(1.0f, 0.0)); assertTrue(tree.getCutValue(1) == (double) Math.nextAfter(2.0f, 1.0)); assertFalse(tree.checkStrictlyContains(1, new float[] { 2 })); assertTrue(tree.checkStrictlyContains(1, new float[] { 1.001f })); } /** * Verify that the tree has the form described in the setUp method. */ @Test public void testInitialTreeState() { int node = tree.getRoot(); // the second double[] is intentional IBoundingBoxView expectedBox = new BoundingBox(new float[] { -1, -1 }).getMergedBox(new float[] { 1, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(1)); assertThat(tree.getCutValue(node), closeTo(-0.5, EPSILON)); assertThat(tree.getMass(), is(5)); assertArrayEquals(new double[] { -1, 2 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, -1 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(1L), 1); // testing inappropriate assertThrows(IllegalArgumentException.class, () -> tree.getLeftChild(Integer.MAX_VALUE)); assertThrows(IllegalArgumentException.class, () -> tree.getRightChild(500)); assertThrows(IllegalArgumentException.class, () -> tree.getCutValue(-1)); assertThrows(IllegalArgumentException.class, () -> tree.getCutDimension(-1)); // pointIndex should have a value at least as large as number of leaves assertThrows(IllegalArgumentException.class, () -> tree.getPointIndex(0)); NodeStoreSmall nodeStoreSmall = (NodeStoreSmall) tree.nodeStore; assert (nodeStoreSmall.getParentIndex(tree.getRightChild(node)) == node); node = tree.getRightChild(node); expectedBox = new BoundingBox(new float[] { -1, 0 }).getMergedBox(new BoundingBox(new float[] { 1, 1 })); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(0)); assertThat(tree.getCutValue(node), closeTo(0.5, EPSILON)); assertThat(tree.getMass(node), is(4)); assertArrayEquals(new double[] { 0.0, 3.0 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getRightChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getRightChild(node))), is(new float[] { 1, 1 })); assertThat(tree.getMass(tree.getRightChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(2L), 1); assert (nodeStoreSmall.getParentIndex(tree.getLeftChild(node)) == node); node = tree.getLeftChild(node); expectedBox = new BoundingBox(new float[] { -1, 0 }).getMergedBox(new float[] { 0, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(0)); assertThat(tree.getCutValue(node), closeTo(-0.5, EPSILON)); assertThat(tree.getMass(node), is(3)); assertArrayEquals(new double[] { -1.0, 2.0 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, 0 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(3L), 1); assertThat(tree.isLeaf(tree.getRightChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getRightChild(node))), is(new float[] { 0, 1 })); assertThat(tree.getMass(tree.getRightChild(node)), is(2)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(4L), 1); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(5L), 1); assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(5, 6)); } @Test public void testDeletePointWithLeafSibling() { tree.deletePoint(2, 3); // root node bounding box and cut remains unchanged, mass and centerOfMass are // updated int node = tree.getRoot(); IBoundingBoxView expectedBox = new BoundingBox(new float[] { -1, -1 }).getMergedBox(new float[] { 1, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(1)); assertThat(tree.getCutValue(node), closeTo(-0.5, EPSILON)); assertThat(tree.getMass(), is(4)); assertArrayEquals(new double[] { 0.0, 2.0 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, -1 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(1L), 1); // sibling node moves up and bounding box recomputed NodeStoreSmall nodeStoreSmall = (NodeStoreSmall) tree.nodeStore; assert (nodeStoreSmall.getParentIndex(tree.getRightChild(node)) == node); node = tree.getRightChild(node); expectedBox = new BoundingBox(new float[] { 0, 1 }).getMergedBox(new float[] { 1, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(0)); assertThat(tree.getCutValue(node), closeTo(0.5, EPSILON)); assertThat(tree.getMass(node), is(3)); assertArrayEquals(new double[] { 1.0, 3.0 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { 0, 1 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(2)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(4L), 1); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(5L), 1); assertThat(tree.isLeaf(tree.getRightChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getRightChild(node))), is(new float[] { 1, 1 })); assertThat(tree.getMass(tree.getRightChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(2L), 1); } @Test public void testDeletePointWithNonLeafSibling() { tree.deletePoint(1, 2); // root node bounding box recomputed int node = tree.getRoot(); IBoundingBoxView expectedBox = new BoundingBox(new float[] { -1, -1 }).getMergedBox(new float[] { 0, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(1)); assertThat(tree.getCutValue(node), closeTo(-0.5, EPSILON)); assertThat(tree.getMass(), is(4)); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, -1 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(1L), 1); // sibling node moves up and bounding box stays the same NodeStoreSmall nodeStoreSmall = (NodeStoreSmall) tree.nodeStore; assert (nodeStoreSmall.getParentIndex(tree.getRightChild(node)) == node); node = tree.getRightChild(node); expectedBox = new BoundingBox(new float[] { -1, 0 }).getMergedBox(new float[] { 0, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(0)); assertThat(tree.getCutValue(node), closeTo(-0.5, EPSILON)); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, 0 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(3L), 1); assertThat(tree.isLeaf(tree.getRightChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getRightChild(node))), is(new float[] { 0, 1 })); assertThat(tree.getMass(tree.getRightChild(node)), is(2)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(4L), 1); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(5L), 1); } @Test public void testDeletePointWithMassGreaterThan1() { assertTrue(tree.boundingBoxCacheFraction == 1.0); tree.setConfig(Config.BOUNDING_BOX_CACHE_FRACTION, 0.5); assertTrue(tree.boundingBoxData != null); assertTrue(tree.boundingBoxData.length == ((tree.numberOfLeaves - 1) / 2) * 4); assertTrue(tree.rangeSumData != null); assertTrue(tree.rangeSumData.length == (tree.numberOfLeaves - 1) / 2); int root = tree.getRoot(); assertTrue(tree.checkStrictlyContains(root, new float[2])); tree.setConfig(Config.BOUNDING_BOX_CACHE_FRACTION, 0.0); assertTrue(tree.boundingBoxData == null); assertTrue(tree.rangeSumData == null); assertFalse(tree.checkStrictlyContains(root, new float[2])); tree.deletePoint(3, 4); tree.setConfig(Config.BOUNDING_BOX_CACHE_FRACTION, 0.5); assertTrue(tree.boundingBoxData != null); assertTrue(tree.boundingBoxData.length == ((tree.numberOfLeaves - 1) / 2) * 4); assertTrue(tree.rangeSumData != null); assertTrue(tree.rangeSumData.length == (tree.numberOfLeaves - 1) / 2); // same as initial state except mass at 0,1 is 1 int node = tree.getRoot(); IBoundingBoxView expectedBox = new BoundingBox(new float[] { -1, -1 }).getMergedBox(new float[] { 1, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(1)); assertThat(tree.getCutValue(node), closeTo(-0.5, EPSILON)); assertThat(tree.getMass(), is(4)); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, -1 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(1L), 1); assertArrayEquals(new double[] { -1.0, 1.0 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, -1 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(1L), 1); node = tree.getRightChild(node); expectedBox = new BoundingBox(new float[] { -1, 0 }).getMergedBox(new float[] { 1, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertThat(tree.getCutDimension(node), is(0)); assertThat(tree.getCutValue(node), closeTo(0.5, EPSILON)); assertThat(tree.getMass(node), is(3)); NodeView nodeView = new NodeView(tree, tree.pointStoreView, node); assertTrue(nodeView.getCutDimension() == 0); assertTrue(nodeView.getCutValue() == 0.5); assertArrayEquals(new double[] { 0.0, 2.0 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getRightChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getRightChild(node))), is(new float[] { 1, 1 })); assertThat(tree.getMass(tree.getRightChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(2L), 1); NodeStoreSmall nodeStoreSmall = (NodeStoreSmall) tree.nodeStore; assert (nodeStoreSmall.getParentIndex(tree.getLeftChild(node)) == node); node = tree.getLeftChild(node); expectedBox = new BoundingBox(new float[] { -1, 0 }).getMergedBox(new float[] { 0, 1 }); assertThat(tree.getBox(node), is(expectedBox)); assertEquals(expectedBox.toString(), tree.getBox(node).toString()); assertThat(tree.getCutDimension(node), is(0)); assertThat(tree.getCutValue(node), closeTo(-0.5, EPSILON)); assertThat(tree.getMass(), is(4)); assertArrayEquals(new double[] { -1.0, 1.0 }, toDoubleArray(tree.getPointSum(node)), EPSILON); assertThat(tree.isLeaf(tree.getLeftChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getLeftChild(node))), is(new float[] { -1, 0 })); assertThat(tree.getMass(tree.getLeftChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getLeftChild(node))).get(3L), 1); assertThat(tree.isLeaf(tree.getRightChild(node)), is(true)); assertThat(tree.pointStoreView.getNumericVector(tree.getPointIndex(tree.getRightChild(node))), is(new float[] { 0, 1 })); assertThat(tree.getMass(tree.getRightChild(node)), is(1)); assertEquals(tree.getSequenceMap(tree.getPointIndex(tree.getRightChild(node))).get(5L), 1); } @Test public void testDeletePointInvalid() { // specified sequence index does not exist assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(2, 99)); // point does not exist in tree assertThrows(IllegalArgumentException.class, () -> tree.deletePoint(7, 3)); } @Test public void testUpdatesOnSmallBoundingBox() { // verifies on small bounding boxes random cuts and tree updates are functional PointStore pointStoreFloat = new PointStore.Builder().indexCapacity(10).capacity(10).currentStoreCapacity(10) .dimensions(1).build(); RandomCutTree tree = RandomCutTree.builder().random(rng).pointStoreView(pointStoreFloat).build(); List<Weighted<double[]>> points = new ArrayList<>(); points.add(new Weighted<>(new double[] { 48.08 }, 0, 1L)); points.add(new Weighted<>(new double[] { 48.08001 }, 0, 2L)); pointStoreFloat.add(toFloatArray(points.get(0).getValue()), 0); pointStoreFloat.add(toFloatArray(points.get(1).getValue()), 1); tree.addPoint(0, points.get(0).getSequenceIndex()); tree.addPoint(1, points.get(1).getSequenceIndex()); assertNotEquals(pointStoreFloat.getNumericVector(0)[0], pointStoreFloat.getNumericVector(1)[0]); for (int i = 0; i < 10000; i++) { Weighted<double[]> point = points.get(i % points.size()); tree.deletePoint(i % points.size(), point.getSequenceIndex()); tree.addPoint(i % points.size(), point.getSequenceIndex()); } } @Test public void testfloat() { float x = 110.13f; double sum = 0; int trials = 230000; for (int i = 0; i < trials; i++) { float z = (x * (trials - i + 1) - x); sum += z; } System.out.println(sum); for (int i = 0; i < trials - 1; i++) { float z = (x * (trials - i + 1) - x); sum -= z; } System.out.println(sum + " " + (double) x + " " + (sum <= (double) x)); float[] possible = new float[trials]; float[] alsoPossible = new float[trials]; for (int i = 0; i < trials; i++) { possible[i] = x; alsoPossible[i] = (trials - i + 1) * x; } BoundingBox box = new BoundingBox(possible, alsoPossible); System.out.println("rangesum " + box.getRangeSum()); double factor = 1.0 - 1e-16; System.out.println(factor); RandomCutTree tree = RandomCutTree.builder().dimension(trials).build(); // tries both path tree.randomCut(factor, possible, box); tree.randomCut(1.0 - 1e-17, possible, box); } @ParameterizedTest @ValueSource(ints = { 100, 10000, 100000 }) void testNodeStore(int size) { PointStore pointStoreFloat = new PointStore.Builder().indexCapacity(100).capacity(100).initialSize(100) .dimensions(2).build(); tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStoreFloat) .capacity(size).storeSequenceIndexesEnabled(true).storeParent(true).dimension(2).build(); long seed = new Random().nextLong(); System.out.println("seed :" + seed); Random rng = new Random(seed); for (int i = 0; i < 100; i++) { pointStoreFloat.add(new double[] { rng.nextDouble(), rng.nextDouble() }, 0L); } ArrayList<Weighted<Integer>> list = new ArrayList<>(); for (int i = 0; i < 100; i++) { tree.addPoint(i, 0L); list.add(new Weighted<>(i, rng.nextFloat(), 0)); } list.sort((o1, o2) -> Float.compare(o1.getWeight(), o2.getWeight())); for (int i = 0; i < 50; i++) { tree.deletePoint(list.remove(0).getValue(), 0L); } AbstractNodeStore nodeStore = tree.getNodeStore(); for (int i = 0; i < 25; i++) { if (!tree.isLeaf(tree.getLeftChild(tree.getRoot()))) { assert (nodeStore.getParentIndex(tree.getLeftChild(tree.getRoot())) == tree.root); } if (!tree.isLeaf(tree.getRightChild(tree.getRoot()))) { assert (nodeStore.getParentIndex(tree.getRightChild(tree.getRoot())) == tree.root); } tree.deletePoint(list.remove(0).getValue(), 0L); } } // spoofs the cut (using a changing box) to hit illegal state @Test public void cutTest1() { BoundingBox box1 = mock(BoundingBox.class); when(box1.getMinValue(anyInt())).thenReturn(0.0).thenReturn(0.0).thenReturn(1.0); assertThrows(IllegalStateException.class, () -> tree.randomCut(1.2, new float[] { 1.0f }, box1)); } // spoofs the cut (usina a changing box) to hit illegal state @Test public void cutTest2() { BoundingBox box1 = mock(BoundingBox.class); when(box1.getMinValue(anyInt())).thenReturn(0.0).thenReturn(0.0).thenReturn(1.0); assertThrows(IllegalStateException.class, () -> tree.randomCut(1.5, new float[] { 1.0f }, box1)); } @Test public void cutTestMultiD() { float[] point = new float[2]; float[] newPoint = new float[] { 0.1f + new Random().nextFloat(), 0.1f + new Random().nextFloat() }; float[] testPoint = new float[] { point[0], newPoint[1] }; float[] testPoint2 = new float[] { newPoint[0], point[1] }; BoundingBox box1 = new BoundingBox(point, point); BoundingBox box2 = new BoundingBox(newPoint, newPoint); assertThrows(IllegalArgumentException.class, () -> tree.randomCut(new Random().nextDouble(), point, box1)); assertDoesNotThrow(() -> tree.randomCut(new Random().nextDouble(), point, box2)); assertDoesNotThrow(() -> tree.randomCut(new Random().nextDouble(), newPoint, box1)); Cut cut1 = tree.randomCut(0, new float[] { 0, 1.0f }, box1); // first dimension is identical assertTrue(cut1.getDimension() == 1); assertTrue(cut1.getValue() == 0f); assertEquals(cut1.toString(), "Cut(1, 0.000000)"); Cut cut2 = tree.randomCut(1.2, point, box2); assertTrue(cut2.getDimension() == 0); assertTrue(cut2.getValue() == Math.nextAfter(newPoint[0], point[0])); Cut largeCut = tree.randomCut(1.2, newPoint, box1); assertTrue(largeCut.getDimension() == 0); assertTrue(largeCut.getValue() == Math.nextAfter(newPoint[0], point[0])); Cut testCut = tree.randomCut(1.2, testPoint, box2); assertTrue(testCut.getDimension() == 0); assertTrue(testCut.getValue() == Math.nextAfter(newPoint[0], testPoint[0])); Cut testCut2 = tree.randomCut(1.2, testPoint2, box2); assertTrue(testCut2.getDimension() == 1); assertTrue(testCut2.getValue() == Math.nextAfter(newPoint[1], point[1])); Cut another = tree.randomCut(1.5, point, box2); assertTrue(another.getDimension() == 1); assertTrue(another.getValue() == Math.nextAfter(newPoint[1], point[1])); Cut anotherLargeCut = tree.randomCut(1.5, newPoint, box1); assertTrue(anotherLargeCut.getDimension() == 1); assertTrue(anotherLargeCut.getValue() == Math.nextAfter(newPoint[1], point[1])); Cut anotherTestCut = tree.randomCut(1.5, testPoint, box1); assertTrue(testCut.getDimension() == 0); assertTrue(testCut.getValue() == Math.nextAfter(newPoint[0], point[0])); Cut anotherTestCut2 = tree.randomCut(1.5, testPoint2, box1); assertTrue(testCut2.getDimension() == 1); assertTrue(testCut2.getValue() == Math.nextAfter(newPoint[1], point[1])); } // the following are tested directly since they are unreachable @Test public void traverseTest() { PointStore pointStoreFloat = new PointStore.Builder().indexCapacity(100).capacity(100).initialSize(100) .dimensions(2).build(); tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStoreFloat) .capacity(188).storeSequenceIndexesEnabled(true).storeParent(true).dimension(2).build(); assertDoesNotThrow(() -> tree.validateAndReconstruct()); assertThrows(IllegalArgumentException.class, () -> tree.traverse(null, null)); assertThrows(IllegalArgumentException.class, () -> tree.traverseMulti(null, null)); } @Test public void invalidNodeTest() { PointStore pointStoreFloat = new PointStore.Builder().indexCapacity(100).capacity(100).initialSize(100) .dimensions(2).build(); tree = RandomCutTree.builder().random(rng).centerOfMassEnabled(true).pointStoreView(pointStoreFloat) .capacity(188).storeSequenceIndexesEnabled(true).storeParent(true).dimension(2).build(); tree.root = 187; assertThrows(IllegalStateException.class, () -> tree.validateAndReconstruct()); assertThrows(IllegalStateException.class, () -> tree.traversePathToLeafAndVisitNodes(null, null, null, tree.root, 0)); assertThrows(IllegalStateException.class, () -> tree.traverseTreeMulti(null, null, null, tree.root, 0)); assertThrows(IllegalStateException.class, () -> tree.growNodeBox(null, pointStoreFloat, 0, 187)); assertThrows(IllegalStateException.class, () -> tree.getBox(187)); } }
401
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/tree/BoxCacheTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static org.junit.jupiter.api.Assertions.assertEquals; import java.util.Random; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.RandomCutForestTest; import com.amazon.randomcutforest.config.Precision; public class BoxCacheTest { @Test public void testChangingBoundingBoxFloat32() { int dimensions = 4; int numberOfTrees = 1; int sampleSize = 64; int dataSize = 1000 * sampleSize; Random random = new Random(); long seed = random.nextLong(); double[][] big = RandomCutForestTest.generateShingledData(dataSize, dimensions, 2); RandomCutForest forest = RandomCutForest.builder().compact(true).dimensions(dimensions) .numberOfTrees(numberOfTrees).sampleSize(sampleSize).precision(Precision.FLOAT_32).randomSeed(seed) .boundingBoxCacheFraction(0).build(); RandomCutForest otherForest = RandomCutForest.builder().compact(true).dimensions(dimensions) .numberOfTrees(numberOfTrees).sampleSize(sampleSize).precision(Precision.FLOAT_32).randomSeed(seed) .boundingBoxCacheFraction(1).build(); int num = 0; for (double[] point : big) { ++num; if (num % sampleSize == 0) { forest.setBoundingBoxCacheFraction(random.nextDouble()); } assertEquals(forest.getAnomalyScore(point), otherForest.getAnomalyScore(point)); forest.update(point); otherForest.update(point); } } }
402
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/tree/BoundingBoxTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.hamcrest.MatcherAssert.assertThat; import static org.hamcrest.Matchers.closeTo; import static org.hamcrest.Matchers.is; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class BoundingBoxTest { private float[] point1; private float[] point2; private BoundingBox box1; private BoundingBox box2; @BeforeEach public void setUp() { point1 = new float[] { 1.5f, 2.7f }; point2 = new float[] { 3.0f, 1.2f }; box1 = new BoundingBox(point1); box2 = new BoundingBox(point2); } @Test public void dimensionTest() { assertThrows(IllegalArgumentException.class, () -> new BoundingBox(point1, new float[1])); assertThrows(IllegalArgumentException.class, () -> box1.getMergedBox(new float[1])); assertThrows(IllegalArgumentException.class, () -> box1.contains(new float[1])); assertThrows(IllegalArgumentException.class, () -> box1.contains(new BoundingBox(new float[1]))); } @Test public void equalsTest() { assertFalse(box1.equals(point1)); assertFalse(box1.equals(box2)); assertFalse(box1.equals(new BoundingBox(point1, new float[] { 3.0f, 2.7f }))); assertTrue(box1.equals(box1.copy())); } @Test public void testNewFromSinglePoint() { assertThat(box1.getDimensions(), is(2)); assertThat((float) box1.getMinValue(0), is(point1[0])); assertThat((float) box1.getMaxValue(0), is(point1[0])); assertThat(box1.getRange(0), is(0.0)); assertThat((float) box1.getMinValue(1), is(point1[1])); assertThat((float) box1.getMaxValue(1), is(point1[1])); assertThat(box1.getRange(1), is(0.0)); assertThat(box1.getRangeSum(), is(0.0)); assertThat(box2.getDimensions(), is(2)); assertThat((float) box2.getMinValue(0), is(point2[0])); assertThat((float) box2.getMaxValue(0), is(point2[0])); assertThat(box2.getRange(0), is(0.0)); assertThat((float) box2.getMinValue(1), is(point2[1])); assertThat((float) box2.getMaxValue(1), is(point2[1])); assertThat(box2.getRange(1), is(0.0)); assertThat(box2.getRangeSum(), is(0.0)); assertTrue(box1.probabilityOfCut(point2) == 1.0); assertTrue(box1.probabilityOfCut(point1) == 0.0); } @Test public void testGetMergedBoxWithOtherBox() { assertThrows(IllegalStateException.class, () -> box1.addBox(box2)); assertThrows(IllegalArgumentException.class, () -> box1.addPoint(new float[1])); assertThrows(IllegalArgumentException.class, () -> box1.addPoint(new float[2])); assertDoesNotThrow(() -> box1.copy().addPoint(new float[2])); BoundingBox mergedBox = box1.getMergedBox(box2); assertThat(mergedBox.getDimensions(), is(2)); assertThat((float) mergedBox.getMinValue(0), is(1.5f)); assertThat((float) mergedBox.getMaxValue(0), is(3.0f)); assertThat(mergedBox.getRange(0), closeTo(3.0 - 1.5, EPSILON)); assertThat((float) mergedBox.getMinValue(1), is(1.2f)); assertThat((float) mergedBox.getMaxValue(1), is(2.7f)); assertThat(mergedBox.getRange(1), closeTo(2.7 - 1.2, EPSILON)); double rangeSum = (3.0 - 1.5) + (2.7 - 1.2); assertThat(mergedBox.getRangeSum(), closeTo(rangeSum, EPSILON)); // check that box1 and box2 were not changed assertThat(box1.getDimensions(), is(2)); assertThat((float) box1.getMinValue(0), is(point1[0])); assertThat((float) box1.getMaxValue(0), is(point1[0])); assertThat(box1.getRange(0), is(0.0)); assertThat((float) box1.getMinValue(1), is(point1[1])); assertThat((float) box1.getMaxValue(1), is(point1[1])); assertThat(box1.getRange(1), is(0.0)); assertThat(box1.getRangeSum(), is(0.0)); assertThat(box2.getDimensions(), is(2)); assertThat((float) box2.getMinValue(0), is(point2[0])); assertThat((float) box2.getMaxValue(0), is(point2[0])); assertThat(box2.getRange(0), is(0.0)); assertThat((float) box2.getMinValue(1), is(point2[1])); assertThat((float) box2.getMaxValue(1), is(point2[1])); assertThat(box2.getRange(1), is(0.0)); assertThat(box2.getRangeSum(), is(0.0)); } @Test public void testContainsBoundingBox() { BoundingBox box1 = new BoundingBox(new float[] { 0.0f, 0.0f }) .getMergedBox(new BoundingBox(new float[] { 10.0f, 10.0f })); BoundingBox box2 = new BoundingBox(new float[] { 2.0f, 2.0f }) .getMergedBox(new BoundingBox(new float[] { 8.0f, 8.0f })); BoundingBox box3 = new BoundingBox(new float[] { -4.0f, -4.0f }) .getMergedBox(new BoundingBox(new float[] { -1.0f, -1.0f })); BoundingBox box4 = new BoundingBox(new float[] { -1.0f, -1.0f }) .getMergedBox(new BoundingBox(new float[] { 5.0f, 5.0f })); // completely contains assertTrue(box1.contains(box2)); assertFalse(box2.contains(box1)); // completely disjoint assertFalse(box1.contains(box3)); assertFalse(box3.contains(box1)); // partially intersect assertFalse(box1.contains(box4)); assertFalse(box4.contains(box1)); } @Test public void testContainsPoint() { BoundingBox box1 = new BoundingBox(new float[] { 0.0f, 0.0f }) .getMergedBox(new BoundingBox(new float[] { 10.0f, 10.0f })); assertTrue(box1.contains(new float[] { 0.0f, 0.1f })); assertTrue(box1.contains(new float[] { 5.5f, 6.5f })); assertFalse(box1.contains(new float[] { -0.7f, -4.5f })); assertFalse(box1.contains(new float[] { 5.0f, 11.0f })); assertFalse(box1.contains(new float[] { -5.0f, 10.0f })); } }
403
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/anomalydetection/DynamicScoreVisitorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import static org.junit.jupiter.api.Assertions.assertEquals; import java.util.function.BiFunction; import org.junit.jupiter.api.Test; public class DynamicScoreVisitorTest { @Test public void testScoringMethods() { BiFunction<Double, Double, Double> scoreSeen = (x, y) -> (x + y) / 2; BiFunction<Double, Double, Double> scoreUneen = (x, y) -> 0.75 * x + 0.25 * y; BiFunction<Double, Double, Double> damp = (x, y) -> Math.sqrt(x * y); DynamicScoreVisitor visitor = new DynamicScoreVisitor(new float[] { 1.1f, -2.2f }, 100, 2, scoreSeen, scoreUneen, damp); int x = 9; int y = 4; assertEquals((x + y) / 2.0, visitor.scoreSeen(x, y)); assertEquals(0.75 * x + 0.25 * y, visitor.scoreUnseen(x, y)); assertEquals(Math.sqrt(x * y), visitor.damp(x, y)); } }
404
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/anomalydetection/AnomalyAttributionVisitorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import static com.amazon.randomcutforest.CommonUtils.defaultScalarNormalizerFunction; import static com.amazon.randomcutforest.CommonUtils.defaultScoreUnseenFunction; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.ArgumentMatchers.any; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.when; import org.junit.jupiter.api.Test; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.ValueSource; import com.amazon.randomcutforest.CommonUtils; import com.amazon.randomcutforest.returntypes.DiVector; import com.amazon.randomcutforest.tree.BoundingBox; import com.amazon.randomcutforest.tree.INodeView; import com.amazon.randomcutforest.tree.NodeView; public class AnomalyAttributionVisitorTest { @Test public void testNew() { float[] point = new float[] { 1.1f, -2.2f, 3.3f }; int treeMass = 99; AnomalyAttributionVisitor visitor = new AnomalyAttributionVisitor(point, treeMass); assertFalse(visitor.pointInsideBox); for (int i = 0; i < point.length; i++) { assertFalse(visitor.coordInsideBox[i]); } assertFalse(visitor.ignoreLeaf); assertEquals(0, visitor.ignoreLeafMassThreshold); DiVector result = visitor.getResult(); double[] zero = new double[point.length]; assertArrayEquals(zero, result.high); assertArrayEquals(zero, result.low); } @Test public void testNewWithIgnoreOptions() { float[] point = new float[] { 1.1f, -2.2f, 3.3f }; int treeMass = 99; AnomalyAttributionVisitor visitor = new AnomalyAttributionVisitor(point, treeMass, 7); assertFalse(visitor.pointInsideBox); for (int i = 0; i < point.length; i++) { assertFalse(visitor.coordInsideBox[i]); } assertTrue(visitor.ignoreLeaf); assertEquals(7, visitor.ignoreLeafMassThreshold); DiVector result = visitor.getResult(); double[] zero = new double[point.length]; assertArrayEquals(zero, result.high); assertArrayEquals(zero, result.low); } @Test public void testAcceptLeafEquals() { float[] point = { 1.1f, -2.2f, 3.3f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafDepth = 100; int leafMass = 10; when(leafNode.getMass()).thenReturn(leafMass); int treeMass = 21; AnomalyAttributionVisitor visitor = new AnomalyAttributionVisitor(point, treeMass, 0); visitor.acceptLeaf(leafNode, leafDepth); assertTrue(visitor.hitDuplicates); assertEquals(visitor.sumOfNewRange, 0); double expectedScoreSum = CommonUtils.defaultDampFunction(leafMass, treeMass) / (leafDepth + Math.log(leafMass + 1) / Math.log(2)); double expectedScore = expectedScoreSum / (2 * point.length); DiVector result = visitor.getResult(); for (int i = 0; i < point.length; i++) { assertEquals(defaultScalarNormalizerFunction(expectedScore, treeMass), result.low[i], EPSILON); assertEquals(defaultScalarNormalizerFunction(expectedScore, treeMass), result.high[i], EPSILON); } } @Test public void testAcceptLeafNotEquals() { float[] point = new float[] { 1.1f, -2.2f, 3.3f }; float[] anotherPoint = new float[] { -4.0f, 5.0f, 6.0f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(anotherPoint); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(anotherPoint, anotherPoint)); int leafDepth = 100; int leafMass = 4; when(leafNode.getMass()).thenReturn(leafMass); int treeMass = 21; AnomalyAttributionVisitor visitor = new AnomalyAttributionVisitor(point, treeMass, 0); visitor.acceptLeaf(leafNode, leafDepth); double expectedScoreSum = defaultScoreUnseenFunction(leafDepth, leafMass); double sumOfNewRange = (1.1 - (-4.0)) + (5.0 - (-2.2)) + (6.0 - 3.3); DiVector result = visitor.getResult(); assertEquals(defaultScalarNormalizerFunction(expectedScoreSum * (1.1 - (-4.0)) / sumOfNewRange, treeMass), result.high[0], EPSILON); assertEquals(0.0, result.low[0]); assertEquals(0.0, result.high[1]); assertEquals(defaultScalarNormalizerFunction(expectedScoreSum * (5.0 - (-2.2)) / sumOfNewRange, treeMass), result.low[1], EPSILON); assertEquals(0.0, result.high[2]); assertEquals(defaultScalarNormalizerFunction(expectedScoreSum * (6.0 - 3.3) / sumOfNewRange, treeMass), result.low[2], EPSILON); visitor = new AnomalyAttributionVisitor(point, treeMass, 3); visitor.acceptLeaf(leafNode, leafDepth); result = visitor.getResult(); assertEquals(defaultScalarNormalizerFunction(expectedScoreSum * (1.1 - (-4.0)) / sumOfNewRange, treeMass), result.high[0], EPSILON); assertEquals(0.0, result.low[0]); assertEquals(0.0, result.high[1]); assertEquals(defaultScalarNormalizerFunction(expectedScoreSum * (5.0 - (-2.2)) / sumOfNewRange, treeMass), result.low[1], EPSILON); assertEquals(0.0, result.high[2]); assertEquals(defaultScalarNormalizerFunction(expectedScoreSum * (6.0 - 3.3) / sumOfNewRange, treeMass), result.low[2], EPSILON); visitor = new AnomalyAttributionVisitor(point, treeMass, 4); visitor.acceptLeaf(leafNode, leafDepth); double expectedScore = expectedScoreSum / (2 * point.length); result = visitor.getResult(); for (int i = 0; i < point.length; i++) { assertEquals(defaultScalarNormalizerFunction(expectedScore, treeMass), result.low[i], EPSILON); assertEquals(defaultScalarNormalizerFunction(expectedScore, treeMass), result.high[i], EPSILON); } } @Test public void testAccept() { float[] pointToScore = { 0.0f, 0.0f }; int treeMass = 50; AnomalyAttributionVisitor visitor = new AnomalyAttributionVisitor(pointToScore, treeMass, 0); INodeView leafNode = mock(NodeView.class); float[] point = new float[] { 1.0f, -2.0f }; when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafMass = 3; when(leafNode.getMass()).thenReturn(leafMass); int depth = 4; visitor.acceptLeaf(leafNode, depth); DiVector result = visitor.getResult(); double expectedScoreSum = defaultScoreUnseenFunction(depth, leafNode.getMass()); double sumOfNewRange = 1.0 + 2.0; double[] expectedUnnormalizedLow = new double[] { expectedScoreSum * 1.0 / sumOfNewRange, 0.0 }; double[] expectedUnnormalizedHigh = new double[] { 0.0, expectedScoreSum * 2.0 / sumOfNewRange }; for (int i = 0; i < pointToScore.length; i++) { assertEquals(defaultScalarNormalizerFunction(expectedUnnormalizedLow[i], treeMass), result.low[i], EPSILON); assertEquals(defaultScalarNormalizerFunction(expectedUnnormalizedHigh[i], treeMass), result.high[i], EPSILON); } // parent does not contain pointToScore depth--; INodeView sibling = mock(NodeView.class); int siblingMass = 2; when(sibling.getMass()).thenReturn(siblingMass); INodeView parent = mock(NodeView.class); int parentMass = leafMass + siblingMass; when(parent.getMass()).thenReturn(parentMass); BoundingBox boundingBox = new BoundingBox(point, new float[] { 2.0f, -0.5f }); when(parent.getBoundingBox()).thenReturn(boundingBox); visitor.accept(parent, depth); result = visitor.getResult(); double expectedSumOfNewRange2 = 2.0 + 2.0; double expectedProbOfCut2 = (1.0 + 0.5) / expectedSumOfNewRange2; double[] expectedDifferenceInRangeVector2 = { 0.0, 1.0, 0.5, 0.0 }; double expectedScore2 = defaultScoreUnseenFunction(depth, parent.getMass()); double[] expectedUnnormalizedLow2 = new double[pointToScore.length]; double[] expectedUnnormalizedHigh2 = new double[pointToScore.length]; for (int i = 0; i < pointToScore.length; i++) { double prob = expectedDifferenceInRangeVector2[2 * i] / expectedSumOfNewRange2; expectedUnnormalizedHigh2[i] = prob * expectedScore2 + (1 - expectedProbOfCut2) * expectedUnnormalizedHigh[i]; prob = expectedDifferenceInRangeVector2[2 * i + 1] / expectedSumOfNewRange2; expectedUnnormalizedLow2[i] = prob * expectedScore2 + (1 - expectedProbOfCut2) * expectedUnnormalizedLow[i]; } for (int i = 0; i < pointToScore.length; i++) { assertEquals(defaultScalarNormalizerFunction(expectedUnnormalizedLow2[i], treeMass), result.low[i], EPSILON); assertEquals(defaultScalarNormalizerFunction(expectedUnnormalizedHigh2[i], treeMass), result.high[i], EPSILON); } // grandparent contains pointToScore assertFalse(visitor.pointInsideBox); depth--; INodeView grandParent = mock(NodeView.class); when(grandParent.getMass()).thenReturn(parentMass + 2); when(grandParent.getBoundingBox()).thenReturn(boundingBox .getMergedBox(new BoundingBox(new float[] { -1.0f, 1.0f }).getMergedBox(new float[] { -0.5f, -1.5f }))); visitor.accept(grandParent, depth); result = visitor.getResult(); for (int i = 0; i < pointToScore.length; i++) { assertEquals(defaultScalarNormalizerFunction(expectedUnnormalizedLow2[i], treeMass), result.low[i], EPSILON); assertEquals(defaultScalarNormalizerFunction(expectedUnnormalizedHigh2[i], treeMass), result.high[i], EPSILON); } } @ParameterizedTest @ValueSource(ints = { 3, 5 }) public void reNormalizeNotEqual(int mass) { float[] pointToScore = { 0.0f, 0.0f }; int treeMass = 50; AnomalyAttributionVisitor visitor = new AnomalyAttributionVisitor(pointToScore, treeMass, 4); INodeView leafNode = mock(NodeView.class); float[] point = new float[] { 1.0f, -2.0f }; when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafMass = mass; when(leafNode.getMass()).thenReturn(leafMass); visitor.acceptLeaf(leafNode, 1); INodeView parent = mock(NodeView.class); int parentMass = leafMass + 2; when(parent.getMass()).thenReturn(parentMass); BoundingBox boundingBox = new BoundingBox(point, new float[] { 2.0f, 2.0f }); when(parent.getBoundingBox()).thenReturn(boundingBox); when(parent.getSiblingBoundingBox(any())).thenReturn(new BoundingBox(new float[] { 2.0f, 2.0f })); visitor.accept(parent, 0); DiVector result = visitor.directionalAttribution; assertEquals(result.getHighLowSum(), visitor.savedScore, 1e-6); } @ParameterizedTest @ValueSource(ints = { 3, 5 }) public void reNormalize(int mass) { float[] pointToScore = { 0.0f, 0.0f }; int treeMass = 50; AnomalyAttributionVisitor visitor = new AnomalyAttributionVisitor(pointToScore, treeMass, 4); INodeView leafNode = mock(NodeView.class); float[] point = pointToScore; when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafMass = mass; when(leafNode.getMass()).thenReturn(leafMass); visitor.acceptLeaf(leafNode, 1); INodeView parent = mock(NodeView.class); int parentMass = leafMass + 2; when(parent.getMass()).thenReturn(parentMass); BoundingBox boundingBox = new BoundingBox(point, new float[] { 2.0f, 2.0f }); when(parent.getBoundingBox()).thenReturn(boundingBox); when(parent.getSiblingBoundingBox(any())).thenReturn(new BoundingBox(new float[] { 2.0f, 2.0f })); visitor.accept(parent, 0); DiVector result = visitor.directionalAttribution; assertEquals(result.getHighLowSum(), visitor.savedScore, 1e-6); } }
405
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/anomalydetection/AnomalyScoreVisitorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static com.amazon.randomcutforest.tree.AbstractNodeStore.Null; import static org.hamcrest.MatcherAssert.assertThat; import static org.hamcrest.Matchers.closeTo; import static org.hamcrest.Matchers.is; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertNotEquals; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.ArgumentMatchers.any; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.when; import java.util.Arrays; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.CommonUtils; import com.amazon.randomcutforest.tree.BoundingBox; import com.amazon.randomcutforest.tree.IBoundingBoxView; import com.amazon.randomcutforest.tree.INodeView; import com.amazon.randomcutforest.tree.NodeView; public class AnomalyScoreVisitorTest { @Test public void testNew() { float[] point = new float[] { 1.0f, 2.0f }; int sampleSize = 9; AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(point, sampleSize); assertFalse(visitor.pointInsideBox); for (int i = 0; i < point.length; i++) { assertFalse(visitor.coordInsideBox[i]); } assertFalse(visitor.ignoreLeafEquals); assertEquals(0, visitor.ignoreLeafMassThreshold); assertThat(visitor.getResult(), is(0.0)); } @Test public void testNewWithIgnoreOptions() { float[] point = new float[] { 1.0f, 2.0f }; int sampleSize = 9; AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(point, sampleSize, 7); assertFalse(visitor.pointInsideBox); for (int i = 0; i < point.length; i++) { assertFalse(visitor.coordInsideBox[i]); } assertTrue(visitor.ignoreLeafEquals); assertEquals(7, visitor.ignoreLeafMassThreshold); assertThat(visitor.getResult(), is(0.0)); } @Test public void testAcceptLeafEquals() { float[] point = { 1.0f, 2.0f, 3.0f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafDepth = 100; int leafMass = 10; when(leafNode.getMass()).thenReturn(leafMass); int subSampleSize = 21; AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(point, subSampleSize); visitor.acceptLeaf(leafNode, leafDepth); double expectedScore = CommonUtils.defaultDampFunction(leafMass, subSampleSize) / (leafDepth + Math.log(leafMass + 1) / Math.log(2)); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, subSampleSize), EPSILON)); assertTrue(visitor.pointInsideBox); visitor = new AnomalyScoreVisitor(point, subSampleSize); visitor.acceptLeaf(leafNode, 0); expectedScore = CommonUtils.defaultDampFunction(leafMass, subSampleSize) / (Math.log(leafMass + 1) / Math.log(2.0)); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, subSampleSize), EPSILON)); assertTrue(visitor.pointInsideBox); AnomalyScoreVisitor anotherVisitor = new AnomalyScoreVisitor(point, subSampleSize, 7); anotherVisitor.acceptLeaf(leafNode, 0); assertEquals(anotherVisitor.score, visitor.score); AnomalyScoreVisitor yetAnotherVisitor = new AnomalyScoreVisitor(point, subSampleSize, 12); yetAnotherVisitor.acceptLeaf(leafNode, 0); assertNotEquals(yetAnotherVisitor.score, visitor.score); } @Test public void testAcceptLeafNotEquals() { float[] point = new float[] { 1.0f, 2.0f, 3.0f }; float[] anotherPoint = new float[] { 4.0f, 5.0f, 6.0f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(anotherPoint); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(anotherPoint, anotherPoint)); int leafDepth = 100; AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(point, 2); visitor.acceptLeaf(leafNode, leafDepth); double expectedScore = 1.0 / (leafDepth + 1); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, 2), EPSILON)); assertFalse(visitor.pointInsideBox); int leafMass = 10; when(leafNode.getMass()).thenReturn(leafMass); AnomalyScoreVisitor anotherVisitor = new AnomalyScoreVisitor(point, 2, 7); anotherVisitor.acceptLeaf(leafNode, 100); assertEquals(anotherVisitor.score, visitor.score); AnomalyScoreVisitor yetAnotherVisitor = new AnomalyScoreVisitor(point, 2, 12); yetAnotherVisitor.acceptLeaf(leafNode, 100); assertEquals(yetAnotherVisitor.score, visitor.score); } @Test public void testAcceptEqualsLeafPoint() { float[] pointToScore = { 0.0f, 0.0f }; int sampleSize = 50; AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(pointToScore, sampleSize); float[] point = Arrays.copyOf(pointToScore, pointToScore.length); INodeView node = mock(NodeView.class); when(node.getLeafPoint()).thenReturn(point); when(node.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int depth = 2; visitor.acceptLeaf(node, depth); double expectedScore = CommonUtils.defaultDampFunction(node.getMass(), sampleSize) / (depth + Math.log(node.getMass() + 1) / Math.log(2)); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, sampleSize), EPSILON)); depth--; IBoundingBoxView boundingBox = node.getBoundingBox().getMergedBox(new float[] { 1.0f, 1.0f }); node = new NodeView(null, null, Null); visitor.accept(node, depth); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, sampleSize), EPSILON)); depth--; boundingBox = boundingBox.getMergedBox(new float[] { -1.0f, -1.0f }); node = new NodeView(null, null, Null); visitor.accept(node, depth); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, sampleSize), EPSILON)); } @Test public void testAccept() { float[] pointToScore = new float[] { 0.0f, 0.0f }; int sampleSize = 50; AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(pointToScore, sampleSize); NodeView node = mock(NodeView.class); float[] otherPoint = new float[] { 1.0f, 1.0f }; when(node.getLeafPoint()).thenReturn(otherPoint); when(node.getBoundingBox()).thenReturn(new BoundingBox(otherPoint, otherPoint)); int depth = 4; visitor.acceptLeaf(node, depth); double expectedScore = 1.0 / (depth + 1); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, sampleSize), EPSILON)); depth--; IBoundingBoxView boundingBox = node.getBoundingBox().getMergedBox(new float[] { 2.0f, 0.0f }); when(node.getBoundingBox()).thenReturn(boundingBox); when(node.probailityOfSeparation(any())).thenReturn(1.0 / 3); visitor.accept(node, depth); double p = visitor.getProbabilityOfSeparation(boundingBox); expectedScore = p * (1.0 / (depth + 1)) + (1 - p) * expectedScore; assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, sampleSize), EPSILON)); depth--; boundingBox = boundingBox.getMergedBox(new float[] { -1.0f, 0.0f }); when(node.getBoundingBox()).thenReturn(boundingBox); when(node.probailityOfSeparation(any())).thenReturn(0.0); visitor.accept(node, depth); p = visitor.getProbabilityOfSeparation(boundingBox); expectedScore = p * (1.0 / (depth + 1)) + (1 - p) * expectedScore; assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, sampleSize), EPSILON)); depth--; boundingBox = boundingBox.getMergedBox(new float[] { -1.0f, -1.0f }); when(node.probailityOfSeparation(any())).thenReturn(0.0); visitor.accept(node, depth); p = visitor.getProbabilityOfSeparation(boundingBox); assertThat(visitor.getResult(), closeTo(CommonUtils.defaultScalarNormalizerFunction(expectedScore, sampleSize), EPSILON)); assertTrue(visitor.pointInsideBox); } @Test public void testGetProbabilityOfSeparation() { float[] minPoint = { 0.0f, 0.0f, 0.0f }; float[] maxPoint = { 1.0f, 2.0f, 3.0f }; IBoundingBoxView boundingBox = new BoundingBox(minPoint); boundingBox = boundingBox.getMergedBox(maxPoint); float[] point = { 0.5f, 0.5f, 0.5f }; int sampleSize = 2; AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(point, sampleSize); double p = visitor.getProbabilityOfSeparation(boundingBox); assertThat(p, closeTo(0.0, EPSILON)); assertTrue(visitor.coordInsideBox[0]); assertTrue(visitor.coordInsideBox[1]); assertTrue(visitor.coordInsideBox[2]); visitor = new AnomalyScoreVisitor(point, sampleSize); visitor.coordInsideBox[1] = visitor.coordInsideBox[2] = true; p = visitor.getProbabilityOfSeparation(boundingBox); assertThat(p, closeTo(0.0, EPSILON)); assertTrue(visitor.coordInsideBox[0]); assertTrue(visitor.coordInsideBox[1]); assertTrue(visitor.coordInsideBox[2]); point = new float[] { 2.0f, 0.5f, 0.5f }; visitor = new AnomalyScoreVisitor(point, sampleSize); p = visitor.getProbabilityOfSeparation(boundingBox); assertThat(p, closeTo(1.0 / (2.0 + 2.0 + 3.0), EPSILON)); assertFalse(visitor.coordInsideBox[0]); assertTrue(visitor.coordInsideBox[1]); assertTrue(visitor.coordInsideBox[2]); visitor = new AnomalyScoreVisitor(point, sampleSize); visitor.coordInsideBox[1] = visitor.coordInsideBox[2] = true; p = visitor.getProbabilityOfSeparation(boundingBox); assertThat(p, closeTo(1.0 / (2.0 + 2.0 + 3.0), EPSILON)); assertFalse(visitor.coordInsideBox[0]); assertTrue(visitor.coordInsideBox[1]); assertTrue(visitor.coordInsideBox[2]); point = new float[] { 0.5f, -3.0f, 4.0f }; visitor = new AnomalyScoreVisitor(point, sampleSize); p = visitor.getProbabilityOfSeparation(boundingBox); assertThat(p, closeTo((3.0 + 1.0) / (1.0 + 5.0 + 4.0), EPSILON)); assertTrue(visitor.coordInsideBox[0]); assertFalse(visitor.coordInsideBox[1]); assertFalse(visitor.coordInsideBox[2]); visitor = new AnomalyScoreVisitor(point, sampleSize); visitor.coordInsideBox[0] = true; p = visitor.getProbabilityOfSeparation(boundingBox); assertThat(p, closeTo((3.0 + 1.0) / (1.0 + 5.0 + 4.0), EPSILON)); assertTrue(visitor.coordInsideBox[0]); assertFalse(visitor.coordInsideBox[1]); assertFalse(visitor.coordInsideBox[2]); } @Test public void test_getProbabilityOfSeparation_leafNode() { float[] point = new float[] { 1.0f, 2.0f, 3.0f }; float[] leafPoint = Arrays.copyOf(point, point.length); BoundingBox boundingBox = new BoundingBox(leafPoint); AnomalyScoreVisitor visitor = new AnomalyScoreVisitor(point, 2); assertThrows(IllegalStateException.class, () -> visitor.getProbabilityOfSeparation(boundingBox)); TransductiveScalarScoreVisitor esotericVisitor = new TransductiveScalarScoreVisitor(leafPoint, 2, CommonUtils::defaultScoreSeenFunction, CommonUtils::defaultScoreUnseenFunction, CommonUtils::defaultDampFunction, b -> new double[3]); assertThrows(IllegalStateException.class, () -> esotericVisitor.getProbabilityOfSeparation(boundingBox)); } }
406
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/anomalydetection/DynamicAttributionVisitorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import static org.junit.jupiter.api.Assertions.assertEquals; import java.util.function.BiFunction; import org.junit.jupiter.api.Test; public class DynamicAttributionVisitorTest { @Test public void testScoringMethods() { BiFunction<Double, Double, Double> scoreSeen = (x, y) -> (x + y) / 2; BiFunction<Double, Double, Double> scoreUneen = (x, y) -> 0.75 * x + 0.25 * y; BiFunction<Double, Double, Double> damp = (x, y) -> Math.sqrt(x * y); DynamicAttributionVisitor visitor = new DynamicAttributionVisitor(new float[] { 1.1f, -2.2f }, 100, 2, scoreSeen, scoreUneen, damp); int x = 9; int y = 4; assertEquals((x + y) / 2.0, visitor.scoreSeen(x, y)); assertEquals(0.75 * x + 0.25 * y, visitor.scoreUnseen(x, y)); assertEquals(Math.sqrt(x * y), visitor.damp(x, y)); } }
407
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/inspect/NearNeighborVisitorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.inspect; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertNotSame; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.when; import java.util.Arrays; import java.util.HashMap; import java.util.List; import java.util.Map; import java.util.Optional; import java.util.stream.Collector; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.returntypes.Neighbor; import com.amazon.randomcutforest.tree.INodeView; import com.amazon.randomcutforest.tree.NodeView; public class NearNeighborVisitorTest { private float[] queryPoint; private double distanceThreshold; private NearNeighborVisitor visitor; @BeforeEach public void setUp() { queryPoint = new float[] { 7.7f, 8.8f, -6.6f }; distanceThreshold = 10.0; visitor = new NearNeighborVisitor(queryPoint, distanceThreshold); } @Test public void acceptLeafNear() { float[] leafPoint = new float[] { 8.8f, 9.9f, -5.5f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(Arrays.copyOf(leafPoint, leafPoint.length)); when(leafNode.getLiftedLeafPoint()).thenReturn(Arrays.copyOf(leafPoint, leafPoint.length)); HashMap<Long, Integer> sequenceIndexes = new HashMap<>(); sequenceIndexes.put(1234L, 1); sequenceIndexes.put(5678L, 1); when(leafNode.getSequenceIndexes()).thenReturn(sequenceIndexes); int depth = 12; visitor.acceptLeaf(leafNode, depth); Optional<Neighbor> optional = visitor.getResult(); assertTrue(optional.isPresent()); Neighbor neighbor = optional.get(); assertNotSame(leafPoint, neighbor.point); assertArrayEquals(leafPoint, neighbor.point); assertEquals(Math.sqrt(3 * 1.1 * 1.1), neighbor.distance, EPSILON); assertNotSame(leafNode.getSequenceIndexes(), neighbor.sequenceIndexes); } @Test public void acceptLeafNearTimestampsDisabled() { float[] leafPoint = new float[] { 8.8f, 9.9f, -5.5f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLiftedLeafPoint()).thenReturn(Arrays.copyOf(leafPoint, leafPoint.length)); when(leafNode.getLeafPoint()).thenReturn(Arrays.copyOf(leafPoint, leafPoint.length)); assertEquals(0, leafNode.getSequenceIndexes().size()); int depth = 12; visitor.acceptLeaf(leafNode, depth); Optional<Neighbor> optional = visitor.getResult(); assertTrue(optional.isPresent()); NearNeighborVisitor nearNeighborVisitor = new NearNeighborVisitor(queryPoint); nearNeighborVisitor.acceptLeaf(leafNode, depth); Map<Integer, Neighbor> map1 = new HashMap<>(); Map<Integer, Neighbor> map2 = new HashMap<>(); // an equality test Collector<Optional<Neighbor>, Map<Integer, Neighbor>, List<Neighbor>> collector = Neighbor.collector(); map1.put(Arrays.hashCode(optional.get().point), optional.get()); map2.put(Arrays.hashCode(nearNeighborVisitor.getResult().get().point), optional.get()); collector.combiner().apply(map1, map2); assertEquals(map1.size(), 1); Neighbor neighbor = optional.get(); assertNotSame(leafPoint, neighbor.point); assertArrayEquals(leafPoint, neighbor.point); assertEquals(Math.sqrt(3 * 1.1 * 1.1), neighbor.distance, EPSILON); assertTrue(neighbor.sequenceIndexes.isEmpty()); } @Test public void acceptLeafNotNear() { float[] leafPoint = new float[] { 108.8f, 209.9f, -305.5f }; INodeView leafNode = mock(NodeView.class); HashMap<Long, Integer> sequenceIndexes = new HashMap<>(); sequenceIndexes.put(1234L, 1); sequenceIndexes.put(5678L, 1); when(leafNode.getLeafPoint()).thenReturn(leafPoint); when(leafNode.getLiftedLeafPoint()).thenReturn(leafPoint); when(leafNode.getSequenceIndexes()).thenReturn(sequenceIndexes); int depth = 12; visitor.acceptLeaf(leafNode, depth); Optional<Neighbor> optional = visitor.getResult(); assertFalse(optional.isPresent()); } }
408
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/util/ArrayPackingTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import static com.amazon.randomcutforest.util.ArrayPacking.pack; import static com.amazon.randomcutforest.util.ArrayPacking.unpackDoubles; import static com.amazon.randomcutforest.util.ArrayPacking.unpackFloats; import static com.amazon.randomcutforest.util.ArrayPacking.unpackInts; import static com.amazon.randomcutforest.util.ArrayPacking.unpackShorts; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import java.util.Arrays; import java.util.Random; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.ValueSource; public class ArrayPackingTest { private Random rng; @BeforeEach public void setUp() { rng = new Random(); ArrayPacking arrayPacking = new ArrayPacking(); } @Test public void testLogMax() { long[] bases = new long[] { 2, 101, 3_456_789 }; Arrays.stream(bases).forEach(base -> { int log = ArrayPacking.logMax(base); assertTrue(Math.pow(base, log + 1) >= Integer.MAX_VALUE); assertTrue(Math.pow(base, log) < Integer.MAX_VALUE); }); } @Test public void testLogMaxInvalid() { assertThrows(IllegalArgumentException.class, () -> ArrayPacking.logMax(1)); assertThrows(IllegalArgumentException.class, () -> ArrayPacking.logMax(0)); assertThrows(IllegalArgumentException.class, () -> ArrayPacking.logMax(-123467890)); } @ParameterizedTest @ValueSource(ints = { 0, 1, 2, 3, 11, 100 }) public void testIntsPackRoundTrip(int inputLength) { int[] inputArray = rng.ints().limit(inputLength).toArray(); assertArrayEquals(inputArray, ArrayPacking.unpackInts(ArrayPacking.pack(inputArray, false), false)); assertArrayEquals(inputArray, ArrayPacking.unpackInts(ArrayPacking.pack(inputArray, true), true)); } @ParameterizedTest @ValueSource(ints = { 0, 1, 2, 3, 17, 100 }) public void testShortsPackRoundTrip(int inputLength) { short[] inputArray = new short[inputLength]; for (int i = 0; i < inputLength; i++) { inputArray[i] = (short) (rng.nextInt() % 100); } assertArrayEquals(inputArray, ArrayPacking.unpackShorts(ArrayPacking.pack(inputArray, false), false)); assertArrayEquals(inputArray, ArrayPacking.unpackShorts(ArrayPacking.pack(inputArray, true), true)); } @ParameterizedTest @ValueSource(ints = { 0, 1, 2, 3, 11, 100 }) public void testIdenticalInts(int inputLength) { int[] inputArray = new int[inputLength]; Arrays.fill(inputArray, rng.nextInt()); assertArrayEquals(inputArray, ArrayPacking.unpackInts(ArrayPacking.pack(inputArray, false), false)); int[] result = ArrayPacking.pack(inputArray, true); assertTrue(result.length == 3 || inputLength < 3 && result.length == inputLength); assertArrayEquals(inputArray, ArrayPacking.unpackInts(result, true)); } @ParameterizedTest @ValueSource(ints = { 0, 1, 2, 3, 17, 100 }) public void testIdenticalShorts(int inputLength) { short item = (short) (rng.nextInt() % 100); short[] inputArray = new short[inputLength]; for (int i = 0; i < inputLength; i++) { inputArray[i] = item; } assertArrayEquals(inputArray, ArrayPacking.unpackShorts(ArrayPacking.pack(inputArray, false), false)); int[] result = ArrayPacking.pack(inputArray, true); assertTrue(result.length == 3 || inputLength < 3 && result.length == inputLength); assertArrayEquals(inputArray, ArrayPacking.unpackShorts(result, true)); } @Test public void testUnpackIntsWithLengthGiven() { int inputLength = 100; int[] inputArray = rng.ints().limit(inputLength).toArray(); assertThrows(IllegalArgumentException.class, () -> pack(inputArray, inputLength + 1, false)); assertThrows(IllegalArgumentException.class, () -> pack(inputArray, inputLength + 1, true)); assertThrows(IllegalArgumentException.class, () -> pack(inputArray, -1, false)); assertThrows(IllegalArgumentException.class, () -> pack(inputArray, -1, true)); assertDoesNotThrow(() -> pack(inputArray, 0, true)); assertDoesNotThrow(() -> pack(inputArray, 0, false)); int[] uncompressed = ArrayPacking.pack(inputArray, false); int[] compressed = ArrayPacking.pack(inputArray, true); int[] result = ArrayPacking.unpackInts(uncompressed, 50, false); assertThrows(IllegalArgumentException.class, () -> unpackInts(compressed, -1, true)); assertEquals(50, result.length); assertArrayEquals(Arrays.copyOf(inputArray, 50), result); result = ArrayPacking.unpackInts(compressed, 50, true); assertEquals(50, result.length); assertArrayEquals(Arrays.copyOf(inputArray, 50), result); result = ArrayPacking.unpackInts(uncompressed, 200, false); assertEquals(200, result.length); assertArrayEquals(inputArray, Arrays.copyOf(result, 100)); for (int i = 100; i < 200; i++) { assertEquals(0, result[i]); } result = ArrayPacking.unpackInts(compressed, 200, true); assertEquals(200, result.length); assertArrayEquals(inputArray, Arrays.copyOf(result, 100)); for (int i = 100; i < 200; i++) { assertEquals(0, result[i]); } } @Test public void testUnpackShortsWithLengthGiven() { int inputLength = 100; short[] inputArray = new short[50]; Arrays.fill(inputArray, (short) 2); short[] test = new short[2]; short[] test2 = new short[3]; int[] uncompressed = ArrayPacking.pack(inputArray, false); int[] compressed = ArrayPacking.pack(inputArray, true); assertArrayEquals(test, unpackShorts(new int[2], true)); assertArrayEquals(test, unpackShorts(new int[2], false)); assertArrayEquals(test2, unpackShorts(new int[3], false)); assertThrows(IllegalArgumentException.class, () -> unpackShorts(uncompressed, -1, false)); short[] result = ArrayPacking.unpackShorts(uncompressed, 50, false); assertEquals(50, result.length); assertArrayEquals(Arrays.copyOf(inputArray, 50), result); result = ArrayPacking.unpackShorts(compressed, 100, true); assertEquals(100, result.length); for (int y = 0; y < 50; y++) { assertTrue(result[y] == 2); } for (int y = 50; y < 100; y++) { assertTrue(result[y] == 0); } } @ParameterizedTest @ValueSource(ints = { 0, 1, 2, 3, 17, 100 }) public void testPackDoublesRoundTrip(int inputLength) { double[] inputArray = rng.doubles().limit(inputLength).toArray(); assertArrayEquals(inputArray, ArrayPacking.unpackDoubles(ArrayPacking.pack(inputArray))); } @ParameterizedTest @ValueSource(ints = { 0, 1, 2, 3, 5, 100 }) public void testPackFloatsRoundTrip(int inputLength) { float[] inputArray = new float[inputLength]; for (int i = 0; i < inputLength; i++) { inputArray[i] = rng.nextFloat(); } assertArrayEquals(inputArray, unpackFloats(ArrayPacking.pack(inputArray))); } @ParameterizedTest @ValueSource(booleans = { true, false }) public void testPackShortsWithLength(boolean compress) { int inputLength = 100; int packLength = 76; short[] inputArray = new short[inputLength]; for (int i = 0; i < inputLength; i++) { inputArray[i] = (short) (rng.nextInt() % 100); } assertThrows(IllegalArgumentException.class, () -> pack(inputArray, inputLength + 10, compress)); assertThrows(IllegalArgumentException.class, () -> pack(inputArray, -10, compress)); int[] array = ArrayPacking.pack(inputArray, packLength, compress); short[] outputArray = ArrayPacking.unpackShorts(array, compress); assertEquals(packLength, outputArray.length); assertArrayEquals(Arrays.copyOf(inputArray, packLength), outputArray); } @Test public void testPackDoublesWithLength() { int inputLength = 100; int packLength = 76; double[] inputArray = rng.doubles().limit(inputLength).toArray(); byte[] bytes = ArrayPacking.pack(inputArray, packLength); double[] outputArray = ArrayPacking.unpackDoubles(bytes); assertEquals(packLength, outputArray.length); assertArrayEquals(Arrays.copyOf(inputArray, packLength), outputArray); assertDoesNotThrow(() -> pack(new double[0], 0)); assertThrows(IllegalArgumentException.class, () -> pack(new double[10], 11)); assertThrows(IllegalArgumentException.class, () -> pack(new double[10], -1)); } @Test public void testPackFloatsWithLength() { int inputLength = 100; int packLength = 76; float[] inputArray = new float[inputLength]; for (int i = 0; i < inputLength; i++) { inputArray[i] = rng.nextFloat(); } byte[] bytes = ArrayPacking.pack(inputArray, packLength); assertThrows(IllegalArgumentException.class, () -> pack(inputArray, inputLength + 10)); float[] outputArray = unpackFloats(bytes); assertEquals(packLength, outputArray.length); assertArrayEquals(Arrays.copyOf(inputArray, packLength), outputArray); assertDoesNotThrow(() -> pack(new float[0], 0)); assertThrows(IllegalArgumentException.class, () -> pack(new float[10], -1)); } @Test public void testUnpackDoublesWithLength() { int inputLength = 100; double[] inputArray = rng.doubles().limit(inputLength).toArray(); byte[] bytes = ArrayPacking.pack(inputArray); int unpackLength1 = 25; double[] outputArray1 = ArrayPacking.unpackDoubles(bytes, unpackLength1); assertEquals(unpackLength1, outputArray1.length); assertArrayEquals(Arrays.copyOf(inputArray, unpackLength1), outputArray1); int unpackLength2 = 123; assertThrows(IllegalArgumentException.class, () -> pack(inputArray, unpackLength2)); double[] outputArray2 = ArrayPacking.unpackDoubles(bytes, unpackLength2); assertEquals(unpackLength2, outputArray2.length); assertArrayEquals(inputArray, Arrays.copyOf(outputArray2, inputLength)); for (int i = inputLength; i < unpackLength2; i++) { assertEquals(0.0, outputArray2[i]); } } @Test public void testUnpackFloatWithLength() { int inputLength = 100; float[] inputArray = new float[inputLength]; for (int i = 0; i < inputLength; i++) { inputArray[i] = rng.nextFloat(); } byte[] bytes = ArrayPacking.pack(inputArray); int unpackLength1 = 25; float[] outputArray1 = unpackFloats(bytes, unpackLength1); assertEquals(unpackLength1, outputArray1.length); assertArrayEquals(Arrays.copyOf(inputArray, unpackLength1), outputArray1); int unpackLength2 = 123; float[] outputArray2 = unpackFloats(bytes, unpackLength2); assertEquals(unpackLength2, outputArray2.length); assertArrayEquals(inputArray, Arrays.copyOf(outputArray2, inputLength)); for (int i = inputLength; i < unpackLength2; i++) { assertEquals(0.0, outputArray2[i]); } } @Test public void testConfig() { byte[] array = new byte[1]; assertThrows(IllegalArgumentException.class, () -> unpackFloats(array, 1)); assertThrows(IllegalArgumentException.class, () -> unpackDoubles(array, 1)); byte[] newArray = new byte[Double.BYTES]; assertDoesNotThrow(() -> unpackDoubles(newArray, 1)); assertDoesNotThrow(() -> unpackFloats(newArray, 1)); assertThrows(IllegalArgumentException.class, () -> unpackFloats(newArray, -1)); assertThrows(IllegalArgumentException.class, () -> unpackDoubles(newArray, -1)); } }
409
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/util/ArrayUtilsTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertNotSame; import java.util.Arrays; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.CsvSource; public class ArrayUtilsTest { ArrayUtils utils = new ArrayUtils(); @ParameterizedTest @CsvSource({ "-0.0,0.0", "0.0,0.0", "-0.0:0.0:1.0,0.0:0.0:1.0" }) public void cleanCopy(String input, String expected) { double[] inputArray = array(input); double[] cleanCopy = ArrayUtils.cleanCopy(inputArray); assertNotSame(inputArray, cleanCopy); assertArrayEquals(array(expected), cleanCopy); } private double[] array(String arrayString) { return Arrays.stream(arrayString.split(":")).mapToDouble(Double::valueOf).toArray(); } }
410
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/util/WeightedTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import static com.amazon.randomcutforest.util.Weighted.createSample; import static com.amazon.randomcutforest.util.Weighted.prefixPick; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import java.util.ArrayList; import java.util.List; import java.util.Random; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class WeightedTest { private Random rng; int size = 10000; int heavyIndex; ArrayList<Weighted<Integer>> list = new ArrayList<>(); @BeforeEach public void setUp() { rng = new Random(); list = new ArrayList<>(); for (int i = 0; i < size; i++) { list.add(new Weighted<>(i, (float) (0.1 + rng.nextDouble()))); } heavyIndex = size + 7; list.add(new Weighted<>(heavyIndex, size)); } @Test public void testCreateSample() { // forcedSample 0 will return a null list assertTrue(createSample(list, 0, 10, 0, 1.0).size() == 0); // the following should add the last item first List<Weighted<Integer>> sampledList = createSample(list, 0, 10, 0.1, 1.0); assertTrue(sampledList.size() > 0); assertTrue(sampledList.get(0).index == heavyIndex); assertTrue(sampledList.get(0).weight == (float) size); } @Test public void testPrefixPick() { double total = list.stream().mapToDouble(e -> e.weight).sum(); assertTrue(total < 2 * size); Weighted<Integer> item = prefixPick(list, size / 3.0); assertTrue(item.index < size); assertTrue(item.weight <= 1.1); // should be the last element Weighted<Integer> heavyItem = prefixPick(list, 3.0 * size / 4); assertTrue(heavyItem.index == heavyIndex); assertTrue(heavyItem.weight == (float) size); // checking extreme weights heavyItem = prefixPick(list, 2 * size); assertTrue(heavyItem.index == heavyIndex); assertTrue(heavyItem.weight == (float) size); } @Test public void emptyList() { List<Weighted<Integer>> list = new ArrayList<>(); assertThrows(IllegalArgumentException.class, () -> prefixPick(list, 1.0f)); } }
411
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/util/ShingleBuilderTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertThrows; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class ShingleBuilderTest { private int dimensions; private int shingleSize; private ShingleBuilder builder; @BeforeEach public void setUp() { dimensions = 2; shingleSize = 3; builder = new ShingleBuilder(dimensions, shingleSize); } @Test public void testNew() { assertEquals(dimensions, builder.getInputPointSize()); assertEquals(dimensions * shingleSize, builder.getShingledPointSize()); assertFalse(builder.isCyclic()); } @Test public void testNewWithInvalidArguments() { assertThrows(IllegalArgumentException.class, () -> new ShingleBuilder(0, shingleSize)); assertThrows(IllegalArgumentException.class, () -> new ShingleBuilder(dimensions, 0)); } @Test public void testAddPoint() { double[] shingle = builder.getShingle(); assertArrayEquals(new double[] { 0, 0, 0, 0, 0, 0 }, shingle); builder.addPoint(new double[] { 9, 10 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 0, 0, 0, 0, 9, 10 }, shingle); builder.addPoint(new double[] { 7, 8 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 0, 0, 9, 10, 7, 8 }, shingle); builder.addPoint(new double[] { 5, 6 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 9, 10, 7, 8, 5, 6 }, shingle); builder.addPoint(new double[] { 3, 4 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 7, 8, 5, 6, 3, 4 }, shingle); } @Test public void testAddPointCyclic() { builder = new ShingleBuilder(dimensions, shingleSize, true); double[] shingle = builder.getShingle(); assertArrayEquals(new double[] { 0, 0, 0, 0, 0, 0 }, shingle); builder.addPoint(new double[] { 9, 10 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 9, 10, 0, 0, 0, 0 }, shingle); builder.addPoint(new double[] { 7, 8 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 9, 10, 7, 8, 0, 0 }, shingle); builder.addPoint(new double[] { 5, 6 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 9, 10, 7, 8, 5, 6 }, shingle); builder.addPoint(new double[] { 3, 4 }); shingle = builder.getShingle(); assertArrayEquals(new double[] { 3, 4, 7, 8, 5, 6 }, shingle); } @Test public void testAddPointWithInvalidArguments() { assertThrows(NullPointerException.class, () -> builder.addPoint(null)); double[] point = new double[9]; // wrong size of array assertThrows(IllegalArgumentException.class, () -> builder.addPoint(point)); } @Test public void testShingleCopy() { double[] buffer = new double[dimensions * shingleSize]; builder.addPoint(new double[] { 2, 1 }); builder.addPoint(new double[] { 4, 3 }); builder.addPoint(new double[] { 6, 5 }); double[] shingle = builder.getShingle(); assertArrayEquals(new double[] { 2, 1, 4, 3, 6, 5 }, shingle); assertArrayEquals(new double[] { 0, 0, 0, 0, 0, 0 }, buffer); builder.getShingle(buffer); assertArrayEquals(shingle, buffer); buffer[0] = 0; assertEquals(2, shingle[0]); } @Test public void testGetShingleWithInvalidArguments() { assertThrows(NullPointerException.class, () -> builder.getShingle(null)); double[] buffer = new double[2]; // wrong size of array assertThrows(IllegalArgumentException.class, () -> builder.getShingle(buffer)); } }
412
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/runner/AnomalyAttributionRunnerTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.io.BufferedReader; import java.io.IOException; import java.io.PrintWriter; import java.util.Arrays; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.returntypes.DiVector; public class AnomalyAttributionRunnerTest { private int numberOfTrees; private int sampleSize; private int shingleSize; private int windowSize; private String delimiter; private boolean headerRow; private AnomalyAttributionRunner runner; private BufferedReader in; private PrintWriter out; @BeforeEach public void setUp() { numberOfTrees = 50; sampleSize = 100; shingleSize = 1; windowSize = 10; delimiter = ","; headerRow = true; runner = new AnomalyAttributionRunner(); runner.parse("--number-of-trees", Integer.toString(numberOfTrees), "--sample-size", Integer.toString(sampleSize), "--shingle-size", Integer.toString(shingleSize), "--window-size", Integer.toString(windowSize), "--delimiter", delimiter, "--header-row", Boolean.toString(headerRow)); in = mock(BufferedReader.class); out = mock(PrintWriter.class); } @Test public void testRun() throws IOException { when(in.readLine()).thenReturn("a,b").thenReturn("1.0,2.0").thenReturn("4.0,5.0").thenReturn(null); runner.run(in, out); verify(out).println("a,b,anomaly_low_0,anomaly_high_0,anomaly_low_1,anomaly_high_1"); verify(out).println("1.0,2.0,0.0,0.0,0.0,0.0"); verify(out).println("4.0,5.0,0.0,0.0,0.0,0.0"); } @Test public void testWriteHeader() { String[] line = new String[] { "a", "b" }; runner.prepareAlgorithm(2); runner.writeHeader(line, out); verify(out).println("a,b,anomaly_low_0,anomaly_high_0,anomaly_low_1,anomaly_high_1"); } @Test public void testProcessLine() { String[] line = new String[] { "1.0", "2.0" }; runner.prepareAlgorithm(2); runner.processLine(line, out); verify(out).println("1.0,2.0,0.0,0.0,0.0,0.0"); } @Test public void testAnomalyAttributionTransformer() { RandomCutForest forest = mock(RandomCutForest.class); when(forest.getDimensions()).thenReturn(2); AnomalyAttributionRunner.AnomalyAttributionTransformer transformer = new AnomalyAttributionRunner.AnomalyAttributionTransformer( forest); DiVector vector = new DiVector(2); vector.low[0] = 1.1; vector.high[1] = 2.2; when(forest.getAnomalyAttribution(new double[] { 1.0, 2.0 })).thenReturn(vector); assertEquals(Arrays.asList("1.1", "0.0", "0.0", "2.2"), transformer.getResultValues(1.0, 2.0)); assertEquals(Arrays.asList("anomaly_low_0", "anomaly_high_0", "anomaly_low_1", "anomaly_high_1"), transformer.getResultColumnNames()); assertEquals(Arrays.asList("NA", "NA", "NA", "NA"), transformer.getEmptyResultValue()); } }
413
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/runner/SimpleDensityRunnerTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.io.BufferedReader; import java.io.IOException; import java.io.PrintWriter; import java.util.Arrays; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.returntypes.DensityOutput; public class SimpleDensityRunnerTest { private int numberOfTrees; private int sampleSize; private int shingleSize; private int windowSize; private String delimiter; private boolean headerRow; private SimpleDensityRunner runner; private BufferedReader in; private PrintWriter out; @BeforeEach public void setUp() { numberOfTrees = 50; sampleSize = 100; shingleSize = 1; windowSize = 10; delimiter = ","; headerRow = true; runner = new SimpleDensityRunner(); runner.parse("--number-of-trees", Integer.toString(numberOfTrees), "--sample-size", Integer.toString(sampleSize), "--shingle-size", Integer.toString(shingleSize), "--window-size", Integer.toString(windowSize), "--delimiter", delimiter, "--header-row", Boolean.toString(headerRow)); in = mock(BufferedReader.class); out = mock(PrintWriter.class); } @Test public void testRun() throws IOException { when(in.readLine()).thenReturn("a,b").thenReturn("1.0,2.0").thenReturn("4.0,5.0").thenReturn(null); runner.run(in, out); verify(out).println("a,b,prob_mass_0_up,prob_mass_0_down,prob_mass_1_up,prob_mass_1_down"); verify(out).println("1.0,2.0,0.000000,0.000000,0.000000,0.000000"); verify(out).println("4.0,5.0,0.000000,0.000000,0.000000,0.000000"); } @Test public void testWriteHeader() { String[] line = new String[] { "a", "b" }; runner.prepareAlgorithm(2); runner.writeHeader(line, out); verify(out).println("a,b,prob_mass_0_up,prob_mass_0_down,prob_mass_1_up,prob_mass_1_down"); } @Test public void testProcessLine() { String[] line = new String[] { "1.0", "2.0" }; runner.prepareAlgorithm(2); runner.processLine(line, out); verify(out).println("1.0,2.0,0.000000,0.000000,0.000000,0.000000"); } @Test public void testSimpleDensityTransformer() { RandomCutForest forest = mock(RandomCutForest.class); when(forest.getDimensions()).thenReturn(2); SimpleDensityRunner.SimpleDensityTransformer transformer = new SimpleDensityRunner.SimpleDensityTransformer( forest); DensityOutput expected = new DensityOutput(2, 1); expected.probMass.high[0] = 0.0; expected.probMass.low[0] = 0.5; expected.probMass.high[1] = 0.25; expected.probMass.low[1] = 0.25; expected.measure.high[0] = 0.0; expected.measure.low[0] = 8.0; expected.measure.high[1] = 8.0; expected.measure.low[1] = 4.0; when(forest.getSimpleDensity(new double[] { 1.0, 2.0 })).thenReturn(expected); assertEquals(Arrays.asList("0.000000", "400.000000", "400.000000", "200.000000"), transformer.getResultValues(1.0, 2.0)); assertEquals(Arrays.asList("prob_mass_0_up", "prob_mass_0_down", "prob_mass_1_up", "prob_mass_1_down"), transformer.getResultColumnNames()); assertEquals(Arrays.asList("NA", "NA", "NA", "NA"), transformer.getEmptyResultValue()); } }
414
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/runner/UpdateOnlyTransformerTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.verify; import java.util.List; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.RandomCutForest; public class UpdateOnlyTransformerTest { private RandomCutForest forest; private UpdateOnlyTransformer transformer; @BeforeEach public void setUp() { forest = mock(RandomCutForest.class); transformer = new UpdateOnlyTransformer(forest); } @Test public void testGetResultValues() { List<String> result = transformer.getResultValues(1.0, 2.0, 3.0); assertTrue(result.isEmpty()); verify(forest).update(new double[] { 1.0, 2.0, 3.0 }); } @Test public void testGetEmptyResultValue() { assertTrue(transformer.getEmptyResultValue().isEmpty()); } @Test public void testGetResultColumnNames() { assertTrue(transformer.getResultColumnNames().isEmpty()); } }
415
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/runner/AnomalyScoreRunnerTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.io.BufferedReader; import java.io.IOException; import java.io.PrintWriter; import java.util.Collections; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.RandomCutForest; public class AnomalyScoreRunnerTest { private int numberOfTrees; private int sampleSize; private int shingleSize; private int windowSize; private String delimiter; private boolean headerRow; private AnomalyScoreRunner runner; private BufferedReader in; private PrintWriter out; @BeforeEach public void setUp() { numberOfTrees = 50; sampleSize = 100; shingleSize = 1; windowSize = 10; delimiter = ","; headerRow = true; runner = new AnomalyScoreRunner(); runner.parse("--number-of-trees", Integer.toString(numberOfTrees), "--sample-size", Integer.toString(sampleSize), "--shingle-size", Integer.toString(shingleSize), "--window-size", Integer.toString(windowSize), "--delimiter", delimiter, "--header-row", Boolean.toString(headerRow)); in = mock(BufferedReader.class); out = mock(PrintWriter.class); } @Test public void testRun() throws IOException { when(in.readLine()).thenReturn("a,b,c").thenReturn("1.0,2.0,3.0").thenReturn("4.0,5.0,6.0").thenReturn(null); runner.run(in, out); verify(out).println("a,b,c,anomaly_score"); verify(out).println("1.0,2.0,3.0,0.0"); verify(out).println("4.0,5.0,6.0,0.0"); } @Test public void testWriteHeader() { String[] line = new String[] { "a", "b", "c" }; runner.prepareAlgorithm(3); runner.writeHeader(line, out); verify(out).println("a,b,c,anomaly_score"); } @Test public void testProcessLine() { String[] line = new String[] { "1.0", "2.0", "3.0" }; runner.prepareAlgorithm(3); runner.processLine(line, out); verify(out).println("1.0,2.0,3.0,0.0"); } @Test public void testAnomalyScoreTransformer() { RandomCutForest forest = mock(RandomCutForest.class); AnomalyScoreRunner.AnomalyScoreTransformer transformer = new AnomalyScoreRunner.AnomalyScoreTransformer(forest); when(forest.getAnomalyScore(new double[] { 1.0, 2.0, 3.0 })).thenReturn(11.0); assertEquals(Collections.singletonList("11.0"), transformer.getResultValues(1.0, 2.0, 3.0)); assertEquals(Collections.singletonList("anomaly_score"), transformer.getResultColumnNames()); assertEquals(Collections.singletonList("NA"), transformer.getEmptyResultValue()); } }
416
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/runner/ImputeRunnerTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.io.BufferedReader; import java.io.PrintWriter; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class ImputeRunnerTest { private int numberOfTrees; private int sampleSize; private int windowSize; private String delimiter; private boolean headerRow; private String missingValueMarker; private ImputeRunner runner; private BufferedReader in; private PrintWriter out; @BeforeEach public void setUp() { numberOfTrees = 50; sampleSize = 100; windowSize = 10; delimiter = ","; missingValueMarker = "X"; headerRow = true; runner = new ImputeRunner(); runner.parse("--number-of-trees", Integer.toString(numberOfTrees), "--sample-size", Integer.toString(sampleSize), "--window-size", Integer.toString(windowSize), "--delimiter", delimiter, "--missing-value-marker", missingValueMarker, "--header-row", Boolean.toString(headerRow)); in = mock(BufferedReader.class); out = mock(PrintWriter.class); } @Test public void testRun() throws Exception { when(in.readLine()).thenReturn("a,b").thenReturn("1.0,2.0").thenReturn("4.0,X").thenReturn(null); runner.run(in, out); verify(out).println("a,b"); verify(out).println("1.0,2.0"); verify(out).println("0.0,0.0"); } @Test public void testWriteHeader() { String[] line = new String[] { "a", "b" }; runner.prepareAlgorithm(2); runner.writeHeader(line, out); verify(out).println("a,b"); } @Test public void testProcessLine() { String[] line = new String[] { "1.0", "2.0" }; runner.prepareAlgorithm(2); runner.processLine(line, out); verify(out).println("1.0,2.0"); line = new String[] { missingValueMarker, "2.0" }; runner.processLine(line, out); verify(out).println("0.0,0.0"); } }
417
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/runner/ArgumentParserTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertTrue; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class ArgumentParserTest { private ArgumentParser parser; @BeforeEach public void setUp() { parser = new ArgumentParser("runner-class", "runner-description"); } @Test public void testNew() { assertEquals(100, parser.getNumberOfTrees()); assertEquals(256, parser.getSampleSize()); assertEquals(0, parser.getWindowSize()); assertEquals(0.0, parser.getTimeDecay()); assertEquals(1, parser.getShingleSize()); assertFalse(parser.getShingleCyclic()); assertEquals(",", parser.getDelimiter()); assertFalse(parser.getHeaderRow()); } @Test public void testParse() { parser.parse("--number-of-trees", "222", "--sample-size", "123", "--window-size", "50", "--shingle-size", "4", "--shingle-cyclic", "true", "--delimiter", "\t", "--header-row", "true"); assertEquals(222, parser.getNumberOfTrees()); assertEquals(123, parser.getSampleSize()); assertEquals(50, parser.getWindowSize()); assertEquals(0.02, parser.getTimeDecay()); assertEquals(4, parser.getShingleSize()); assertTrue(parser.getShingleCyclic()); assertEquals("\t", parser.getDelimiter()); assertTrue(parser.getHeaderRow()); } @Test public void testParseShortFlags() { parser.parse("-n", "222", "-s", "123", "-w", "50", "-g", "4", "-c", "true", "-d", "\t"); assertEquals(222, parser.getNumberOfTrees()); assertEquals(123, parser.getSampleSize()); assertEquals(50, parser.getWindowSize()); assertEquals(0.02, parser.getTimeDecay()); assertEquals(4, parser.getShingleSize()); assertEquals("\t", parser.getDelimiter()); } }
418
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/state/RandomCutForestMapperTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.state; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import java.util.Random; import java.util.stream.Stream; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.MethodSource; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.config.Precision; import com.amazon.randomcutforest.executor.PointStoreCoordinator; import com.amazon.randomcutforest.store.PointStore; import com.amazon.randomcutforest.testutils.NormalMixtureTestData; public class RandomCutForestMapperTest { private static int dimensions = 5; private static int sampleSize = 128; private static Stream<RandomCutForest> compactForestProvider() { RandomCutForest.Builder<?> builder = RandomCutForest.builder().compact(true).dimensions(dimensions) .sampleSize(sampleSize); RandomCutForest cachedDouble = builder.boundingBoxCacheFraction(new Random().nextDouble()) .precision(Precision.FLOAT_64).build(); RandomCutForest cachedFloat = builder.boundingBoxCacheFraction(new Random().nextDouble()) .precision(Precision.FLOAT_32).build(); RandomCutForest uncachedDouble = builder.boundingBoxCacheFraction(0.0).precision(Precision.FLOAT_64).build(); RandomCutForest uncachedFloat = builder.boundingBoxCacheFraction(0.0).precision(Precision.FLOAT_32).build(); return Stream.of(cachedDouble, cachedFloat, uncachedDouble, uncachedFloat); } private RandomCutForestMapper mapper; @BeforeEach public void setUp() { mapper = new RandomCutForestMapper(); mapper.setSaveExecutorContextEnabled(true); } public void assertCompactForestEquals(RandomCutForest forest, RandomCutForest forest2) { assertEquals(forest.getDimensions(), forest2.getDimensions()); assertEquals(forest.getSampleSize(), forest2.getSampleSize()); assertEquals(forest.getOutputAfter(), forest2.getOutputAfter()); assertEquals(forest.getNumberOfTrees(), forest2.getNumberOfTrees()); assertEquals(forest.getTimeDecay(), forest2.getTimeDecay()); assertEquals(forest.isStoreSequenceIndexesEnabled(), forest2.isStoreSequenceIndexesEnabled()); assertEquals(forest.isCompact(), forest2.isCompact()); assertEquals(forest.getPrecision(), forest2.getPrecision()); assertEquals(forest.getBoundingBoxCacheFraction(), forest2.getBoundingBoxCacheFraction()); assertEquals(forest.isCenterOfMassEnabled(), forest2.isCenterOfMassEnabled()); assertEquals(forest.isParallelExecutionEnabled(), forest2.isParallelExecutionEnabled()); assertEquals(forest.getThreadPoolSize(), forest2.getThreadPoolSize()); PointStoreCoordinator coordinator = (PointStoreCoordinator) forest.getUpdateCoordinator(); PointStoreCoordinator coordinator2 = (PointStoreCoordinator) forest2.getUpdateCoordinator(); PointStore store = (PointStore) coordinator.getStore(); PointStore store2 = (PointStore) coordinator2.getStore(); assertArrayEquals(store.getRefCount(), store2.getRefCount()); assertArrayEquals(store.getStore(), store2.getStore()); assertEquals(store.getCapacity(), store2.getCapacity()); assertEquals(store.size(), store2.size()); } @ParameterizedTest @MethodSource("compactForestProvider") public void testRoundTripForCompactForest(RandomCutForest forest) { NormalMixtureTestData testData = new NormalMixtureTestData(); for (double[] point : testData.generateTestData(sampleSize, dimensions)) { forest.update(point); } RandomCutForest forest2 = mapper.toModel(mapper.toState(forest)); assertCompactForestEquals(forest, forest2); } @ParameterizedTest @MethodSource("compactForestProvider") public void testRoundTripForCompactForestSaveTreeState(RandomCutForest forest) { mapper.setSaveTreeStateEnabled(true); testRoundTripForCompactForest(forest); } @ParameterizedTest @MethodSource("compactForestProvider") public void testRoundTripForCompactForestSaveTreeStatePartial(RandomCutForest forest) { mapper.setSaveTreeStateEnabled(true); mapper.setPartialTreeStateEnabled(true); testRoundTripForCompactForest(forest); } @Test public void testRoundTripForEmptyForest() { Precision precision = Precision.FLOAT_64; RandomCutForest forest = RandomCutForest.builder().compact(true).dimensions(dimensions).sampleSize(sampleSize) .precision(precision).numberOfTrees(1).build(); mapper.setSaveTreeStateEnabled(true); RandomCutForest forest2 = mapper.toModel(mapper.toState(forest)); assertCompactForestEquals(forest, forest2); } @Test public void testRoundTripForSingleNodeForest() { int dimensions = 10; long seed = new Random().nextLong(); System.out.println(" Seed " + seed); RandomCutForest forest = RandomCutForest.builder().compact(true).dimensions(dimensions).numberOfTrees(1) .precision(Precision.FLOAT_32).internalShinglingEnabled(false).randomSeed(seed).build(); Random r = new Random(seed + 1); double[] point = r.ints(dimensions, 0, 50).asDoubleStream().toArray(); for (int i = 0; i < new Random().nextInt(1000); i++) { forest.update(point); } RandomCutForestMapper mapper = new RandomCutForestMapper(); mapper.setSaveExecutorContextEnabled(true); mapper.setSaveTreeStateEnabled(true); mapper.setPartialTreeStateEnabled(true); RandomCutForest copyForest = mapper.toModel(mapper.toState(forest)); for (int i = 0; i < new Random(seed + 2).nextInt(1000); i++) { double[] anotherPoint = r.ints(dimensions, 0, 50).asDoubleStream().toArray(); assertEquals(forest.getAnomalyScore(anotherPoint), copyForest.getAnomalyScore(anotherPoint), 1e-10); forest.update(anotherPoint); copyForest.update(anotherPoint); } } private static float[] generate(int input) { return new float[] { (float) (20 * Math.sin(input / 10.0)), (float) (20 * Math.cos(input / 10.0)) }; } @Test void benchmarkMappers() { long seed = new Random().nextLong(); System.out.println(" Seed " + seed); Random random = new Random(seed); RandomCutForest rcf = RandomCutForest.builder().dimensions(2 * 10).shingleSize(10).sampleSize(628) .internalShinglingEnabled(true).randomSeed(random.nextLong()).build(); for (int i = 0; i < 10000; i++) { rcf.update(generate(i)); } RandomCutForestMapper mapper = new RandomCutForestMapper(); mapper.setSaveExecutorContextEnabled(true); mapper.setSaveTreeStateEnabled(true); for (int j = 0; j < 1000; j++) { RandomCutForest newRCF = mapper.toModel(mapper.toState(rcf)); float[] test = generate(10000 + j); assertEquals(newRCF.getAnomalyScore(test), rcf.getAnomalyScore(test), 1e-6); rcf.update(test); } } }
419
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/state
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/state/sampler/CompactSamplerMapperTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.state.sampler; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertNull; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import java.util.Random; import java.util.stream.Stream; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.Arguments; import org.junit.jupiter.params.provider.MethodSource; import com.amazon.randomcutforest.sampler.CompactSampler; public class CompactSamplerMapperTest { private static int sampleSize = 20; private static double lambda = 0.01; private static long seed = 4444; public static Stream<Arguments> nonemptySamplerProvider() { CompactSampler fullSampler1 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda).randomSeed(seed) .storeSequenceIndexesEnabled(false).build(); CompactSampler fullSampler2 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda).randomSeed(seed) .storeSequenceIndexesEnabled(true).build(); Random random = new Random(); long baseIndex = 10_000; for (int i = 0; i < 100; i++) { int pointReference = random.nextInt(); fullSampler1.update(pointReference, baseIndex + i); fullSampler2.update(pointReference, baseIndex + i); } CompactSampler partiallyFullSampler1 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda) .randomSeed(seed).storeSequenceIndexesEnabled(false).build(); CompactSampler partiallyFullSampler2 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda) .randomSeed(seed).storeSequenceIndexesEnabled(true).build(); for (int i = 0; i < sampleSize / 2; i++) { int pointReference = random.nextInt(); partiallyFullSampler1.update(pointReference, baseIndex + i); partiallyFullSampler2.update(pointReference, baseIndex + i); } return Stream.of(Arguments.of("full sampler without sequence indexes", fullSampler1), Arguments.of("full sampler with sequence indexes", fullSampler2), Arguments.of("partially full sampler without sequence indexes", partiallyFullSampler1), Arguments.of("partially full sampler with sequence indexes", partiallyFullSampler2)); } public static Stream<Arguments> samplerProvider() { CompactSampler emptySampler1 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda).randomSeed(seed) .storeSequenceIndexesEnabled(false).build(); CompactSampler emptySampler2 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda).randomSeed(seed) .storeSequenceIndexesEnabled(true).build(); return Stream.concat(nonemptySamplerProvider(), Stream.of(Arguments.of("empty sampler without sequence indexes", emptySampler1), Arguments.of("empty sampler with sequence indexes", emptySampler2))); } private CompactSamplerMapper mapper; @BeforeEach public void setUp() { mapper = new CompactSamplerMapper(); mapper.setValidateHeapEnabled(false); } private void assertValidMapping(CompactSampler original, CompactSampler mapped) { assertArrayEquals(original.getWeightArray(), mapped.getWeightArray(), "different weight arrays"); assertArrayEquals(original.getPointIndexArray(), mapped.getPointIndexArray(), "different point index arrays"); assertEquals(original.getCapacity(), mapped.getCapacity()); assertEquals(original.size(), mapped.size()); assertEquals(original.getTimeDecay(), mapped.getTimeDecay()); assertFalse(mapped.getEvictedPoint().isPresent()); if (original.isStoreSequenceIndexesEnabled()) { assertTrue(mapped.isStoreSequenceIndexesEnabled()); assertArrayEquals(original.getSequenceIndexArray(), mapped.getSequenceIndexArray(), "different sequence index arrays"); } else { assertFalse(mapped.isStoreSequenceIndexesEnabled()); assertNull(mapped.getSequenceIndexArray()); } } @ParameterizedTest @MethodSource("nonemptySamplerProvider") public void testRoundTripInvalidHeap(String description, CompactSampler sampler) { mapper.setValidateHeapEnabled(true); CompactSamplerState state = mapper.toState(sampler); // swap to weights in the weight array in order to violate the heap property float[] weights = state.getWeight(); int index = state.getSize() / 4; float temp = weights[index]; weights[index] = weights[2 * index + 1]; weights[2 * index + 1] = temp; assertThrows(IllegalStateException.class, () -> mapper.toModel(state)); mapper.setValidateHeapEnabled(false); CompactSampler sampler2 = mapper.toModel(state); assertArrayEquals(sampler.getWeightArray(), sampler2.getWeightArray()); } }
420
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/state
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/state/store/PointStoreMapperTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.state.store; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.store.PointStore; import com.amazon.randomcutforest.store.PointStoreSmall; public class PointStoreMapperTest { private PointStoreMapper mapper; @BeforeEach public void setUp() { mapper = new PointStoreMapper(); } @Test public void testRoundTrip() { int dimensions = 2; int capacity = 4; PointStore store = new PointStoreSmall(dimensions, capacity); float[] point1 = { 1.1f, -22.2f }; int index1 = store.add(point1, 1); float[] point2 = { 3.3f, -4.4f }; int index2 = store.add(point2, 2); float[] point3 = { 10.1f, 100.1f }; int index3 = store.add(point3, 3); PointStore store2 = mapper.toModel(mapper.toState(store)); assertEquals(capacity, store2.getCapacity()); assertEquals(3, store2.size()); assertEquals(dimensions, store2.getDimensions()); assertArrayEquals(store.getStore(), store2.getStore()); } }
421
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/sampler/CompactSamplerTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.sampler; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertNotEquals; import static org.junit.jupiter.api.Assertions.assertNotNull; import static org.junit.jupiter.api.Assertions.assertNull; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.ArgumentMatchers.any; import static org.mockito.Mockito.reset; import static org.mockito.Mockito.spy; import static org.mockito.Mockito.times; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.util.Arrays; import java.util.Comparator; import java.util.List; import java.util.Random; import java.util.stream.Stream; import org.junit.jupiter.api.Test; import org.junit.jupiter.api.extension.ExtensionContext; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.Arguments; import org.junit.jupiter.params.provider.ArgumentsProvider; import org.junit.jupiter.params.provider.ArgumentsSource; import com.amazon.randomcutforest.config.Config; public class CompactSamplerTest { private static int sampleSize = 256; private static double lambda = 0.01; private static long seed = 42L; private static class SamplerProvider implements ArgumentsProvider { @Override public Stream<? extends Arguments> provideArguments(ExtensionContext context) throws Exception { Random random1 = spy(new Random(seed)); CompactSampler sampler1 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda).random(random1) .initialAcceptFraction(0.1).storeSequenceIndexesEnabled(false).build(); Random random2 = spy(new Random(seed)); CompactSampler sampler2 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda).random(random2) .initialAcceptFraction(0.1).storeSequenceIndexesEnabled(true).build(); CompactSampler sampler3 = CompactSampler.builder().capacity(sampleSize).timeDecay(lambda).random(random1) .initialAcceptFraction(1.0).storeSequenceIndexesEnabled(false).build(); return Stream.of(Arguments.of(random1, sampler1), Arguments.of(random2, sampler2), Arguments.of(random1, sampler3)); } } @ParameterizedTest @ArgumentsSource(SamplerProvider.class) public void testNew(Random random, CompactSampler sampler) { // test CompactSampler fields not defined in the IStreamSampler interface assertEquals(lambda, sampler.getTimeDecay()); assertNotNull(sampler.getWeightArray()); assertNotNull(sampler.getPointIndexArray()); long seq = new Random().nextLong(); sampler.setMaxSequenceIndex(seq); assertEquals(sampler.getMaxSequenceIndex(), seq); assertFalse(sampler.isFull()); assertFalse(sampler.isReady()); double newLambda = new Random().nextDouble(); sampler.setTimeDecay(newLambda); assertEquals(sampler.getConfig(Config.TIME_DECAY), newLambda); sampler.setConfig(Config.TIME_DECAY, lambda + newLambda); assertEquals(sampler.getTimeDecay(), lambda + newLambda, 1e-10); assertEquals(sampler.getMostRecentTimeDecayUpdate(), seq); sampler.setMostRecentTimeDecayUpdate(0L); assertEquals(sampler.getMostRecentTimeDecayUpdate(), 0L); assertThrows(IllegalArgumentException.class, () -> sampler.getConfig("foo")); assertThrows(IllegalArgumentException.class, () -> sampler.setConfig("bar", 0L)); if (sampler.isStoreSequenceIndexesEnabled()) { assertNotNull(sampler.getSequenceIndexArray()); } else { assertNull(sampler.getSequenceIndexArray()); } assertThrows(IllegalStateException.class, () -> sampler.addPoint(1)); assertDoesNotThrow(() -> sampler.addPoint(null)); } @Test public void testNewFromExistingWeightsParameters() { int sampleSize = 3; double lambda = 0.1; // weight array is valid heap float[] weight = { 0.4f, 0.3f, 0.2f }; int[] pointIndex = { 1, 2, 3 }; assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(sampleSize).size(weight.length).timeDecay(lambda) .random(new Random()).weight(weight).pointIndex(pointIndex).sequenceIndex(null) .storeSequenceIndexesEnabled(true).validateHeap(true).build()); } @Test public void testNewFromExistingWeights() { int sampleSize = 3; double lambda = 0.1; // weight array is valid heap float[] weight = { 0.4f, 0.3f, 0.2f }; int[] pointIndex = { 1, 2, 3 }; CompactSampler sampler = new CompactSampler.Builder<>().capacity(sampleSize).size(weight.length) .timeDecay(lambda).random(new Random()).weight(weight).pointIndex(pointIndex).sequenceIndex(null) .validateHeap(true).build(); assertFalse(sampler.getEvictedPoint().isPresent()); assertFalse(sampler.isStoreSequenceIndexesEnabled()); assertEquals(3, sampler.size()); assertNull(sampler.getSequenceIndexArray()); for (int i = 0; i < 3; i++) { assertEquals(weight[i], sampler.weight[i]); assertEquals(pointIndex[i], sampler.pointIndex[i]); } sampler.setMaxSequenceIndex(10L); sampler.setTimeDecay(lambda * 2); assertNotEquals(sampler.accumuluatedTimeDecay, 0); sampler.getWeightedSample(); assertEquals(sampler.accumuluatedTimeDecay, 0); } @Test public void testUniformSampler() { CompactSampler uniformSampler = CompactSampler.uniformSampler(sampleSize, seed, false); assertFalse(uniformSampler.getEvictedPoint().isPresent()); assertFalse(uniformSampler.isReady()); assertFalse(uniformSampler.isFull()); assertEquals(sampleSize, uniformSampler.getCapacity()); assertEquals(0, uniformSampler.size()); assertEquals(0.0, uniformSampler.getTimeDecay()); } @Test public void testBuilderClass() { assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(0).initialAcceptFraction(0.5).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1).initialAcceptFraction(0).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1).size(1).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1).validateHeap(true).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1).weight(new float[] { 0 }).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1) .sequenceIndex(new long[] { 0 }).storeSequenceIndexesEnabled(true).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1).pointIndex(new int[] { 0 }).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1) .weight(new float[0]).pointIndex(new int[] { 0 }).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1) .weight(new float[] { 0 }).pointIndex(new int[0]).build()); assertThrows(IllegalArgumentException.class, () -> new CompactSampler.Builder<>().capacity(1).weight(new float[] { 0 }).pointIndex(new int[] { 0 }) .sequenceIndex(new long[0]).storeSequenceIndexesEnabled(true).build()); assertDoesNotThrow(() -> new CompactSampler.Builder<>().capacity(1).weight(new float[] { 0 }) .pointIndex(new int[] { 0 }).sequenceIndex(new long[] { 0 }).storeSequenceIndexesEnabled(true).build()); assertDoesNotThrow(() -> new CompactSampler.Builder<>().capacity(1).weight(new float[] { 0 }) .pointIndex(new int[] { 0 }).build()); } @ParameterizedTest @ArgumentsSource(SamplerProvider.class) public void testAddPoint(Random random, CompactSampler sampler) { when(random.nextDouble()).thenReturn(0.0).thenReturn(0.5).thenReturn(0.0).thenReturn(0.01).thenReturn(0.0) .thenReturn(0.99); sampler.acceptPoint(10L); double weight1 = sampler.acceptPointState.getWeight(); sampler.addPoint(1); sampler.acceptPoint(11L); double weight2 = sampler.acceptPointState.getWeight(); // acceptstate is non-null assertThrows(IllegalArgumentException.class, () -> sampler.addPoint(12, 2.0f, 0L)); sampler.addPoint(12); assertThrows(IllegalArgumentException.class, () -> sampler.acceptPoint(12L, -1f)); sampler.acceptPoint(12L, 0f); assertNull(sampler.acceptPointState); sampler.acceptPoint(12L); double weight3 = sampler.acceptPointState.getWeight(); sampler.addPoint(123); assertEquals(3, sampler.size()); assertEquals(sampleSize, sampler.getCapacity()); List<Weighted<Integer>> samples = sampler.getWeightedSample(); samples.sort(Comparator.comparing(Weighted<Integer>::getWeight)); assertEquals(3, samples.size()); assertEquals(123, samples.get(0).getValue()); assertEquals(weight3, samples.get(0).getWeight()); assertEquals(1, samples.get(1).getValue()); assertEquals(weight1, samples.get(1).getWeight()); assertEquals(12, samples.get(2).getValue()); assertEquals(weight2, samples.get(2).getWeight()); } @ParameterizedTest @ArgumentsSource(SamplerProvider.class) public void testAcceptPoint(Random random, CompactSampler sampler) { assertThrows(IllegalArgumentException.class, () -> sampler.addPoint(null, 1, 0L)); assertThrows(IllegalArgumentException.class, () -> sampler.addPoint(null, -1, 0L)); assertDoesNotThrow(() -> sampler.addPoint(0, 0f, 0L)); assertEquals(sampler.size, 1); // The sampler should accept all samples until initial fraction for (int i = 0; i < sampleSize * sampler.initialAcceptFraction; i++) { assertTrue(sampler.acceptPoint(i)); assertNotNull(sampler.acceptPointState); sampler.addPoint(i); } assertTrue(sampler.initialAcceptProbability(sampler.size) < 1.0); for (int i = 0; i < sampleSize * 10; i++) { if (sampler.acceptPoint(i)) { sampler.addPoint(i); } } assertTrue(sampler.isFull()); assertTrue(sampler.isReady()); assertThrows(IllegalStateException.class, () -> sampler.addPoint(sampleSize)); assertThrows(IllegalArgumentException.class, () -> sampler.addPoint(sampleSize, 1.0f, 0L)); sampler.setTimeDecay(0); // we should only accept sequences of value samplesize - 1 or higher assertThrows(IllegalStateException.class, () -> sampler.acceptPoint(sampleSize - 2)); // In subsequent calls to sample, either the result is empty or else // the new weight is smaller than the evicted weight int numAccepted = 0; for (int i = 10 * sampleSize; i < 12 * sampleSize; i++) { if (sampler.acceptPoint(i)) { numAccepted++; assertTrue(sampler.getEvictedPoint().isPresent()); assertNotNull(sampler.acceptPointState); Weighted<Integer> evictedPoint = (Weighted<Integer>) sampler.getEvictedPoint().get(); assertTrue(sampler.acceptPointState.getWeight() < evictedPoint.getWeight()); sampler.addPoint(i); } } assertTrue(numAccepted > 0, "the sampler did not accept any points"); } @ParameterizedTest @ArgumentsSource(SamplerProvider.class) public void testUpdate(Random random, CompactSampler compactSampler) { CompactSampler sampler = spy(compactSampler); for (int i = 0; i < sampleSize * sampler.initialAcceptFraction; i++) { assertTrue(sampler.update(i, i)); } int num = (int) Math.ceil(sampleSize * sampler.initialAcceptFraction); // all points should be added to the sampler until the sampler is full assertEquals(num, sampler.size()); verify(sampler, times(num)).addPoint(any()); reset(sampler); int numSampled = 0; for (int i = num; i < 2 * sampleSize; i++) { if (sampler.update(i, i)) { numSampled++; } } assertTrue(numSampled > 0, "no new values were sampled"); assertTrue(sampler.initialAcceptFraction > 0.5 || numSampled < 2 * sampleSize - num, "all values were sampled"); verify(sampler, times(numSampled)).addPoint(any()); } @ParameterizedTest @ArgumentsSource(SamplerProvider.class) public void testGetScore(Random random, CompactSampler sampler) { when(random.nextDouble()).thenReturn(0.0).thenReturn(0.25).thenReturn(0.0).thenReturn(0.75).thenReturn(0.0) .thenReturn(0.50).thenReturn(0.5).thenReturn(0.1).thenReturn(1.3); sampler.update(1, 101); sampler.update(2, 102); sampler.update(3, 103); double[] expectedScores = new double[3]; expectedScores[0] = -lambda * 101L + Math.log(-Math.log(0.25)); expectedScores[1] = -lambda * 102L + Math.log(-Math.log(0.75)); expectedScores[2] = -lambda * 103L + Math.log(-Math.log(0.50)); Arrays.sort(expectedScores); assertFalse(sampler.acceptPoint(104)); List<Weighted<Integer>> samples = sampler.getWeightedSample(); samples.sort(Comparator.comparing(Weighted<Integer>::getWeight)); for (int i = 0; i < 3; i++) { assertEquals(expectedScores[i], samples.get(i).getWeight(), EPSILON); } } @ParameterizedTest @ArgumentsSource(SamplerProvider.class) public void testValidateHeap(Random random, CompactSampler sampler) { // populate the heap for (int i = 0; i < 2 * sampleSize; i++) { sampler.update(i, i); } float[] weightArray = sampler.getWeightArray(); // swapping a weight value with one of its children will break the heap property int i = sampleSize / 4; float f = weightArray[i]; weightArray[i] = weightArray[2 * i + 1]; weightArray[2 * i + 1] = f; assertThrows(IllegalStateException.class, () -> new CompactSampler.Builder<>().capacity(sampleSize).size(sampleSize).timeDecay(lambda) .random(random).weight(weightArray).pointIndex(sampler.getPointIndexArray()) .sequenceIndex(sampler.getSequenceIndexArray()).validateHeap(true).build()); } }
422
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/imputation/ConditionalSampleSummarizerTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.imputation; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertNotNull; import static org.junit.jupiter.api.Assertions.assertNull; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import java.util.ArrayList; import java.util.Collections; import java.util.Random; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.returntypes.ConditionalTreeSample; import com.amazon.randomcutforest.returntypes.SampleSummary; import com.amazon.randomcutforest.summarization.Summarizer; public class ConditionalSampleSummarizerTest { private float[] queryPoint; private int[] missingIndexes; ConditionalSampleSummarizer summarizer; @BeforeEach public void setUp() { queryPoint = new float[] { 50, 70, 90, 100 }; missingIndexes = new int[] { 2, 3 }; summarizer = new ConditionalSampleSummarizer(missingIndexes, queryPoint, 0.2, true); } @Test public void testSummarize() { assertThrows(IllegalArgumentException.class, () -> summarizer.summarize(Collections.emptyList())); Random random = new Random(42); ArrayList<ConditionalTreeSample> list = new ArrayList<>(); for (int i = 0; i < 999; i++) { float[] point = new float[] { 50, 70, 90, 100 + 2 * random.nextFloat() }; list.add(new ConditionalTreeSample(i, null, Summarizer.L1distance(point, queryPoint), point)); } list.add(new ConditionalTreeSample(999, null, 100, new float[] { 50, 70, 90, 200 })); SampleSummary summary = summarizer.summarize(list, false); assertNull(summary.summaryPoints); SampleSummary summaryTwo = summarizer.summarize(list, true); assertNotNull(summaryTwo.summaryPoints); for (float[] element : summaryTwo.summaryPoints) { assertEquals(element.length, 4); assertEquals(element[0], 50); assertEquals(element[1], 70); assertEquals(element[2], 90); assertTrue(100 < element[3] && element[3] < 102); } assertEquals(2, summaryTwo.mean.length); assertEquals(0, summaryTwo.deviation[0]); } @Test public void testZero() { ArrayList<ConditionalTreeSample> list = new ArrayList<>(); for (int i = 0; i < 1000; i++) { list.add(new ConditionalTreeSample(i, null, 0, queryPoint)); } assert (summarizer.summarize(list, true).summaryPoints.length == 1); } }
423
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/imputation/ImputeVisitorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.imputation; import static com.amazon.randomcutforest.CommonUtils.defaultScoreSeenFunction; import static com.amazon.randomcutforest.CommonUtils.defaultScoreUnseenFunction; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertNotEquals; import static org.junit.jupiter.api.Assertions.assertNotSame; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.ArgumentMatchers.any; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.when; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.CommonUtils; import com.amazon.randomcutforest.tree.BoundingBox; import com.amazon.randomcutforest.tree.IBoundingBoxView; import com.amazon.randomcutforest.tree.INodeView; import com.amazon.randomcutforest.tree.NodeView; public class ImputeVisitorTest { private float[] queryPoint; private int numberOfMissingValues; private int[] missingIndexes; private ImputeVisitor visitor; private ImputeVisitor anotherVisitor; @BeforeEach public void setUp() { // create a point where the 2nd value is missing // The second value of queryPoint and the 2nd and 3rd values of missingIndexes // should be ignored in all tests queryPoint = new float[] { -1.0f, 1000.0f, 3.0f }; numberOfMissingValues = 1; missingIndexes = new int[] { 1, 99, -888 }; visitor = new ImputeVisitor(queryPoint, numberOfMissingValues, missingIndexes); anotherVisitor = new ImputeVisitor(queryPoint, queryPoint, null, null, 0.8, 42); } @Test public void testNew() { assertArrayEquals(queryPoint, visitor.getResult().leafPoint); assertNotSame(queryPoint, visitor.getResult()); assertEquals(ImputeVisitor.DEFAULT_INIT_VALUE, visitor.getAnomalyRank()); assertEquals(ImputeVisitor.DEFAULT_INIT_VALUE, visitor.adjustedRank()); assertEquals(ImputeVisitor.DEFAULT_INIT_VALUE, anotherVisitor.getAnomalyRank()); assertNotEquals(ImputeVisitor.DEFAULT_INIT_VALUE, anotherVisitor.adjustedRank()); assertEquals(visitor.getDistance(), Double.MAX_VALUE); assertFalse(visitor.isConverged()); assertThrows(IllegalArgumentException.class, () -> new ImputeVisitor(queryPoint, queryPoint, null, null, -1.0, 42)); assertThrows(IllegalArgumentException.class, () -> new ImputeVisitor(queryPoint, queryPoint, null, null, 2.0, 42)); assertThrows(IllegalArgumentException.class, () -> new ImputeVisitor(queryPoint, queryPoint, null, new int[] { -1 }, 1.0, 42)); assertThrows(IllegalArgumentException.class, () -> new ImputeVisitor(queryPoint, queryPoint, null, new int[] { 4 }, 1.0, 42)); } @Test public void testCopyConstructor() { ImputeVisitor copy = new ImputeVisitor(visitor); assertArrayEquals(queryPoint, copy.getResult().leafPoint); assertNotSame(copy.getResult(), visitor.getResult()); assertEquals(ImputeVisitor.DEFAULT_INIT_VALUE, visitor.getAnomalyRank()); } @Test public void testAcceptLeafEquals() { float[] point = { queryPoint[0], 2.0f, queryPoint[2] }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getLiftedLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafDepth = 100; int leafMass = 10; when(leafNode.getMass()).thenReturn(leafMass); visitor.acceptLeaf(leafNode, leafDepth); anotherVisitor.acceptLeaf(leafNode, leafDepth); float[] expected = new float[] { -1.0f, 2.0f, 3.0f }; assertArrayEquals(expected, visitor.getResult().leafPoint); assertEquals(visitor.getDistance(), 0, 1e-6); assertEquals(defaultScoreSeenFunction(leafDepth, leafMass), visitor.getAnomalyRank()); } @Test public void testAcceptLeafEqualsZeroDepth() { float[] point = { queryPoint[0], 2.0f, queryPoint[2] }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getLiftedLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafDepth = 0; int leafMass = 10; when(leafNode.getMass()).thenReturn(leafMass); visitor.acceptLeaf(leafNode, leafDepth); float[] expected = new float[] { -1.0f, 2.0f, 3.0f }; assertArrayEquals(expected, visitor.getResult().leafPoint); assertEquals(0.0, visitor.getAnomalyRank()); } @Test public void testAcceptLeafNotEquals() { float[] point = { queryPoint[0], 2.0f, -111.11f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getLiftedLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafDepth = 100; int leafMass = 10; when(leafNode.getMass()).thenReturn(leafMass); visitor.acceptLeaf(leafNode, leafDepth); float[] expected = new float[] { -1.0f, 2.0f, 3.0f }; assertArrayEquals(expected, visitor.getResult().leafPoint); assertEquals(defaultScoreUnseenFunction(leafDepth, leafMass), visitor.getAnomalyRank()); } @Test public void testAccept() { float[] point = { queryPoint[0], 2.0f, -111.11f }; INodeView node = mock(NodeView.class); when(node.getLeafPoint()).thenReturn(point); when(node.getLiftedLeafPoint()).thenReturn(point); when(node.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int depth = 100; int leafMass = 10; when(node.getMass()).thenReturn(leafMass); visitor.acceptLeaf(node, depth); float[] expected = new float[] { -1.0f, 2.0f, 3.0f }; assertArrayEquals(expected, visitor.getResult().leafPoint); assertEquals(defaultScoreUnseenFunction(depth, leafMass), visitor.getAnomalyRank()); depth--; IBoundingBoxView boundingBox = node.getBoundingBox().getMergedBox(new float[] { 99.0f, 4.0f, -19.0f }); when(node.getBoundingBox()).thenReturn(boundingBox); when(node.probailityOfSeparation(any())) .thenReturn(CommonUtils.getProbabilityOfSeparation(boundingBox, expected)); when(node.getMass()).thenReturn(leafMass + 2); double oldRank = visitor.getAnomalyRank(); visitor.accept(node, depth); assertArrayEquals(expected, visitor.getResult().leafPoint); double p = CommonUtils.getProbabilityOfSeparation(boundingBox, expected); double expectedRank = p * defaultScoreUnseenFunction(depth, node.getMass()) + (1 - p) * oldRank; assertEquals(expectedRank, visitor.getAnomalyRank(), EPSILON); } @Test public void testNewCopy() { ImputeVisitor copy = (ImputeVisitor) visitor.newCopy(); assertArrayEquals(queryPoint, copy.getResult().leafPoint); assertNotSame(copy.getResult(), visitor.getResult()); assertEquals(ImputeVisitor.DEFAULT_INIT_VALUE, visitor.getAnomalyRank()); } @Test public void testMerge() { float[] otherPoint = new float[] { 99, 100, 101 }; ImputeVisitor other = new ImputeVisitor(otherPoint, 0, new int[0]); // set other.rank to a small value NodeView node = mock(NodeView.class); when(node.getLeafPoint()).thenReturn(new float[] { 0, 0, 0 }); when(node.getLiftedLeafPoint()).thenReturn(new float[] { 0, 0, 0 }); when(node.getBoundingBox()).thenReturn(new BoundingBox(new float[] { 0, 0, 0 })); other.acceptLeaf(node, 99); assertTrue(other.getAnomalyRank() < visitor.getAnomalyRank()); other.combine(visitor); assertArrayEquals(otherPoint, other.getResult().leafPoint); visitor.combine(other); assertArrayEquals(otherPoint, visitor.getResult().leafPoint); } }
424
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/executor/UpdateResultTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.executor; import static org.junit.jupiter.api.Assertions.assertFalse; import org.junit.jupiter.api.Test; public class UpdateResultTest { @Test public void testNoop() { UpdateResult<Integer> result = UpdateResult.noop(); assertFalse(result.getAddedPoint().isPresent()); assertFalse(result.getDeletedPoint().isPresent()); assertFalse(result.isStateChange()); } }
425
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/executor/SamplerPlusTreeTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.executor; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.ArgumentMatchers.any; import static org.mockito.ArgumentMatchers.anyLong; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.never; import static org.mockito.Mockito.times; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.util.Optional; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import org.junit.jupiter.api.extension.ExtendWith; import org.mockito.Mock; import org.mockito.junit.jupiter.MockitoExtension; import com.amazon.randomcutforest.sampler.ISampled; import com.amazon.randomcutforest.sampler.IStreamSampler; import com.amazon.randomcutforest.tree.ITree; @ExtendWith(MockitoExtension.class) public class SamplerPlusTreeTest { @Mock private ITree<Integer, double[]> tree; @Mock private IStreamSampler<Integer> sampler; private SamplerPlusTree<Integer, double[]> samplerPlusTree; @BeforeEach public void setUp() { samplerPlusTree = new SamplerPlusTree<>(sampler, tree); } @Test public void testUpdateAddPoint() { int pointReference = 2; long sequenceIndex = 100L; int existingPointReference = 222; when(sampler.acceptPoint(sequenceIndex)).thenReturn(true); when(sampler.getEvictedPoint()).thenReturn(Optional.empty()); when(tree.addPoint(pointReference, sequenceIndex)).thenReturn(existingPointReference); UpdateResult<Integer> result = samplerPlusTree.update(pointReference, sequenceIndex); assertTrue(result.getAddedPoint().isPresent()); assertEquals(existingPointReference, result.getAddedPoint().get()); assertFalse(result.getDeletedPoint().isPresent()); verify(tree, never()).deletePoint(any(), anyLong()); verify(sampler, times(1)).addPoint(existingPointReference); } @Test public void testUpdateAddAndDeletePoint() { int pointReference = 2; long sequenceIndex = 100L; int existingPointReference = 222; int evictedPoint = 333; long evictedSequenceIndex = 50L; ISampled<Integer> evictedPointSampled = mock(ISampled.class); when(evictedPointSampled.getValue()).thenReturn(evictedPoint); when(evictedPointSampled.getSequenceIndex()).thenReturn(evictedSequenceIndex); when(sampler.acceptPoint(sequenceIndex)).thenReturn(true); when(sampler.getEvictedPoint()).thenReturn(Optional.of(evictedPointSampled)); when(tree.addPoint(pointReference, sequenceIndex)).thenReturn(existingPointReference); UpdateResult<Integer> result = samplerPlusTree.update(pointReference, sequenceIndex); assertTrue(result.getAddedPoint().isPresent()); assertEquals(existingPointReference, result.getAddedPoint().get()); assertTrue(result.getDeletedPoint().isPresent()); assertEquals(evictedPoint, result.getDeletedPoint().get()); verify(tree, times(1)).deletePoint(evictedPoint, evictedSequenceIndex); verify(sampler, times(1)).addPoint(existingPointReference); } @Test public void testRejectPoint() { when(sampler.acceptPoint(anyLong())).thenReturn(false); UpdateResult<Integer> result = samplerPlusTree.update(2, 100L); assertFalse(result.isStateChange()); verify(tree, never()).addPoint(any(), anyLong()); verify(tree, never()).deletePoint(any(), anyLong()); verify(sampler, never()).addPoint(any()); } }
426
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/executor/ForestUpdateExecutorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.executor; import static com.amazon.randomcutforest.util.ArrayUtils.cleanCopy; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertNotSame; import static org.mockito.ArgumentMatchers.any; import static org.mockito.ArgumentMatchers.anyLong; import static org.mockito.ArgumentMatchers.eq; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.spy; import static org.mockito.Mockito.times; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.util.List; import java.util.stream.Stream; import org.junit.jupiter.api.extension.ExtendWith; import org.junit.jupiter.api.extension.ExtensionContext; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.Arguments; import org.junit.jupiter.params.provider.ArgumentsProvider; import org.junit.jupiter.params.provider.ArgumentsSource; import org.mockito.ArgumentCaptor; import org.mockito.Captor; import org.mockito.junit.jupiter.MockitoExtension; import com.amazon.randomcutforest.ComponentList; import com.amazon.randomcutforest.IComponentModel; import com.amazon.randomcutforest.store.PointStore; @ExtendWith(MockitoExtension.class) public class ForestUpdateExecutorTest { private static final int numberOfTrees = 10; private static final int threadPoolSize = 2; @Captor private ArgumentCaptor<List<UpdateResult<Integer>>> updateResultCaptor; private static class TestExecutorProvider implements ArgumentsProvider { @Override public Stream<? extends Arguments> provideArguments(ExtensionContext context) throws Exception { ComponentList<Integer, float[]> sequentialComponents = new ComponentList<>(); ComponentList<Integer, float[]> parallelComponents = new ComponentList<>(); for (int i = 0; i < numberOfTrees; i++) { sequentialComponents.add(mock(IComponentModel.class)); parallelComponents.add(mock(IComponentModel.class)); } PointStore pointStore = mock(PointStore.class); IStateCoordinator<Integer, float[]> sequentialUpdateCoordinator = spy( new PointStoreCoordinator<>(pointStore)); AbstractForestUpdateExecutor<Integer, float[]> sequentialExecutor = new SequentialForestUpdateExecutor<>( sequentialUpdateCoordinator, sequentialComponents); IStateCoordinator<Integer, float[]> parallelUpdateCoordinator = spy( new PointStoreCoordinator<>(pointStore)); AbstractForestUpdateExecutor<Integer, float[]> parallelExecutor = new ParallelForestUpdateExecutor<>( parallelUpdateCoordinator, parallelComponents, threadPoolSize); return Stream.of(sequentialExecutor, parallelExecutor).map(Arguments::of); } } @ParameterizedTest @ArgumentsSource(TestExecutorProvider.class) public void testUpdate(AbstractForestUpdateExecutor<Integer, float[]> executor) { int addAndDelete = 4; int addOnly = 4; ComponentList<Integer, ?> components = executor.components; for (int i = 0; i < addAndDelete; i++) { IComponentModel<Integer, ?> model = components.get(i); UpdateResult<Integer> result = new UpdateResult<>(i, 2 * i); when(model.update(any(), anyLong())).thenReturn(result); } for (int i = addAndDelete; i < addAndDelete + addOnly; i++) { IComponentModel<Integer, ?> model = components.get(i); UpdateResult<Integer> result = UpdateResult.<Integer>builder().addedPoint(i).build(); when(model.update(any(), anyLong())).thenReturn(result); } for (int i = addAndDelete + addOnly; i < numberOfTrees; i++) { IComponentModel<Integer, ?> model = components.get(i); when(model.update(any(), anyLong())).thenReturn(UpdateResult.noop()); } float[] point = new float[] { 1.0f }; executor.update(point); executor.components.forEach(model -> verify(model).update(any(), eq(0L))); IStateCoordinator<Integer, ?> coordinator = executor.updateCoordinator; verify(coordinator, times(1)).completeUpdate(updateResultCaptor.capture(), any()); List<UpdateResult<Integer>> updateResults = updateResultCaptor.getValue(); assertEquals(addAndDelete + addOnly, updateResults.size()); int actualAddAndAndDelete = 0; int actualAddOnly = 0; for (int i = 0; i < updateResults.size(); i++) { UpdateResult<Integer> result = updateResults.get(i); if (result.getDeletedPoint().isPresent()) { actualAddAndAndDelete++; } else { actualAddOnly++; } } assertEquals(addAndDelete, actualAddAndAndDelete); assertEquals(addOnly, actualAddOnly); } @ParameterizedTest @ArgumentsSource(TestExecutorProvider.class) public void testCleanCopy(AbstractForestUpdateExecutor<double[], ?> executor) { float[] point1 = new float[] { 1.0f, -22.2f, 30.9f }; float[] point1Copy = cleanCopy(point1); assertNotSame(point1, point1Copy); assertArrayEquals(point1, point1Copy); float[] point2 = new float[] { -0.0f, -22.2f, 30.9f }; float[] point2Copy = cleanCopy(point2); assertNotSame(point2, point2Copy); assertEquals(0.0, point2Copy[0]); point2Copy[0] = -0.0f; assertArrayEquals(point2, point2Copy); } }
427
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/executor/ForestTraversalExecutorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.executor; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertTrue; import static org.mockito.AdditionalMatchers.aryEq; import static org.mockito.ArgumentMatchers.any; import static org.mockito.Mockito.atMost; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.spy; import static org.mockito.Mockito.times; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.util.Arrays; import java.util.List; import java.util.stream.Stream; import org.junit.jupiter.api.extension.ExtensionContext; import org.junit.jupiter.params.ParameterizedTest; import org.junit.jupiter.params.provider.Arguments; import org.junit.jupiter.params.provider.ArgumentsProvider; import org.junit.jupiter.params.provider.ArgumentsSource; import com.amazon.randomcutforest.ComponentList; import com.amazon.randomcutforest.IComponentModel; import com.amazon.randomcutforest.TestUtils; import com.amazon.randomcutforest.returntypes.ConvergingAccumulator; import com.amazon.randomcutforest.sampler.CompactSampler; import com.amazon.randomcutforest.tree.ITree; import com.amazon.randomcutforest.tree.RandomCutTree; public class ForestTraversalExecutorTest { private static int numberOfTrees = 10; private static int threadPoolSize = 2; private static class TestExecutorProvider implements ArgumentsProvider { @Override public Stream<? extends Arguments> provideArguments(ExtensionContext context) throws Exception { ComponentList<Integer, float[]> sequentialExecutors = new ComponentList<>(); ComponentList<Integer, float[]> parallelExecutors = new ComponentList<>(); for (int i = 0; i < numberOfTrees; i++) { CompactSampler sampler = mock(CompactSampler.class); RandomCutTree tree = mock(RandomCutTree.class); sequentialExecutors.add(spy(new SamplerPlusTree<>(sampler, tree))); } for (int i = 0; i < numberOfTrees; i++) { CompactSampler sampler = mock(CompactSampler.class); RandomCutTree tree = mock(RandomCutTree.class); parallelExecutors.add(spy(new SamplerPlusTree<>(sampler, tree))); } SequentialForestTraversalExecutor sequentialExecutor = new SequentialForestTraversalExecutor( sequentialExecutors); ParallelForestTraversalExecutor parallelExecutor = new ParallelForestTraversalExecutor(parallelExecutors, threadPoolSize); return Stream.of(sequentialExecutor, parallelExecutor).map(Arguments::of); } } @ParameterizedTest @ArgumentsSource(TestExecutorProvider.class) public void testTraverseForestBinaryAccumulator(AbstractForestTraversalExecutor executor) { float[] point = new float[] { 1.2f, -3.4f }; double expectedResult = 0.0; for (int i = 0; i < numberOfTrees; i++) { double treeResult = Math.random(); ITree<?, ?> tree = ((SamplerPlusTree<?, ?>) executor.components.get(i)).getTree(); when(tree.traverse(aryEq(point), any())).thenReturn(treeResult); expectedResult += treeResult; } expectedResult /= numberOfTrees; double result = executor.traverseForest(point, TestUtils.DUMMY_GENERIC_VISITOR_FACTORY, Double::sum, x -> x / 10.0); for (IComponentModel<?, ?> component : executor.components) { verify(component, times(1)).traverse(aryEq(point), any()); } assertEquals(expectedResult, result, EPSILON); } @ParameterizedTest @ArgumentsSource(TestExecutorProvider.class) public void testTraverseForestCollector(AbstractForestTraversalExecutor executor) { float[] point = new float[] { 1.2f, -3.4f }; double[] expectedResult = new double[numberOfTrees]; for (int i = 0; i < numberOfTrees; i++) { double treeResult = Math.random(); ITree<?, ?> tree = ((SamplerPlusTree<?, ?>) executor.components.get(i)).getTree(); when(tree.traverse(aryEq(point), any())).thenReturn(treeResult); expectedResult[i] = treeResult; } Arrays.sort(expectedResult); List<Double> result = executor.traverseForest(point, TestUtils.DUMMY_GENERIC_VISITOR_FACTORY, TestUtils.SORTED_LIST_COLLECTOR); for (IComponentModel<?, ?> component : executor.components) { verify(component, times(1)).traverse(aryEq(point), any()); } assertEquals(numberOfTrees, result.size()); for (int i = 0; i < numberOfTrees; i++) { assertEquals(expectedResult[i], result.get(i), EPSILON); } } @ParameterizedTest @ArgumentsSource(TestExecutorProvider.class) public void testTraverseForestConverging(AbstractForestTraversalExecutor executor) { float[] point = new float[] { 1.2f, -3.4f }; for (int i = 0; i < numberOfTrees; i++) { double treeResult = Math.random(); ITree<?, ?> tree = ((SamplerPlusTree<?, ?>) executor.components.get(i)).getTree(); when(tree.traverse(aryEq(point), any())).thenReturn(treeResult); } int convergenceThreshold = numberOfTrees / 2; ConvergingAccumulator<Double> accumulator = TestUtils.convergeAfter(convergenceThreshold); double result = executor.traverseForest(point, TestUtils.DUMMY_GENERIC_VISITOR_FACTORY, accumulator, x -> x / accumulator.getValuesAccepted()); for (IComponentModel<?, ?> component : executor.components) { verify(component, atMost(1)).traverse(aryEq(point), any()); } assertTrue(accumulator.getValuesAccepted() >= convergenceThreshold); assertTrue(accumulator.getValuesAccepted() < numberOfTrees); assertEquals(accumulator.getAccumulatedValue() / accumulator.getValuesAccepted(), result, EPSILON); } @ParameterizedTest @ArgumentsSource(TestExecutorProvider.class) public void testTraverseForestMultiBinaryAccumulator(AbstractForestTraversalExecutor executor) { float[] point = new float[] { 1.2f, -3.4f }; double expectedResult = 0.0; for (int i = 0; i < numberOfTrees; i++) { double treeResult = Math.random(); ITree<?, ?> tree = ((SamplerPlusTree<?, ?>) executor.components.get(i)).getTree(); when(tree.traverseMulti(aryEq(point), any())).thenReturn(treeResult); expectedResult += treeResult; } expectedResult /= numberOfTrees; double result = executor.traverseForestMulti(point, TestUtils.DUMMY_GENERIC_MULTI_VISITOR_FACTORY, Double::sum, x -> x / 10.0); for (IComponentModel<?, ?> component : executor.components) { verify(component, times(1)).traverseMulti(aryEq(point), any()); } assertEquals(expectedResult, result, EPSILON); } @ParameterizedTest @ArgumentsSource(TestExecutorProvider.class) public void testTraverseForestMultiCollector(AbstractForestTraversalExecutor executor) { float[] point = new float[] { 1.2f, -3.4f }; double[] expectedResult = new double[numberOfTrees]; for (int i = 0; i < numberOfTrees; i++) { double treeResult = Math.random(); ITree<?, ?> tree = ((SamplerPlusTree<?, ?>) executor.components.get(i)).getTree(); when(tree.traverseMulti(aryEq(point), any())).thenReturn(treeResult); expectedResult[i] = treeResult; } Arrays.sort(expectedResult); List<Double> result = executor.traverseForestMulti(point, TestUtils.DUMMY_GENERIC_MULTI_VISITOR_FACTORY, TestUtils.SORTED_LIST_COLLECTOR); for (IComponentModel<?, ?> component : executor.components) { verify(component, times(1)).traverseMulti(aryEq(point), any()); } assertEquals(numberOfTrees, result.size()); for (int i = 0; i < numberOfTrees; i++) { assertEquals(expectedResult[i], result.get(i), EPSILON); } } }
428
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/executor/PointStoreCoordinatorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.executor; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.mockito.ArgumentMatchers.anyLong; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.times; import static org.mockito.Mockito.verify; import static org.mockito.Mockito.when; import java.util.ArrayList; import java.util.Collections; import java.util.List; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import org.mockito.ArgumentCaptor; import com.amazon.randomcutforest.store.PointStore; public class PointStoreCoordinatorTest { private PointStore store; private PointStoreCoordinator coordinator; @BeforeEach public void setUp() { store = mock(PointStore.class); coordinator = new PointStoreCoordinator(store); } @Test public void testInitUpdate() { float[] point = { 1.2f, -3.4f }; int index = 123; ArgumentCaptor<float[]> captor = ArgumentCaptor.forClass(float[].class); when(store.add(captor.capture(), anyLong())).thenReturn(index); int result = coordinator.initUpdate(point, 0); verify(store, times(1)).add(point, 0); assertEquals(result, index); } @Test public void testCompleteUpdate() { List<UpdateResult<Integer>> updateResults = new ArrayList<>(); UpdateResult<Integer> result1 = UpdateResult.<Integer>builder().addedPoint(1).deletedPoint(100).build(); updateResults.add(result1); UpdateResult<Integer> result2 = UpdateResult.<Integer>builder().addedPoint(2).deletedPoint(200).build(); updateResults.add(result2); UpdateResult<Integer> result3 = UpdateResult.<Integer>builder().addedPoint(3).build(); updateResults.add(result3); UpdateResult<Integer> result4 = UpdateResult.noop(); updateResults.add(result4); // order shouldn't matter Collections.shuffle(updateResults); Integer updateInput = 1000; coordinator.completeUpdate(updateResults, updateInput); ArgumentCaptor<Integer> captor1 = ArgumentCaptor.forClass(Integer.class); verify(store, times(3)).incrementRefCount(captor1.capture()); List<Integer> arguments = captor1.getAllValues(); Collections.sort(arguments); assertEquals(1, arguments.get(0)); assertEquals(2, arguments.get(1)); assertEquals(3, arguments.get(2)); ArgumentCaptor<Integer> captor2 = ArgumentCaptor.forClass(Integer.class); verify(store, times(3)).decrementRefCount(captor2.capture()); arguments = captor2.getAllValues(); Collections.sort(arguments); assertEquals(100, arguments.get(0)); assertEquals(200, arguments.get(1)); assertEquals(1000, arguments.get(2)); } }
429
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/RangeVectorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertThrows; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class RangeVectorTest { int dimensions; private RangeVector vector; @BeforeEach public void setUp() { dimensions = 3; vector = new RangeVector(dimensions); } @Test public void testNew() { assertThrows(IllegalArgumentException.class, () -> new RangeVector(0)); assertThrows(IllegalArgumentException.class, () -> new RangeVector(new float[0])); float[] expected = new float[dimensions]; assertArrayEquals(expected, vector.values); assertArrayEquals(expected, vector.upper); assertArrayEquals(expected, vector.lower); float[] another = new float[0]; assertThrows(IllegalArgumentException.class, () -> new RangeVector(another, another, another)); assertThrows(IllegalArgumentException.class, () -> new RangeVector(expected, expected, new float[dimensions + 1])); assertThrows(IllegalArgumentException.class, () -> new RangeVector(expected, new float[dimensions + 1], expected)); assertThrows(IllegalArgumentException.class, () -> new RangeVector(new float[dimensions + 1], expected, expected)); assertDoesNotThrow(() -> new RangeVector(expected, expected, expected)); assertThrows(IllegalArgumentException.class, () -> new RangeVector(expected, new float[] { -1f, 0f, 0f }, expected)); assertDoesNotThrow(() -> new RangeVector(expected, expected, new float[] { -1f, 0f, 0f })); assertThrows(IllegalArgumentException.class, () -> new RangeVector(expected, new float[] { 1f, 0f, 0f }, new float[] { 1f, 0f, 0f })); assertDoesNotThrow(() -> new RangeVector(expected, new float[] { 1f, 0f, 0f }, new float[] { -1f, 0f, 0f })); } @Test public void testScale() { vector.upper[0] = 1.1f; vector.upper[2] = 3.1f; vector.upper[1] = 3.1f; vector.lower[1] = -2.2f; float z = 9.9f; assertThrows(IllegalArgumentException.class, () -> vector.scale(0, -1.0f)); assertThrows(IllegalArgumentException.class, () -> vector.scale(-1, 1.0f)); assertThrows(IllegalArgumentException.class, () -> vector.scale(dimensions + 1, 1.0f)); vector.scale(0, z); float[] expected = new float[] { 1.1f * 9.9f, 3.1f, 3.1f }; assertArrayEquals(expected, vector.upper, 1e-6f); expected = new float[] { 0.0f, -2.2f, 0.0f }; assertArrayEquals(expected, vector.lower); vector.scale(1, 2 * z); assertArrayEquals(new float[] { 1.1f * 9.9f, 3.1f * 2 * z, 3.1f }, vector.upper, 1e-6f); assertArrayEquals(new float[] { 0f, -2.2f * 2 * z, 0f }, vector.lower, 1e-6f); } @Test public void testShift() { vector.upper[0] = 1.1f; vector.upper[2] = 3.1f; vector.lower[1] = -2.2f; float z = -9.9f; assertThrows(IllegalArgumentException.class, () -> vector.shift(-1, z)); assertThrows(IllegalArgumentException.class, () -> vector.shift(dimensions + 1, z)); vector.shift(0, z); float[] expected = new float[] { 1.1f - 9.9f, 0.0f, 3.1f }; assertArrayEquals(expected, vector.upper, 1e-6f); expected = new float[] { z, -2.2f, 0.0f }; assertArrayEquals(expected, vector.lower); assertArrayEquals(new float[] { z, 0, 0 }, vector.values, 1e-6f); } }
430
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/DiVectorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertNull; import static org.junit.jupiter.api.Assertions.assertSame; import static org.junit.jupiter.api.Assertions.assertThrows; import java.util.Arrays; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.state.returntypes.DiVectorMapper; import com.amazon.randomcutforest.state.returntypes.DiVectorState; public class DiVectorTest { int dimensions; private DiVector vector; @BeforeEach public void setUp() { dimensions = 3; vector = new DiVector(dimensions); } @Test public void testNew() { double[] expected = new double[dimensions]; assertEquals(dimensions, vector.getDimensions()); assertArrayEquals(expected, vector.high); assertArrayEquals(expected, vector.low); assertThrows(IllegalArgumentException.class, () -> new DiVector(0)); assertThrows(IllegalArgumentException.class, () -> new DiVector(new double[10], new double[9])); assertDoesNotThrow(() -> new DiVector(new double[10], new double[10])); } @Test public void testAddToLeft() { DiVector left = new DiVector(dimensions); DiVector right = new DiVector(dimensions); for (int i = 0; i < dimensions; i++) { left.low[i] = Math.random(); left.high[i] = Math.random(); right.low[i] = Math.random(); right.high[i] = Math.random(); } assertThrows(IllegalArgumentException.class, () -> DiVector.addToLeft(left, new DiVector(dimensions + 1))); DiVector leftCopy = new DiVector(dimensions); System.arraycopy(left.low, 0, leftCopy.low, 0, dimensions); System.arraycopy(left.high, 0, leftCopy.high, 0, dimensions); DiVector rightCopy = new DiVector(dimensions); System.arraycopy(right.low, 0, rightCopy.low, 0, dimensions); System.arraycopy(right.high, 0, rightCopy.high, 0, dimensions); DiVector result = DiVector.addToLeft(left, right); assertSame(result, left); assertArrayEquals(rightCopy.low, right.low); assertArrayEquals(rightCopy.high, right.high); for (int i = 0; i < dimensions; i++) { assertEquals(leftCopy.low[i] + right.low[i], left.low[i]); assertEquals(leftCopy.high[i] + right.high[i], left.high[i]); } } @Test public void testScale() { vector.high[0] = 1.1; vector.high[2] = 3.1; vector.low[1] = 2.2; double z = 9.9; DiVector result = vector.scale(z); double[] expected = new double[] { 1.1 * 9.9, 0.0, 3.1 * 9.9 }; assertArrayEquals(expected, result.high); expected = new double[] { 0.0, 2.2 * 9.9, 0.0 }; assertArrayEquals(expected, result.low); DiVector emptyVector = new DiVector(dimensions); emptyVector.scale(123.0); expected = new double[dimensions]; assertArrayEquals(expected, emptyVector.low); assertArrayEquals(expected, emptyVector.high); } @Test public void testGetHighLowSum() { vector.high[2] = 3.1; vector.low[1] = 2.2; assertEquals(3.1 + 2.2, vector.getHighLowSum()); } @Test public void testRenormalize() { DiVector testVector = new DiVector(10); // cannot renormalize really testVector.renormalize(100); assertEquals(testVector.getHighLowSum(), 0); vector.high[0] = 1.1; vector.high[2] = 3.1; vector.low[1] = 2.2; assertEquals(1.1 + 3.1 + 2.2, vector.getHighLowSum()); vector.renormalize(100.0); assertEquals(100.0, vector.getHighLowSum()); } @Test public void testComponentwiseTransform() { vector.high[0] = 1.1; vector.high[1] = 2.1; vector.high[2] = 3.1; vector.low[0] = 101.1; vector.low[1] = 202.1; vector.low[2] = 303.1; double[] highCopy = Arrays.copyOf(vector.high, dimensions); double[] lowCopy = Arrays.copyOf(vector.low, dimensions); vector.componentwiseTransform(x -> 2 * x - 1); for (int i = 0; i < dimensions; i++) { assertEquals(2 * highCopy[i] - 1, vector.high[i], EPSILON); assertEquals(2 * lowCopy[i] - 1, vector.low[i], EPSILON); } } @Test public void testMapper() { DiVector left = new DiVector(dimensions); for (int i = 0; i < dimensions; i++) { left.low[i] = Math.random(); left.high[i] = Math.random(); } DiVectorMapper mapper = new DiVectorMapper(); DiVector another = mapper.toModel(mapper.toState(left)); assertArrayEquals(another.high, left.high, 1e-10); assertArrayEquals(another.low, left.low, 1e-10); assertNull(mapper.toModel(mapper.toState(null))); DiVectorState state = new DiVectorState(); state.setHigh(left.high); assertNull(mapper.toModel(state)); state.setHigh(null); state.setLow(left.low); assertNull(mapper.toModel(state)); } }
431
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/OneSidedConvergingDoubleAccumulatorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertTrue; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; /** * This test doubles as a test of the abstract OneSidedStdDevAccumulator class */ public class OneSidedConvergingDoubleAccumulatorTest { private boolean highIsCritical; private double precision; private int minValuesAccepted; private int maxValuesAccepted; private OneSidedConvergingDoubleAccumulator accumulator; @BeforeEach public void setUp() { highIsCritical = true; precision = 0.1; minValuesAccepted = 5; maxValuesAccepted = 100; accumulator = new OneSidedConvergingDoubleAccumulator(highIsCritical, precision, minValuesAccepted, maxValuesAccepted); } @Test public void testGetConvergingValue() { assertEquals(1.23, accumulator.getConvergingValue(1.23)); assertEquals(-1001.1001, accumulator.getConvergingValue(-1001.1001)); } @Test public void testAccumulateValue() { double sum = 0.0; for (int i = 0; i < 10; i++) { double value = Math.random(); accumulator.accept(value); sum += value; assertEquals(sum, accumulator.getAccumulatedValue()); } } @Test public void testConvergenceHighIsCritical() { accumulator.accept(0.0); accumulator.accept(10.0); accumulator.accept(0.0); accumulator.accept(10.0); // less than minValuesAccepted assertEquals(4, accumulator.getValuesAccepted()); assertFalse(accumulator.isConverged()); double expectedSum = 20.0; assertEquals(expectedSum, accumulator.getAccumulatedValue()); // each high value should result in a witness to convergence // we need 1.0 / precision witnesses in order to converge for (int i = 0; i < 1.0 / precision - 1; i++) { accumulator.accept(0.0); accumulator.accept(10.0); assertEquals(6 + 2 * i, accumulator.getValuesAccepted()); assertFalse(accumulator.isConverged()); expectedSum += 10.0; assertEquals(expectedSum, accumulator.getAccumulatedValue()); } accumulator.accept(0.0); assertFalse(accumulator.isConverged()); // the last required high value accumulator.accept(10.0); assertTrue(accumulator.isConverged()); expectedSum += 10.0; assertEquals(expectedSum, accumulator.getAccumulatedValue()); } @Test public void testConvergenceLowIsCritical() { highIsCritical = false; accumulator = new OneSidedConvergingDoubleAccumulator(highIsCritical, precision, minValuesAccepted, maxValuesAccepted); accumulator.accept(0.0); accumulator.accept(10.0); accumulator.accept(0.0); accumulator.accept(10.0); // less than minValuesAccepted assertFalse(accumulator.isConverged()); double expectedSum = 20.0; assertEquals(expectedSum, accumulator.getAccumulatedValue()); // each high value should result in a witness to convergence // we need 1.0 / precision witnesses in order to converge for (int i = 0; i < 1.0 / precision - 1; i++) { accumulator.accept(0.0); accumulator.accept(10.0); assertFalse(accumulator.isConverged()); expectedSum += 10.0; assertEquals(expectedSum, accumulator.getAccumulatedValue()); } accumulator.accept(10.0); assertFalse(accumulator.isConverged()); // the last required low value accumulator.accept(0.0); assertTrue(accumulator.isConverged()); expectedSum += 10.0; assertEquals(expectedSum, accumulator.getAccumulatedValue()); } }
432
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/SampleSummaryTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static com.amazon.randomcutforest.CommonUtils.toFloatArray; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertThrows; import java.util.ArrayList; import java.util.Arrays; import java.util.Collections; import java.util.Random; import java.util.function.BiFunction; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.summarization.Summarizer; import com.amazon.randomcutforest.testutils.NormalMixtureTestData; import com.amazon.randomcutforest.util.Weighted; public class SampleSummaryTest { /** * this class tests the return type data structure whereas * randomcutforest.SampleSummaryTest tests tha summarization algorithms. */ int dataSize = 20000; int newDimensions = 2; Random random = new Random(); @Test public void testConstructor() { assertThrows(IllegalArgumentException.class, () -> new SampleSummary(Collections.emptyList(), 0.6)); float[][] points = getData(dataSize, newDimensions, random.nextInt(), Summarizer::L2distance); ArrayList<Weighted<float[]>> weighted = new ArrayList<>(); for (float[] point : points) { // testing 0 weight weighted.add(new Weighted<>(point, 0.0f)); } assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted, 0.1)); assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted, 1.3)); assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted)); weighted.get(0).weight = Float.NaN; assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted)); weighted.get(0).weight = Float.POSITIVE_INFINITY; assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted)); weighted.get(0).weight = -1.0f; assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted)); weighted.get(0).weight = 1.0f; assertDoesNotThrow(() -> new SampleSummary(weighted)); weighted.get(1).index = new float[newDimensions + 1]; assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted)); weighted.get(1).index = new float[newDimensions]; weighted.get(1).index[0] = Float.NaN; assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted)); weighted.get(1).index[0] = Float.NEGATIVE_INFINITY; assertThrows(IllegalArgumentException.class, () -> new SampleSummary(weighted)); weighted.get(1).index[0] = -1.0f; SampleSummary summary = new SampleSummary(weighted); } @Test public void addTypicalTest() { float[][] points = getData(dataSize, newDimensions, random.nextInt(), Summarizer::L2distance); ArrayList<Weighted<float[]>> weighted = new ArrayList<>(); for (float[] point : points) { // testing 0 weight weighted.add(new Weighted<>(point, 1.0f)); } SampleSummary summary = new SampleSummary(weighted); assertThrows(IllegalArgumentException.class, () -> summary.addTypical(new float[1][2], new float[2])); assertDoesNotThrow(() -> summary.addTypical(new float[0][2], new float[0])); assertDoesNotThrow(() -> summary.addTypical(new float[2][4], new float[2])); assertThrows(IllegalArgumentException.class, () -> summary.addTypical(new float[][] { new float[2], new float[3] }, new float[2])); assertThrows(IllegalArgumentException.class, () -> summary.addTypical(new float[][] { new float[2], new float[3] }, new float[2])); } public float[][] getData(int dataSize, int newDimensions, int seed, BiFunction<float[], float[], Double> distance) { double baseMu = 0.0; double baseSigma = 1.0; double anomalyMu = 0.0; double anomalySigma = 1.0; double transitionToAnomalyProbability = 0.0; // ignoring anomaly cluster for now double transitionToBaseProbability = 1.0; Random prg = new Random(0); NormalMixtureTestData generator = new NormalMixtureTestData(baseMu, baseSigma, anomalyMu, anomalySigma, transitionToAnomalyProbability, transitionToBaseProbability); double[][] data = generator.generateTestData(dataSize, newDimensions, seed); float[][] floatData = new float[dataSize][]; float[] allZero = new float[newDimensions]; float[] sigma = new float[newDimensions]; Arrays.fill(sigma, 1f); double scale = distance.apply(allZero, sigma); for (int i = 0; i < dataSize; i++) { // shrink, shift at random int nextD = prg.nextInt(newDimensions); for (int j = 0; j < newDimensions; j++) { data[i][j] *= 1.0 / (3.0); // standard deviation adds up across dimension; taking square root // and using s 3 sigma ball if (j == nextD) { if (prg.nextDouble() < 0.5) data[i][j] += 2.0 * scale; else data[i][j] -= 2.0 * scale; } } floatData[i] = toFloatArray(data[i]); } return floatData; } }
433
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/InterpolationMeasureTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertDoesNotThrow; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertThrows; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class InterpolationMeasureTest { private int dimensions; private int sampleSize; private InterpolationMeasure output; @BeforeEach public void setUp() { dimensions = 3; sampleSize = 99; output = new InterpolationMeasure(dimensions, sampleSize); } @Test public void testNew() { double[] zero = new double[3]; assertArrayEquals(zero, output.measure.high); assertArrayEquals(zero, output.distances.high); assertArrayEquals(zero, output.probMass.high); assertArrayEquals(zero, output.measure.low); assertArrayEquals(zero, output.distances.low); assertArrayEquals(zero, output.probMass.low); assertEquals(output.getSampleSize(), sampleSize); assertThrows(IllegalArgumentException.class, () -> new InterpolationMeasure(0, sampleSize)); assertThrows(IllegalArgumentException.class, () -> new InterpolationMeasure(1, new DiVector(1), new DiVector(2), new DiVector(3))); assertThrows(IllegalArgumentException.class, () -> new InterpolationMeasure(1, new DiVector(2), new DiVector(2), new DiVector(3))); assertDoesNotThrow(() -> new InterpolationMeasure(1, new DiVector(2), new DiVector(2), new DiVector(2))); } @Test public void testAddToLeft() { InterpolationMeasure other1 = new InterpolationMeasure(dimensions, sampleSize); InterpolationMeasure other2 = new InterpolationMeasure(dimensions, sampleSize); assertThrows(IllegalArgumentException.class, () -> InterpolationMeasure.addToLeft(other1, new InterpolationMeasure(dimensions + 1, sampleSize))); for (int i = 0; i < dimensions; i++) { output.probMass.high[i] = 2 * i; output.probMass.low[i] = 2 * i + 1; output.distances.high[i] = 4 * i; output.distances.low[i] = 4 * i + 2; output.measure.high[i] = 6 * i; output.measure.low[i] = 6 * i + 3; other1.probMass.high[i] = other2.probMass.high[i] = 8 * i; other1.distances.high[i] = other2.distances.high[i] = 10 * i; other1.measure.high[i] = other2.measure.high[i] = 12 * i; other1.probMass.low[i] = other2.probMass.low[i] = 8 * i + 4; other1.distances.low[i] = other2.distances.low[i] = 10 * i + 5; other1.measure.low[i] = other2.measure.low[i] = 12 * i + 6; } assertArrayEquals(other1.probMass.high, other2.probMass.high); assertArrayEquals(other1.distances.high, other2.distances.high); assertArrayEquals(other1.measure.high, other2.measure.high); assertArrayEquals(other1.probMass.low, other2.probMass.low); assertArrayEquals(other1.distances.low, other2.distances.low); assertArrayEquals(other1.measure.low, other2.measure.low); InterpolationMeasure.addToLeft(output, other1); for (int i = 0; i < dimensions; i++) { assertEquals(2 * i + 8 * i, output.probMass.high[i]); assertEquals(4 * i + 10 * i, output.distances.high[i]); assertEquals(6 * i + 12 * i, output.measure.high[i]); assertEquals(2 * i + 8 * i + 5, output.probMass.low[i]); assertEquals(4 * i + 10 * i + 7, output.distances.low[i]); assertEquals(6 * i + 12 * i + 9, output.measure.low[i]); } assertArrayEquals(other1.probMass.high, other2.probMass.high); assertArrayEquals(other1.distances.high, other2.distances.high); assertArrayEquals(other1.measure.high, other2.measure.high); assertArrayEquals(other1.probMass.low, other2.probMass.low); assertArrayEquals(other1.distances.low, other2.distances.low); assertArrayEquals(other1.measure.low, other2.measure.low); } @Test public void testScale() { InterpolationMeasure copy = new InterpolationMeasure(dimensions, sampleSize); for (int i = 0; i < dimensions; i++) { output.probMass.high[i] = copy.probMass.high[i] = 2 * i; output.distances.high[i] = copy.distances.high[i] = 4 * i; output.measure.high[i] = copy.measure.high[i] = 6 * i; output.probMass.low[i] = copy.probMass.low[i] = 2 * i + 1; output.distances.low[i] = copy.distances.low[i] = 4 * i + 2; output.measure.low[i] = copy.measure.low[i] = 6 * i + 3; } assertArrayEquals(copy.probMass.high, output.probMass.high); assertArrayEquals(copy.distances.high, output.distances.high); assertArrayEquals(copy.measure.high, output.measure.high); assertArrayEquals(copy.probMass.low, output.probMass.low); assertArrayEquals(copy.distances.low, output.distances.low); assertArrayEquals(copy.measure.low, output.measure.low); InterpolationMeasure result = output.scale(0.9); assertArrayEquals(copy.probMass.low, output.probMass.low); assertArrayEquals(copy.distances.low, output.distances.low); assertArrayEquals(copy.measure.low, output.measure.low); assertArrayEquals(copy.probMass.high, output.probMass.high); assertArrayEquals(copy.distances.high, output.distances.high); assertArrayEquals(copy.measure.high, output.measure.high); for (int i = 0; i < dimensions; i++) { assertEquals(2 * i * 0.9, result.probMass.high[i]); assertEquals(4 * i * 0.9, result.distances.high[i]); assertEquals(6 * i * 0.9, result.measure.high[i]); assertEquals((2 * i + 1) * 0.9, result.probMass.low[i]); assertEquals((4 * i + 2) * 0.9, result.distances.low[i]); assertEquals((6 * i + 3) * 0.9, result.measure.low[i]); } } }
434
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/NeighborTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static org.hamcrest.MatcherAssert.assertThat; import static org.hamcrest.Matchers.containsInAnyOrder; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import java.util.ArrayList; import java.util.List; import org.junit.jupiter.api.Test; public class NeighborTest { @Test public void testNew() { float[] point = new float[] { 1.0f, -2.0f, 3.3f }; double distance = 1234.5; List<Long> timestamps = new ArrayList<>(); timestamps.add(99999L); timestamps.add(99L); Neighbor neighbor = new Neighbor(point, distance, timestamps); assertArrayEquals(point, neighbor.point); assertEquals(distance, neighbor.distance); assertThat(neighbor.sequenceIndexes, containsInAnyOrder(timestamps.toArray())); } }
435
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/DensityOutputTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertTrue; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class DensityOutputTest { private int dimensions; private int sampleSize; private DensityOutput output; @BeforeEach public void setUp() { dimensions = 3; sampleSize = 99; output = new DensityOutput(dimensions, sampleSize); } @Test public void testNew() { double[] zero = new double[3]; assertArrayEquals(zero, output.measure.high); assertArrayEquals(zero, output.distances.high); assertArrayEquals(zero, output.probMass.high); assertArrayEquals(zero, output.measure.low); assertArrayEquals(zero, output.distances.low); assertArrayEquals(zero, output.probMass.low); } @Test public void testAddToLeft() { DensityOutput other1 = new DensityOutput(dimensions, sampleSize); DensityOutput other2 = new DensityOutput(dimensions, sampleSize); for (int i = 0; i < dimensions; i++) { output.probMass.high[i] = 2 * i; output.probMass.low[i] = 2 * i + 1; output.distances.high[i] = 4 * i; output.distances.low[i] = 4 * i + 2; output.measure.high[i] = 6 * i; output.measure.low[i] = 6 * i + 3; other1.probMass.high[i] = other2.probMass.high[i] = 8 * i; other1.distances.high[i] = other2.distances.high[i] = 10 * i; other1.measure.high[i] = other2.measure.high[i] = 12 * i; other1.probMass.low[i] = other2.probMass.low[i] = 8 * i + 4; other1.distances.low[i] = other2.distances.low[i] = 10 * i + 5; other1.measure.low[i] = other2.measure.low[i] = 12 * i + 6; } assertArrayEquals(other1.probMass.high, other2.probMass.high); assertArrayEquals(other1.distances.high, other2.distances.high); assertArrayEquals(other1.measure.high, other2.measure.high); assertArrayEquals(other1.probMass.low, other2.probMass.low); assertArrayEquals(other1.distances.low, other2.distances.low); assertArrayEquals(other1.measure.low, other2.measure.low); DensityOutput.addToLeft(output, other1); for (int i = 0; i < dimensions; i++) { assertEquals(2 * i + 8 * i, output.probMass.high[i]); assertEquals(4 * i + 10 * i, output.distances.high[i]); assertEquals(6 * i + 12 * i, output.measure.high[i]); assertEquals(2 * i + 8 * i + 5, output.probMass.low[i]); assertEquals(4 * i + 10 * i + 7, output.distances.low[i]); assertEquals(6 * i + 12 * i + 9, output.measure.low[i]); } assertArrayEquals(other1.probMass.high, other2.probMass.high); assertArrayEquals(other1.distances.high, other2.distances.high); assertArrayEquals(other1.measure.high, other2.measure.high); assertArrayEquals(other1.probMass.low, other2.probMass.low); assertArrayEquals(other1.distances.low, other2.distances.low); assertArrayEquals(other1.measure.low, other2.measure.low); } @Test public void testGetDensity() { assertTrue(output.getDensity(0.5, 3) == 0); for (int i = 0; i < dimensions; i++) { output.probMass.high[i] = 2 * i; output.distances.high[i] = 4 * i; output.measure.high[i] = 6 * i; output.probMass.low[i] = 2 * i; output.distances.low[i] = 4 * i + 2; output.measure.low[i] = 6 * i + 3; } double q = 0.5; double density = output.getDensity(q, 3); DiVector densityVector = output.getDirectionalDensity(q, 3); double sumOfPoints = output.measure.getHighLowSum() / sampleSize; double sumOfFactors = 0.0; for (int i = 0; i < dimensions; i++) { double mass = output.probMass.getHighLowSum(i); double distance = output.distances.getHighLowSum(i); double t = (mass != 0) ? distance / mass : 0; t = Math.pow(t, dimensions) * mass; sumOfFactors += t; } assertEquals(sumOfPoints / (q * sumOfPoints + sumOfFactors), density, EPSILON); // for contrib, do not scale sum of points by sample size sumOfPoints = output.measure.getHighLowSum(); for (int i = 0; i < dimensions; i++) { assertEquals(output.measure.high[i] * density / sumOfPoints, densityVector.high[i], EPSILON); assertEquals(output.measure.low[i] * density / sumOfPoints, densityVector.low[i], EPSILON); } assertEquals(output.getDensity(DensityOutput.DEFAULT_SUM_OF_POINTS_SCALING_FACTOR, dimensions), output.getDensity()); densityVector = output.getDirectionalDensity(DensityOutput.DEFAULT_SUM_OF_POINTS_SCALING_FACTOR, dimensions); DiVector defaultDensityVector = output.getDirectionalDensity(); for (int i = 0; i < dimensions; i++) { assertEquals(densityVector.high[i], defaultDensityVector.high[i], EPSILON); assertEquals(densityVector.low[i], defaultDensityVector.low[i], EPSILON); } } }
436
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/returntypes/OneSidedConvergingDiVectorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.returntypes; import static com.amazon.randomcutforest.TestUtils.EPSILON; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class OneSidedConvergingDiVectorTest { private boolean highIsCritical; private double precision; private int minValuesAccepted; private int maxValuesAccepted; private int dimensions; private OneSidedConvergingDiVectorAccumulator accumulator; @BeforeEach public void setUp() { highIsCritical = true; precision = 0.1; minValuesAccepted = 5; maxValuesAccepted = 100; dimensions = 2; accumulator = new OneSidedConvergingDiVectorAccumulator(dimensions, highIsCritical, precision, minValuesAccepted, maxValuesAccepted); } @Test public void testGetConvergingValue() { DiVector vector = new DiVector(dimensions); vector.high[0] = 1.1; vector.low[1] = 2.3; vector.high[1] = 9.6; assertEquals(1.1 + 2.3 + 9.6, accumulator.getConvergingValue(vector), EPSILON); } @Test public void testAccumulateValue() { assertEquals(accumulator.getWitnesses(), 0); assertEquals(accumulator.getMean(), 0); assertEquals(accumulator.getDeviation(), 0); DiVector vector1 = new DiVector(dimensions); vector1.high[0] = 1.1; vector1.low[1] = 2.3; vector1.high[1] = 9.6; accumulator.accept(vector1); DiVector result = accumulator.getAccumulatedValue(); assertArrayEquals(vector1.high, result.high, EPSILON); assertArrayEquals(vector1.low, result.low, EPSILON); DiVector vector2 = new DiVector(dimensions); vector2.high[0] = 1.1; vector2.low[1] = 2.3; vector2.high[1] = 9.6; accumulator.accept(vector2); result = accumulator.getAccumulatedValue(); DiVector.addToLeft(vector1, vector2); assertArrayEquals(vector1.high, result.high, EPSILON); assertArrayEquals(vector1.low, result.low, EPSILON); for (int i = 0; i < 5; i++) { accumulator.accept(vector2); } assertEquals(accumulator.getWitnesses(), 3); assertEquals(accumulator.getDeviation(), 0, 1e-6f); assertEquals(accumulator.getMean(), 13, 1e-6f); } }
437
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/interpolation/SimpleInterpolationVisitorTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.interpolation; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.mockito.ArgumentMatchers.any; import static org.mockito.Mockito.mock; import static org.mockito.Mockito.when; import java.util.Arrays; import org.junit.jupiter.api.Test; import com.amazon.randomcutforest.returntypes.InterpolationMeasure; import com.amazon.randomcutforest.tree.BoundingBox; import com.amazon.randomcutforest.tree.INodeView; import com.amazon.randomcutforest.tree.NodeView; public class SimpleInterpolationVisitorTest { private static final int SEED = 1002; @Test public void testNew() { float[] point = { 1.0f, 2.0f }; int sampleSize = 9; SimpleInterpolationVisitor visitor = new SimpleInterpolationVisitor(point, sampleSize, 1, false); assertFalse(visitor.pointInsideBox); assertEquals(2, visitor.coordInsideBox.length); for (int i = 0; i < point.length; i++) { assertFalse(visitor.coordInsideBox[i]); } InterpolationMeasure output = visitor.getResult(); double[] zero = new double[point.length]; assertArrayEquals(zero, output.measure.high); assertArrayEquals(zero, output.distances.high); assertArrayEquals(zero, output.probMass.high); assertArrayEquals(zero, output.measure.low); assertArrayEquals(zero, output.distances.low); assertArrayEquals(zero, output.probMass.low); } @Test public void testAcceptLeafEquals() { float[] point = { 1.0f, 2.0f, 3.0f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafDepth = 100; int leafMass = 10; when(leafNode.getMass()).thenReturn(leafMass); int sampleSize = 21; SimpleInterpolationVisitor visitor = new SimpleInterpolationVisitor(point, sampleSize, 1, false); visitor.acceptLeaf(leafNode, leafDepth); InterpolationMeasure result = visitor.getResult(); double[] expected = new double[point.length]; Arrays.fill(expected, 0.5 * (1 + leafMass) / point.length); assertArrayEquals(expected, result.measure.high); assertArrayEquals(expected, result.measure.low); Arrays.fill(expected, 0.5 / point.length); assertArrayEquals(expected, result.probMass.high); assertArrayEquals(expected, result.probMass.low); Arrays.fill(expected, 0.0); assertArrayEquals(expected, result.distances.high); assertArrayEquals(expected, result.distances.low); } @Test public void testAcceptLeafNotEquals() { float[] point = { 1.0f, 9.0f, 4.0f }; float[] anotherPoint = { 4.0f, 5.0f, 6.0f }; INodeView leafNode = mock(NodeView.class); when(leafNode.getLeafPoint()).thenReturn(anotherPoint); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(anotherPoint, anotherPoint)); when(leafNode.getMass()).thenReturn(4); int leafDepth = 100; int sampleSize = 99; SimpleInterpolationVisitor visitor = new SimpleInterpolationVisitor(point, sampleSize, 1, false); visitor.acceptLeaf(leafNode, leafDepth); InterpolationMeasure result = visitor.getResult(); double expectedSumOfNewRange = 3.0 + 4.0 + 2.0; double[] expectedDifferenceInRangeVector = { 0.0, 3.0, 4.0, 0.0, 0.0, 2.0 }; double[] expectedProbVector = Arrays.stream(expectedDifferenceInRangeVector).map(x -> x / expectedSumOfNewRange) .toArray(); double[] expectedmeasure = Arrays.stream(expectedProbVector).toArray(); double[] expectedDistances = new double[2 * point.length]; for (int i = 0; i < 2 * point.length; i++) { expectedDistances[i] = expectedProbVector[i] * expectedDifferenceInRangeVector[i]; } for (int i = 0; i < 2 * point.length; i++) { expectedmeasure[i] = expectedmeasure[i] * 5; } for (int i = 0; i < point.length; i++) { assertEquals(expectedProbVector[2 * i], result.probMass.high[i]); assertEquals(expectedProbVector[2 * i + 1], result.probMass.low[i]); assertEquals(expectedmeasure[2 * i], result.measure.high[i]); assertEquals(expectedmeasure[2 * i + 1], result.measure.low[i]); assertEquals(expectedDistances[2 * i], result.distances.high[i]); assertEquals(expectedDistances[2 * i + 1], result.distances.low[i]); } } @Test public void testAcceptEqualsLeafPoint() { float[] pointToScore = { 0.0f, 0.0f }; int sampleSize = 50; SimpleInterpolationVisitor visitor = new SimpleInterpolationVisitor(pointToScore, sampleSize, 1, false); float[] point = Arrays.copyOf(pointToScore, pointToScore.length); INodeView node = mock(NodeView.class); when(node.getLeafPoint()).thenReturn(point); when(node.getBoundingBox()).thenReturn(new BoundingBox(point, point)); when(node.getMass()).thenReturn(1); int depth = 2; visitor.acceptLeaf(node, depth); InterpolationMeasure result = visitor.getResult(); double[] expected = new double[point.length]; Arrays.fill(expected, 0.5 * (1 + node.getMass()) / point.length); assertArrayEquals(expected, result.measure.high); assertArrayEquals(expected, result.measure.low); Arrays.fill(expected, 0.5 / point.length); assertArrayEquals(expected, result.probMass.high); assertArrayEquals(expected, result.probMass.low); Arrays.fill(expected, 0.0); assertArrayEquals(expected, result.distances.high); assertArrayEquals(expected, result.distances.low); depth--; float[] siblingPoint = { 1.0f, -2.0f }; INodeView sibling = mock(NodeView.class); int siblingMass = 2; when(sibling.getMass()).thenReturn(siblingMass); INodeView parent = mock(NodeView.class); when(parent.getMass()).thenReturn(1 + siblingMass); BoundingBox boundingBox = new BoundingBox(point, siblingPoint); when(parent.getBoundingBox()).thenReturn(boundingBox); when(parent.getSiblingBoundingBox(any())).thenReturn(new BoundingBox(siblingPoint)); visitor.accept(parent, depth); result = visitor.getResult(); // compute using shadow box (sibling leaf node at {1.0, -2.0} and parent // bounding box double[] directionalDistance = { 0.0, 1.0, 2.0, 0.0 }; double[] differenceInRange = { 0.0, 1.0, 2.0, 0.0 }; double sumOfNewRange = 1.0 + 2.0; double[] probVector = Arrays.stream(differenceInRange).map(x -> x / sumOfNewRange).toArray(); expected = new double[2 * pointToScore.length]; for (int i = 0; i < expected.length; i++) { expected[i] = probVector[i] * (1 + node.getMass() + parent.getMass()); } for (int i = 0; i < pointToScore.length; i++) { assertEquals(expected[2 * i], result.measure.high[i]); assertEquals(expected[2 * i + 1], result.measure.low[i]); } for (int i = 0; i < expected.length; i++) { expected[i] = probVector[i]; } for (int i = 0; i < pointToScore.length; i++) { assertEquals(expected[2 * i], result.probMass.high[i]); assertEquals(expected[2 * i + 1], result.probMass.low[i]); } for (int i = 0; i < expected.length; i++) { expected[i] = probVector[i] * directionalDistance[i]; } for (int i = 0; i < pointToScore.length; i++) { assertEquals(expected[2 * i], result.distances.high[i]); assertEquals(expected[2 * i + 1], result.distances.low[i]); } // reset to probmass for (int i = 0; i < expected.length; i++) { expected[i] = probVector[i]; } // testing shawbox setup for grandparent INodeView uncle = mock(NodeView.class); int uncleMass = 2; when(sibling.getMass()).thenReturn(uncleMass); INodeView grandParent = mock(NodeView.class); when(grandParent.getMass()).thenReturn(1 + siblingMass + uncleMass); BoundingBox grandBox = boundingBox.getMergedBox(new float[] { 2.0f, 2.0f }); when(grandParent.getBoundingBox()).thenReturn(grandBox); when(grandParent.getSiblingBoundingBox(any())).thenReturn(new BoundingBox(new float[] { 2.0f, 2.0f })); visitor.accept(grandParent, depth - 1); result = visitor.getResult(); directionalDistance = new double[] { 0.0, 2.0, 0.0, 0.0 }; differenceInRange = new double[] { 0.0, 1.0, 0.0, 0.0 }; double newSumOfNewRange = 1.0 + 2.0 + 1.0 + 2.0; probVector = Arrays.stream(differenceInRange).map(x -> x / newSumOfNewRange).toArray(); double prob = Arrays.stream(probVector).sum(); for (int i = 0; i < expected.length; i++) { expected[i] = probVector[i] + (1 - prob) * expected[i]; } for (int i = 0; i < pointToScore.length; i++) { System.out.println(i); assertEquals(expected[2 * i], result.probMass.high[i]); assertEquals(expected[2 * i + 1], result.probMass.low[i]); } } @Test public void testAccept() { float[] pointToScore = { 0.0f, 0.0f }; int sampleSize = 50; SimpleInterpolationVisitor visitor = new SimpleInterpolationVisitor(pointToScore, sampleSize, 1, false); INodeView leafNode = mock(NodeView.class); float[] point = new float[] { 1.0f, -2.0f }; when(leafNode.getLeafPoint()).thenReturn(point); when(leafNode.getBoundingBox()).thenReturn(new BoundingBox(point, point)); int leafMass = 3; when(leafNode.getMass()).thenReturn(leafMass); int depth = 4; visitor.acceptLeaf(leafNode, depth); InterpolationMeasure result = visitor.getResult(); double expectedSumOfNewRange = 1.0 + 2.0; double[] expectedDifferenceInRangeVector = { 0.0, 1.0, 2.0, 0.0 }; double[] expectedProbVector = Arrays.stream(expectedDifferenceInRangeVector).map(x -> x / expectedSumOfNewRange) .toArray(); double[] expectedNumPts = Arrays.stream(expectedProbVector).toArray(); double[] expectedDistances = new double[2 * pointToScore.length]; for (int i = 0; i < 2 * pointToScore.length; i++) { expectedDistances[i] = expectedProbVector[i] * expectedDifferenceInRangeVector[i]; } for (int i = 0; i < 2 * pointToScore.length; i++) { expectedNumPts[i] = expectedNumPts[i] * 4; } for (int i = 0; i < pointToScore.length; i++) { assertEquals(expectedProbVector[2 * i], result.probMass.high[i]); assertEquals(expectedProbVector[2 * i + 1], result.probMass.low[i]); assertEquals(expectedNumPts[2 * i], result.measure.high[i]); assertEquals(expectedNumPts[2 * i + 1], result.measure.low[i]); assertEquals(expectedDistances[2 * i], result.distances.high[i]); assertEquals(expectedDistances[2 * i + 1], result.distances.low[i]); } // parent does not contain pointToScore depth--; INodeView sibling = mock(NodeView.class); int siblingMass = 2; when(sibling.getMass()).thenReturn(siblingMass); INodeView parent = mock(NodeView.class); int parentMass = leafMass + siblingMass; when(parent.getMass()).thenReturn(parentMass); when(parent.getBoundingBox()).thenReturn(new BoundingBox(point, new float[] { 2.0f, -0.5f })); visitor.accept(parent, depth); result = visitor.getResult(); double expectedSumOfNewRange2 = 2.0 + 2.0; double expectedProbOfCut2 = (1.0 + 0.5) / expectedSumOfNewRange2; double[] expectedDifferenceInRangeVector2 = { 0.0, 1.0, 0.5, 0.0 }; double[] expectedDirectionalDistanceVector2 = { 0.0, 2.0, 2.0, 0.0 }; for (int i = 0; i < 2 * pointToScore.length; i++) { double prob = expectedDifferenceInRangeVector2[i] / expectedSumOfNewRange2; expectedProbVector[i] = prob + (1 - expectedProbOfCut2) * expectedProbVector[i]; expectedNumPts[i] = prob * (1 + parent.getMass()) + (1 - expectedProbOfCut2) * expectedNumPts[i]; expectedDistances[i] = prob * expectedDirectionalDistanceVector2[i] + (1 - expectedProbOfCut2) * expectedDistances[i]; } for (int i = 0; i < pointToScore.length; i++) { assertEquals(expectedProbVector[2 * i], result.probMass.high[i]); assertEquals(expectedProbVector[2 * i + 1], result.probMass.low[i]); assertEquals(expectedNumPts[2 * i], result.measure.high[i]); assertEquals(expectedNumPts[2 * i + 1], result.measure.low[i]); assertEquals(expectedDistances[2 * i], result.distances.high[i]); assertEquals(expectedDistances[2 * i + 1], result.distances.low[i]); } // grandparent contains pointToScore assertFalse(visitor.pointInsideBox); depth--; } }
438
0
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/test/java/com/amazon/randomcutforest/store/PointStoreTest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.store; import static org.junit.jupiter.api.Assertions.assertArrayEquals; import static org.junit.jupiter.api.Assertions.assertEquals; import static org.junit.jupiter.api.Assertions.assertFalse; import static org.junit.jupiter.api.Assertions.assertNotEquals; import static org.junit.jupiter.api.Assertions.assertNotSame; import static org.junit.jupiter.api.Assertions.assertThrows; import static org.junit.jupiter.api.Assertions.assertTrue; import java.util.Arrays; import java.util.Random; import org.junit.jupiter.api.BeforeEach; import org.junit.jupiter.api.Test; public class PointStoreTest { private int dimensions; private int capacity; private PointStore pointStore; @BeforeEach public void setUp() { dimensions = 2; capacity = 4; pointStore = new PointStoreSmall(dimensions, capacity); } @Test public void testNew() { assertEquals(dimensions, pointStore.getDimensions()); assertEquals(capacity, pointStore.getCapacity()); assertEquals(0, pointStore.size()); for (int i = 0; i < pointStore.getIndexCapacity(); i++) { assertEquals(0, pointStore.getRefCount(i)); } } @Test public void testAdd() { float[] point1 = { 1.2f, -3.4f }; int offset1 = pointStore.add(point1, 1); assertTrue(offset1 >= 0 && offset1 < capacity); assertEquals(1, pointStore.getRefCount(offset1)); assertEquals(1, pointStore.size()); float[] retrievedPoint1 = pointStore.getNumericVector(offset1); assertNotSame(point1, retrievedPoint1); assertArrayEquals(point1, retrievedPoint1); float[] point2 = { 111.2f, -333.4f }; int offset2 = pointStore.add(point2, 2); assertTrue(offset2 >= 0 && offset2 < capacity); assertEquals(1, pointStore.getRefCount(offset2)); assertEquals(2, pointStore.size()); assertNotEquals(offset1, offset2); float[] retrievedPoint2 = pointStore.getNumericVector(offset2); assertNotSame(point2, retrievedPoint2); assertArrayEquals(point2, retrievedPoint2); // check that adding a second point didn't change the first stored point's value retrievedPoint1 = pointStore.getNumericVector(offset1); assertNotSame(point1, retrievedPoint1); assertArrayEquals(point1, retrievedPoint1); } @Test public void testAddInvalid() { assertThrows(IllegalArgumentException.class, () -> pointStore.add(new float[] { 1.1f, -2.2f, 3.0f }, 0)); for (int i = 0; i < capacity; i++) { float[] point = new float[dimensions]; point[0] = (float) Math.random(); point[1] = (float) Math.random(); pointStore.add(point, i + 2); } // point store is full assertThrows(IllegalStateException.class, () -> pointStore.add(new float[] { 1.1f, -2.2f }, 0)); } @Test public void testGetInvalid() { assertThrows(IllegalArgumentException.class, () -> pointStore.getNumericVector(-1)); assertThrows(IllegalArgumentException.class, () -> pointStore.getNumericVector(capacity)); } @Test public void testIncrementRefCount() { float[] point = { 1.2f, -3.4f }; int offset = pointStore.add(point, 0); assertEquals(1, pointStore.getRefCount(offset)); pointStore.incrementRefCount(offset); assertEquals(2, pointStore.getRefCount(offset)); } @Test public void testIncrementRefCountInvalid() { assertThrows(IllegalArgumentException.class, () -> pointStore.incrementRefCount(-1)); assertThrows(IllegalArgumentException.class, () -> pointStore.incrementRefCount(0)); } @Test public void testDecrementRefCount() { float[] point = { 1.2f, -3.4f }; int offset = pointStore.add(point, 0); pointStore.incrementRefCount(offset); assertEquals(2, pointStore.getRefCount(offset)); assertEquals(1, pointStore.size()); pointStore.decrementRefCount(offset); assertEquals(1, pointStore.getRefCount(offset)); assertEquals(1, pointStore.size()); pointStore.decrementRefCount(offset); assertEquals(0, pointStore.getRefCount(offset)); assertEquals(0, pointStore.size()); } @Test public void testDecrementRefCountInvalid() { assertThrows(IllegalArgumentException.class, () -> pointStore.decrementRefCount(-1)); assertThrows(IllegalArgumentException.class, () -> pointStore.decrementRefCount(0)); } @Test public void testPointEquals() { float[] point = { 1.2f, -3.4f }; int offset = pointStore.add(point, 0); assertArrayEquals(pointStore.getNumericVector(offset), point); assertNotEquals(pointStore.getNumericVector(offset), new float[] { 5.6f, -7.8f }); } @Test public void testPointEqualsInvalid() { float[] point = { 1.2f, -3.4f }; assertThrows(IllegalArgumentException.class, () -> pointStore.getNumericVector(-1)); assertThrows(IllegalArgumentException.class, () -> pointStore.getNumericVector(0)); } @Test public void internalshinglingTestNoRotation() { int shinglesize = 10; PointStore store = new PointStore.Builder().capacity(20 * shinglesize).dimensions(shinglesize) .shingleSize(shinglesize).indexCapacity(shinglesize).internalShinglingEnabled(true) .currentStoreCapacity(1).build(); assertFalse(store.isInternalRotationEnabled()); Random random = new Random(0); float[] shingle = new float[shinglesize]; for (int i = 0; i < 10 * shinglesize - 3; i++) { shingle[(i + 3) % shinglesize] = (float) random.nextDouble(); store.add(new float[] { shingle[(i + 3) % shinglesize] }, i); } assertArrayEquals(store.getNumericVector(9 * shinglesize - 3), shingle, (float) 1e-6); assertArrayEquals(store.getInternalShingle(), shingle, (float) 1e-6); assertArrayEquals(store.transformIndices(new int[] { 0 }), new int[] { shinglesize - 1 }); assertThrows(IllegalArgumentException.class, () -> store.transformIndices(new int[] { 1 })); assertThrows(IllegalArgumentException.class, () -> store.transformIndices(new int[] { 0, 0 })); assertArrayEquals(store.transformToShingledPoint(new float[] { 0.0f }), store.transformToShingledPoint(new float[] { -0.0f }), (float) 1e-6); assertThrows(IllegalArgumentException.class, () -> store.add(new float[] { 0, 0 }, 0)); } @Test public void internalshinglingTestWithRotation() { int shinglesize = 10; PointStore store = new PointStore.Builder().capacity(20 * shinglesize).dimensions(shinglesize) .shingleSize(shinglesize).indexCapacity(shinglesize).internalShinglingEnabled(true) .internalRotationEnabled(true).currentStoreCapacity(1).build(); assertTrue(store.isInternalRotationEnabled()); Random random = new Random(0); float[] shingle = new float[shinglesize]; float[] temp = null; for (int i = 0; i < 10 * shinglesize + 5; i++) { shingle[i % shinglesize] = (float) random.nextDouble(); temp = store.transformToShingledPoint(new float[] { shingle[i % shinglesize] }); store.add(new float[] { shingle[i % shinglesize] }, i); } assertEquals(store.getNextSequenceIndex(), 10 * shinglesize + 5); assertArrayEquals(temp, shingle, (float) 1e-6); assertArrayEquals(store.getNumericVector(9 * shinglesize + 5), shingle, (float) 1e-6); assertNotEquals(store.internalShingle, store.getInternalShingle()); assertArrayEquals(store.getNumericVector(9 * shinglesize + 5), shingle); assertNotEquals(store.getNumericVector(9 * shinglesize + 4), shingle); assertArrayEquals(store.getInternalShingle(), shingle, (float) 1e-6); assertArrayEquals(store.transformIndices(new int[] { 0 }), new int[] { 5 }); assertThrows(IllegalArgumentException.class, () -> store.transformIndices(new int[] { 1 })); assertEquals(store.transformToShingledPoint(new float[] { 1, 2 }).length, 2); assertArrayEquals(store.transformToShingledPoint(new float[] { 0.0f }), store.transformToShingledPoint(new float[] { -0.0f }), (float) 1e-6); } @Test public void checkRotationAndCompact() { int shinglesize = 4; PointStore store = new PointStore.Builder().capacity(2 * shinglesize).dimensions(shinglesize) .shingleSize(shinglesize).indexCapacity(shinglesize).internalShinglingEnabled(true) .internalRotationEnabled(true).currentStoreCapacity(1).build(); for (int i = 0; i < 2 * shinglesize; i++) { store.add(new float[] { -i - 1 }, i); } for (int i = 0; i < 2 * shinglesize - shinglesize + 1; i++) { if (i != shinglesize - 1) { store.decrementRefCount(i); } } assertThrows(IllegalArgumentException.class, () -> store.getNumericVector(0)); float[] test = new float[shinglesize]; for (int i = 0; i < shinglesize; i++) { test[i] = -(i + shinglesize + 1); } test[shinglesize - 1] = -shinglesize; assertArrayEquals(store.getNumericVector(shinglesize - 1), test, 1e-6f); store.compact(); for (int i = 2 * shinglesize; i < 4 * shinglesize - 1; i++) { store.add(new float[] { -i - 1 }, i); } assertThrows(IllegalStateException.class, () -> store.add(new float[] { -4 * shinglesize }, 0)); for (int i = 0; i < 2 * shinglesize; i++) { if (i != shinglesize - 1) { store.decrementRefCount(i); } } assertEquals(store.toString(shinglesize - 1), Arrays.toString(test)); for (int i = 4 * shinglesize; i < 6 * shinglesize - 1; i++) { store.add(new float[] { -i - 1 }, i); } assertThrows(IllegalStateException.class, () -> store.add(new float[] { -6 * shinglesize }, 6 * shinglesize - 1)); store.decrementRefCount(shinglesize - 1); store.add(new float[] { -6 * shinglesize }, 6 * shinglesize - 1); store.decrementRefCount(shinglesize); store.compact(); } @Test void CompactionTest() { int shinglesize = 2; PointStore store = new PointStore.Builder().capacity(6).dimensions(shinglesize).shingleSize(shinglesize) .indexCapacity(6).directLocationEnabled(false).internalShinglingEnabled(true).build(); store.add(new float[] { 0 }, 0L); for (int i = 0; i < 5; i++) { store.add(new float[] { i + 1 }, 0L); } int finalIndex = store.add(new float[] { 4 + 2 }, 0L); assertArrayEquals(store.getNumericVector(finalIndex), new float[] { 5, 6 }); store.decrementRefCount(1); store.decrementRefCount(2); int index = store.add(new float[] { 7 }, 0L); assertArrayEquals(store.getNumericVector(index), new float[] { 6, 7 }); store.decrementRefCount(index); assertTrue(store.size() < store.capacity); index = store.add(new float[] { 8 }, 0L); assertArrayEquals(store.getNumericVector(index), new float[] { 7, 8 }); } }
439
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/VisitorFactory.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import java.util.function.BiFunction; import com.amazon.randomcutforest.tree.ITree; /** * This is the interface for a visitor factory the factory corresponds to * mapping a (tree,point) pair to a visitor and a mapping for the inverse result */ public class VisitorFactory<R> implements IVisitorFactory<R> { private final BiFunction<ITree<?, ?>, float[], Visitor<R>> newVisitor; private final BiFunction<ITree<?, ?>, R, R> liftResult; public VisitorFactory(BiFunction<ITree<?, ?>, float[], Visitor<R>> newVisitor, BiFunction<ITree<?, ?>, R, R> liftResult) { this.newVisitor = newVisitor; this.liftResult = liftResult; } public VisitorFactory(BiFunction<ITree<?, ?>, float[], Visitor<R>> newVisitor) { this(newVisitor, (tree, x) -> x); } @Override public Visitor<R> newVisitor(ITree<?, ?> tree, float[] point) { return newVisitor.apply(tree, point); } @Override public R liftResult(ITree<?, ?> tree, R result) { return liftResult.apply(tree, result); } }
440
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/RandomCutForest.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.checkNotNull; import static com.amazon.randomcutforest.CommonUtils.toDoubleArray; import static com.amazon.randomcutforest.CommonUtils.toFloatArray; import static com.amazon.randomcutforest.summarization.Summarizer.DEFAULT_SEPARATION_RATIO_FOR_MERGE; import static java.lang.Math.max; import java.util.ArrayList; import java.util.Arrays; import java.util.Collections; import java.util.List; import java.util.Optional; import java.util.Random; import java.util.function.BiFunction; import java.util.function.BinaryOperator; import java.util.function.Function; import java.util.stream.Collector; import com.amazon.randomcutforest.anomalydetection.AnomalyAttributionVisitor; import com.amazon.randomcutforest.anomalydetection.AnomalyScoreVisitor; import com.amazon.randomcutforest.anomalydetection.DynamicAttributionVisitor; import com.amazon.randomcutforest.anomalydetection.DynamicScoreVisitor; import com.amazon.randomcutforest.anomalydetection.SimulatedTransductiveScalarScoreVisitor; import com.amazon.randomcutforest.config.Config; import com.amazon.randomcutforest.config.Precision; import com.amazon.randomcutforest.executor.AbstractForestTraversalExecutor; import com.amazon.randomcutforest.executor.AbstractForestUpdateExecutor; import com.amazon.randomcutforest.executor.IStateCoordinator; import com.amazon.randomcutforest.executor.ParallelForestTraversalExecutor; import com.amazon.randomcutforest.executor.ParallelForestUpdateExecutor; import com.amazon.randomcutforest.executor.PointStoreCoordinator; import com.amazon.randomcutforest.executor.SamplerPlusTree; import com.amazon.randomcutforest.executor.SequentialForestTraversalExecutor; import com.amazon.randomcutforest.executor.SequentialForestUpdateExecutor; import com.amazon.randomcutforest.imputation.ConditionalSampleSummarizer; import com.amazon.randomcutforest.imputation.ImputeVisitor; import com.amazon.randomcutforest.inspect.NearNeighborVisitor; import com.amazon.randomcutforest.interpolation.SimpleInterpolationVisitor; import com.amazon.randomcutforest.returntypes.ConditionalTreeSample; import com.amazon.randomcutforest.returntypes.ConvergingAccumulator; import com.amazon.randomcutforest.returntypes.DensityOutput; import com.amazon.randomcutforest.returntypes.DiVector; import com.amazon.randomcutforest.returntypes.InterpolationMeasure; import com.amazon.randomcutforest.returntypes.Neighbor; import com.amazon.randomcutforest.returntypes.OneSidedConvergingDiVectorAccumulator; import com.amazon.randomcutforest.returntypes.OneSidedConvergingDoubleAccumulator; import com.amazon.randomcutforest.returntypes.RangeVector; import com.amazon.randomcutforest.returntypes.SampleSummary; import com.amazon.randomcutforest.sampler.CompactSampler; import com.amazon.randomcutforest.sampler.IStreamSampler; import com.amazon.randomcutforest.store.IPointStore; import com.amazon.randomcutforest.store.PointStore; import com.amazon.randomcutforest.summarization.ICluster; import com.amazon.randomcutforest.summarization.Summarizer; import com.amazon.randomcutforest.tree.IBoundingBoxView; import com.amazon.randomcutforest.tree.ITree; import com.amazon.randomcutforest.tree.RandomCutTree; import com.amazon.randomcutforest.util.ShingleBuilder; /** * The RandomCutForest class is the interface to the algorithms in this package, * and includes methods for anomaly detection, anomaly detection with * attribution, density estimation, imputation, and forecasting. A Random Cut * Forest is a collection of Random Cut Trees and stream samplers. When an * update call is made to a Random Cut Forest, each sampler is independently * updated with the submitted (and if the point is accepted by the sampler, then * the corresponding Random Cut Tree is also updated. Similarly, when an * algorithm method is called, the Random Cut Forest proxies to the trees which * implement the actual scoring logic. The Random Cut Forest then combines * partial results into a final results. */ public class RandomCutForest { /** * Default sample size. This is the number of points retained by the stream * sampler. */ public static final int DEFAULT_SAMPLE_SIZE = 256; /** * Default fraction used to compute the amount of points required by stream * samplers before results are returned. */ public static final double DEFAULT_OUTPUT_AFTER_FRACTION = 0.25; /** * If the user doesn't specify an explicit time decay value, then we set it to * the inverse of this coefficient times sample size. */ public static final double DEFAULT_SAMPLE_SIZE_COEFFICIENT_IN_TIME_DECAY = 10.0; /** * Default number of trees to use in the forest. */ public static final int DEFAULT_NUMBER_OF_TREES = 50; /** * By default, trees will not store sequence indexes. */ public static final boolean DEFAULT_STORE_SEQUENCE_INDEXES_ENABLED = false; /** * By default, trees will accept every point until full. */ public static final double DEFAULT_INITIAL_ACCEPT_FRACTION = 1.0; /** * By default, the collection of points stored in the forest will increase from * a small size, as needed to maximum capacity */ public static final boolean DEFAULT_DYNAMIC_RESIZING_ENABLED = true; /** * By default, shingling will be external */ public static final boolean DEFAULT_INTERNAL_SHINGLING_ENABLED = false; /** * By default, shingles will be a sliding window and not a cyclic buffer */ public static final boolean DEFAULT_INTERNAL_ROTATION_ENABLED = false; /** * By default, point stores will favor speed of size for larger shingle sizes */ public static final boolean DEFAULT_DIRECT_LOCATION_MAP = false; /** * Default floating-point precision for internal data structures. */ public static final Precision DEFAULT_PRECISION = Precision.FLOAT_64; /** * fraction of bounding boxes maintained by each tree */ public static final double DEFAULT_BOUNDING_BOX_CACHE_FRACTION = 1.0; /** * By default, nodes will not store center of mass. */ public static final boolean DEFAULT_CENTER_OF_MASS_ENABLED = false; /** * By default RCF is unaware of shingle size */ public static final int DEFAULT_SHINGLE_SIZE = 1; /** * Parallel execution is enabled by default. */ public static final boolean DEFAULT_PARALLEL_EXECUTION_ENABLED = false; public static final boolean DEFAULT_APPROXIMATE_ANOMALY_SCORE_HIGH_IS_CRITICAL = true; public static final double DEFAULT_APPROXIMATE_DYNAMIC_SCORE_PRECISION = 0.1; public static final int DEFAULT_APPROXIMATE_DYNAMIC_SCORE_MIN_VALUES_ACCEPTED = 5; /** * Random number generator used by the forest. */ protected Random random; /** * The number of dimensions in the input data. */ protected final int dimensions; /** * The sample size used by stream samplers in this forest. */ protected final int sampleSize; /** * The shingle size (if known) */ protected final int shingleSize; /** * The input dimensions for known shingle size and internal shingling */ protected final int inputDimensions; /** * The number of points required by stream samplers before results are returned. */ protected final int outputAfter; /** * The number of trees in this forest. */ protected final int numberOfTrees; /** * The decay factor used by stream samplers in this forest. */ protected double timeDecay; /** * Store the time information */ protected final boolean storeSequenceIndexesEnabled; /** * enables internal shingling */ protected final boolean internalShinglingEnabled; /** * The following can be set between 0 and 1 (inclusive) to achieve tradeoff * between smaller space, lower throughput and larger space, larger throughput */ protected final double boundingBoxCacheFraction; /** * Enable center of mass at internal nodes */ protected final boolean centerOfMassEnabled; /** * Enable parallel execution. */ protected final boolean parallelExecutionEnabled; /** * Number of threads to use in the thread pool if parallel execution is enabled. */ protected final int threadPoolSize; /** * A string to define an "execution mode" that can be used to set multiple * configuration options. This field is not currently in use. */ protected String executionMode; protected IStateCoordinator<?, float[]> stateCoordinator; protected ComponentList<?, float[]> components; /** * This flag is initialized to false. It is set to true when all component * models are ready. */ private boolean outputReady; /** * used for initializing the compact forests */ private final int initialPointStoreSize; private final int pointStoreCapacity; /** * An implementation of forest traversal algorithms. */ protected AbstractForestTraversalExecutor traversalExecutor; /** * An implementation of forest update algorithms. */ protected AbstractForestUpdateExecutor<?, float[]> updateExecutor; public <P> RandomCutForest(Builder<?> builder, IStateCoordinator<P, float[]> stateCoordinator, ComponentList<P, float[]> components, Random random) { this(builder, false); checkNotNull(stateCoordinator, "updateCoordinator must not be null"); checkNotNull(components, "componentModels must not be null"); checkNotNull(random, "random must not be null"); this.stateCoordinator = stateCoordinator; this.components = components; this.random = random; initExecutors(stateCoordinator, components); } public RandomCutForest(Builder<?> builder) { this(builder, false); random = builder.getRandom(); PointStore tempStore = PointStore.builder().internalRotationEnabled(builder.internalRotationEnabled) .capacity(pointStoreCapacity).initialSize(initialPointStoreSize) .internalShinglingEnabled(internalShinglingEnabled).shingleSize(shingleSize).dimensions(dimensions) .build(); IStateCoordinator<Integer, float[]> stateCoordinator = new PointStoreCoordinator<>(tempStore); ComponentList<Integer, float[]> components = new ComponentList<>(numberOfTrees); for (int i = 0; i < numberOfTrees; i++) { ITree<Integer, float[]> tree = new RandomCutTree.Builder().capacity(sampleSize) .randomSeed(random.nextLong()).pointStoreView(tempStore) .boundingBoxCacheFraction(boundingBoxCacheFraction).centerOfMassEnabled(centerOfMassEnabled) .storeSequenceIndexesEnabled(storeSequenceIndexesEnabled).outputAfter(1).build(); IStreamSampler<Integer> sampler = CompactSampler.builder().capacity(sampleSize).timeDecay(timeDecay) .randomSeed(random.nextLong()).storeSequenceIndexesEnabled(storeSequenceIndexesEnabled) .initialAcceptFraction(builder.initialAcceptFraction).build(); components.add(new SamplerPlusTree<>(sampler, tree)); } this.stateCoordinator = stateCoordinator; this.components = components; initExecutors(stateCoordinator, components); } protected <PointReference> void initExecutors(IStateCoordinator<PointReference, float[]> updateCoordinator, ComponentList<PointReference, float[]> components) { if (parallelExecutionEnabled) { traversalExecutor = new ParallelForestTraversalExecutor(components, threadPoolSize); updateExecutor = new ParallelForestUpdateExecutor<>(updateCoordinator, components, threadPoolSize); } else { traversalExecutor = new SequentialForestTraversalExecutor(components); updateExecutor = new SequentialForestUpdateExecutor<>(updateCoordinator, components); } } /** * This constructor is responsible for initializing a forest's configuration * variables from a builder. The method signature contains a boolean argument * that isn't used. This argument exists only to create a distinct method * signature so that we can expose {@link #RandomCutForest(Builder)} as a * protected constructor. * * @param builder A Builder instance giving the desired random cut forest * configuration. * @param notUsed This parameter is not used. */ protected RandomCutForest(Builder<?> builder, boolean notUsed) { checkArgument(builder.numberOfTrees > 0, "numberOfTrees must be greater than 0"); checkArgument(builder.sampleSize > 0, "sampleSize must be greater than 0"); builder.outputAfter.ifPresent(n -> { checkArgument(n > 0, "outputAfter must be greater than 0"); }); checkArgument(builder.dimensions > 0, "dimensions must be greater than 0"); builder.timeDecay.ifPresent(timeDecay -> { checkArgument(timeDecay >= 0, "timeDecay must be greater than or equal to 0"); }); builder.threadPoolSize.ifPresent(n -> checkArgument((n > 0) || ((n == 0) && !builder.parallelExecutionEnabled), "threadPoolSize must be greater/equal than 0. To disable thread pool, set parallel execution to 'false'.")); checkArgument(builder.internalShinglingEnabled || builder.shingleSize == 1 || builder.dimensions % builder.shingleSize == 0, "wrong shingle size"); // checkArgument(!builder.internalShinglingEnabled || builder.shingleSize > 1, // " need shingle size > 1 for internal shingling"); if (builder.internalRotationEnabled) { checkArgument(builder.internalShinglingEnabled, " enable internal shingling"); } builder.initialPointStoreSize.ifPresent(n -> { checkArgument(n > 0, "initial point store must be greater than 0"); checkArgument(n > builder.sampleSize * builder.numberOfTrees || builder.dynamicResizingEnabled, " enable dynamic resizing "); }); checkArgument(builder.boundingBoxCacheFraction >= 0 && builder.boundingBoxCacheFraction <= 1, "incorrect cache fraction range"); numberOfTrees = builder.numberOfTrees; sampleSize = builder.sampleSize; outputAfter = builder.outputAfter.orElse(max(1, (int) (sampleSize * DEFAULT_OUTPUT_AFTER_FRACTION))); internalShinglingEnabled = builder.internalShinglingEnabled; shingleSize = builder.shingleSize; dimensions = builder.dimensions; timeDecay = builder.timeDecay.orElse(1.0 / (DEFAULT_SAMPLE_SIZE_COEFFICIENT_IN_TIME_DECAY * sampleSize)); storeSequenceIndexesEnabled = builder.storeSequenceIndexesEnabled; centerOfMassEnabled = builder.centerOfMassEnabled; parallelExecutionEnabled = builder.parallelExecutionEnabled; boundingBoxCacheFraction = builder.boundingBoxCacheFraction; builder.directLocationMapEnabled = builder.directLocationMapEnabled || shingleSize == 1; inputDimensions = (internalShinglingEnabled) ? dimensions / shingleSize : dimensions; pointStoreCapacity = max(sampleSize * numberOfTrees + 1, 2 * sampleSize); initialPointStoreSize = builder.initialPointStoreSize.orElse(2 * sampleSize); if (parallelExecutionEnabled) { threadPoolSize = builder.threadPoolSize.orElse(Runtime.getRuntime().availableProcessors() - 1); } else { threadPoolSize = 0; } } /** * @return a new RandomCutForest builder. */ public static Builder builder() { return new Builder(); } /** * Create a new RandomCutForest with optional arguments set to default values. * * @param dimensions The number of dimension in the input data. * @param randomSeed The random seed to use to create the forest random number * generator * @return a new RandomCutForest with optional arguments set to default values. */ public static RandomCutForest defaultForest(int dimensions, long randomSeed) { return builder().dimensions(dimensions).randomSeed(randomSeed).build(); } /** * Create a new RandomCutForest with optional arguments set to default values. * * @param dimensions The number of dimension in the input data. * @return a new RandomCutForest with optional arguments set to default values. */ public static RandomCutForest defaultForest(int dimensions) { return builder().dimensions(dimensions).build(); } /** * @return the number of trees in the forest. */ public int getNumberOfTrees() { return numberOfTrees; } /** * @return the sample size used by stream samplers in this forest. */ public int getSampleSize() { return sampleSize; } /** * @return the shingle size used by the point store. */ public int getShingleSize() { return shingleSize; } /** * @return the number of points required by stream samplers before results are * returned. */ public int getOutputAfter() { return outputAfter; } /** * @return the number of dimensions in the data points accepted by this forest. */ public int getDimensions() { return dimensions; } /** * @return return the decay factor used by stream samplers in this forest. */ public double getTimeDecay() { return timeDecay; } /** * @return true if points are saved with sequence indexes, false otherwise. */ public boolean isStoreSequenceIndexesEnabled() { return storeSequenceIndexesEnabled; } /** * For compact forests, users can choose to specify the desired floating-point * precision to use internally to store points. Choosing single-precision will * reduce the memory size of the model at the cost of requiring double/float * conversions. * * @return the desired precision to use internally to store points. */ public Precision getPrecision() { return Precision.FLOAT_32; } @Deprecated public boolean isCompact() { return true; } /** * @return true if internal shingling is performed, false otherwise. */ public boolean isInternalShinglingEnabled() { return internalShinglingEnabled; } /** * @return true if tree nodes retain the center of mass, false otherwise. */ public boolean isCenterOfMassEnabled() { return centerOfMassEnabled; } /** * @return true if parallel execution is enabled, false otherwise. */ public boolean isParallelExecutionEnabled() { return parallelExecutionEnabled; } public double getBoundingBoxCacheFraction() { return boundingBoxCacheFraction; } /** * @return the number of threads in the thread pool if parallel execution is * enabled, 0 otherwise. */ public int getThreadPoolSize() { return threadPoolSize; } public IStateCoordinator<?, ?> getUpdateCoordinator() { return stateCoordinator; } public ComponentList<?, ?> getComponents() { return components; } /** * used for scoring and other function, expands to a shingled point in either * case performs a clean copy * * @param point input point * @return a shingled copy or a clean copy */ public float[] transformToShingledPoint(float[] point) { return stateCoordinator.getStore().transformToShingledPoint(point); } /** * does the pointstore use rotated shingles * * @return true/false based on pointstore */ public boolean isRotationEnabled() { return stateCoordinator.getStore().isInternalRotationEnabled(); } /** * transforms the missing indices on the input point to the corresponding * indices of a shingled point * * @param indexList input array of missing values * @param length length of the input array * @return output array of missing values corresponding to shingle */ protected int[] transformIndices(int[] indexList, int length) { return (internalShinglingEnabled && length == inputDimensions) ? stateCoordinator.getStore().transformIndices(indexList) : indexList; } /** * * @return the last known shingled point seen */ public float[] lastShingledPoint() { checkArgument(internalShinglingEnabled, "incorrect use"); return stateCoordinator.getStore().getInternalShingle(); } /** * * @return the sequence index of the last known shingled point. If internal * shingling is not enabled, then this would correspond to the number of * updates */ public long nextSequenceIndex() { return stateCoordinator.getStore().getNextSequenceIndex(); } /** * Update the forest with the given point. The point is submitted to each * sampler in the forest. If the sampler accepts the point, the point is * submitted to the update method in the corresponding Random Cut Tree. * * @param point The point used to update the forest. */ public void update(double[] point) { update(toFloatArray(point)); } public void update(float[] point) { checkNotNull(point, "point must not be null"); checkArgument(internalShinglingEnabled || point.length == dimensions, String.format("point.length must equal %d", dimensions)); checkArgument(!internalShinglingEnabled || point.length == inputDimensions, String.format("point.length must equal %d for internal shingling", inputDimensions)); updateExecutor.update(point); } /** * Update the forest with the given point and a timestamp. The point is * submitted to each sampler in the forest as if that timestamp was the correct * stamp. storeSequenceIndexes must be false since the algorithm will not verify * the correctness of the timestamp. * * @param point The point used to update the forest. * @param sequenceNum The timestamp of the corresponding point */ public void update(double[] point, long sequenceNum) { checkNotNull(point, "point must not be null"); update(toFloatArray(point), sequenceNum); } public void update(float[] point, long sequenceNum) { checkNotNull(point, "point must not be null"); checkArgument(!internalShinglingEnabled, "cannot be applied with internal shingling"); checkArgument(point.length == dimensions, () -> "point.length must equal to " + dimensions); updateExecutor.update(point, sequenceNum); } /** * Update the forest such that each tree caches a fraction of the bounding * boxes. This allows for a tradeoff between speed and storage. * * @param cacheFraction The (approximate) fraction of bounding boxes used in * caching. */ public void setBoundingBoxCacheFraction(double cacheFraction) { checkArgument(0 <= cacheFraction && cacheFraction <= 1, "cacheFraction must be between 0 and 1 (inclusive)"); updateExecutor.getComponents().forEach(c -> c.setConfig(Config.BOUNDING_BOX_CACHE_FRACTION, cacheFraction)); } /** * changes the setting of time dependent sampling on the fly * * @param timeDecay new value of sampling rate */ public void setTimeDecay(double timeDecay) { checkArgument(0 <= timeDecay, "timeDecay must be greater than or equal to 0"); this.timeDecay = timeDecay; updateExecutor.getComponents().forEach(c -> c.setConfig(Config.TIME_DECAY, timeDecay)); } /** * Visit each of the trees in the forest and combine the individual results into * an aggregate result. A visitor is constructed for each tree using the visitor * factory, and then submitted to * {@link RandomCutTree#traverse(float[], IVisitorFactory)}. The results from * all the trees are combined using the accumulator and then transformed using * the finisher before being returned. Trees are visited in parallel using * {@link java.util.Collection#parallelStream()}. * * @param point The point that defines the traversal path. * @param visitorFactory A factory method which is invoked for each tree to * construct a visitor. * @param accumulator A function that combines the results from individual * trees into an aggregate result. * @param finisher A function called on the aggregate result in order to * produce the final result. * @param <R> The visitor result type. This is the type that will be * returned after traversing each individual tree. * @param <S> The final type, after any final normalization at the * forest level. * @return The aggregated and finalized result after sending a visitor through * each tree in the forest. */ public <R, S> S traverseForest(float[] point, IVisitorFactory<R> visitorFactory, BinaryOperator<R> accumulator, Function<R, S> finisher) { checkNotNull(point, "point must not be null"); checkArgument(point.length == dimensions, () -> "point.length must equal to " + dimensions); checkNotNull(visitorFactory, "visitorFactory must not be null"); checkNotNull(accumulator, "accumulator must not be null"); checkNotNull(finisher, "finisher must not be null"); return traversalExecutor.traverseForest(point, visitorFactory, accumulator, finisher); } /** * Visit each of the trees in the forest and combine the individual results into * an aggregate result. A visitor is constructed for each tree using the visitor * factory, and then submitted to * {@link RandomCutTree#traverse(float[], IVisitorFactory)}. The results from * individual trees are collected using the {@link java.util.stream.Collector} * and returned. Trees are visited in parallel using * {@link java.util.Collection#parallelStream()}. * * @param point The point that defines the traversal path. * @param visitorFactory A factory method which is invoked for each tree to * construct a visitor. * @param collector A collector used to aggregate individual tree results * into a final result. * @param <R> The visitor result type. This is the type that will be * returned after traversing each individual tree. * @param <S> The final type, after any final normalization at the * forest level. * @return The aggregated and finalized result after sending a visitor through * each tree in the forest. */ public <R, S> S traverseForest(float[] point, IVisitorFactory<R> visitorFactory, Collector<R, ?, S> collector) { checkNotNull(point, "point must not be null"); checkArgument(point.length == dimensions, () -> "point.length must equal to " + dimensions); checkNotNull(visitorFactory, "visitorFactory must not be null"); checkNotNull(collector, "collector must not be null"); return traversalExecutor.traverseForest(point, visitorFactory, collector); } /** * Visit each of the trees in the forest sequentially and combine the individual * results into an aggregate result. A visitor is constructed for each tree * using the visitor factory, and then submitted to * {@link RandomCutTree#traverse(float[], IVisitorFactory)}. The results from * all the trees are combined using the {@link ConvergingAccumulator}, and the * method stops visiting trees after convergence is reached. The result is * transformed using the finisher before being returned. * * @param point The point that defines the traversal path. * @param visitorFactory A factory method which is invoked for each tree to * construct a visitor. * @param accumulator An accumulator that combines the results from * individual trees into an aggregate result and checks to * see if the result can be returned without further * processing. * @param finisher A function called on the aggregate result in order to * produce the final result. * @param <R> The visitor result type. This is the type that will be * returned after traversing each individual tree. * @param <S> The final type, after any final normalization at the * forest level. * @return The aggregated and finalized result after sending a visitor through * each tree in the forest. */ public <R, S> S traverseForest(float[] point, IVisitorFactory<R> visitorFactory, ConvergingAccumulator<R> accumulator, Function<R, S> finisher) { checkNotNull(point, "point must not be null"); checkArgument(point.length == dimensions, () -> "point.length must equal to " + dimensions); checkNotNull(visitorFactory, "visitorFactory must not be null"); checkNotNull(accumulator, "accumulator must not be null"); checkNotNull(finisher, "finisher must not be null"); return traversalExecutor.traverseForest(point, visitorFactory, accumulator, finisher); } /** * Visit each of the trees in the forest and combine the individual results into * an aggregate result. A multi-visitor is constructed for each tree using the * visitor factory, and then submitted to * {@link RandomCutTree#traverseMulti(float[], IMultiVisitorFactory)}. The * results from all the trees are combined using the accumulator and then * transformed using the finisher before being returned. * * @param point The point that defines the traversal path. * @param visitorFactory A factory method which is invoked for each tree to * construct a multi-visitor. * @param accumulator A function that combines the results from individual * trees into an aggregate result. * @param finisher A function called on the aggregate result in order to * produce the final result. * @param <R> The visitor result type. This is the type that will be * returned after traversing each individual tree. * @param <S> The final type, after any final normalization at the * forest level. * @return The aggregated and finalized result after sending a visitor through * each tree in the forest. */ public <R, S> S traverseForestMulti(float[] point, IMultiVisitorFactory<R> visitorFactory, BinaryOperator<R> accumulator, Function<R, S> finisher) { checkNotNull(point, "point must not be null"); checkArgument(point.length == dimensions, () -> "point.length must equal to " + dimensions); checkNotNull(visitorFactory, "visitorFactory must not be null"); checkNotNull(accumulator, "accumulator must not be null"); checkNotNull(finisher, "finisher must not be null"); return traversalExecutor.traverseForestMulti(point, visitorFactory, accumulator, finisher); } /** * Visit each of the trees in the forest and combine the individual results into * an aggregate result. A multi-visitor is constructed for each tree using the * visitor factory, and then submitted to * {@link RandomCutTree#traverseMulti(float[], IMultiVisitorFactory)}. The * results from individual trees are collected using the * {@link java.util.stream.Collector} and returned. Trees are visited in * parallel using {@link java.util.Collection#parallelStream()}. * * @param point The point that defines the traversal path. * @param visitorFactory A factory method which is invoked for each tree to * construct a visitor. * @param collector A collector used to aggregate individual tree results * into a final result. * @param <R> The visitor result type. This is the type that will be * returned after traversing each individual tree. * @param <S> The final type, after any final normalization at the * forest level. * @return The aggregated and finalized result after sending a visitor through * each tree in the forest. */ public <R, S> S traverseForestMulti(float[] point, IMultiVisitorFactory<R> visitorFactory, Collector<R, ?, S> collector) { checkNotNull(point, "point must not be null"); checkArgument(point.length == dimensions, () -> "point.length must equal to " + dimensions); checkNotNull(visitorFactory, "visitorFactory must not be null"); checkNotNull(collector, "collector must not be null"); return traversalExecutor.traverseForestMulti(point, visitorFactory, collector); } /** * Compute an anomaly score for the given point. The point being scored is * compared with the points in the sample to compute a measure of how anomalous * it is. Scores are greater than 0, with higher scores corresponding to bing * more anomalous. A threshold of 1.0 is commonly used to distinguish anomalous * points from non-anomalous ones. * <p> * See {@link AnomalyScoreVisitor} for more details about the anomaly score * algorithm. * * @param point The point being scored. * @return an anomaly score for the given point. */ @Deprecated public double getAnomalyScore(double[] point) { return getAnomalyScore(toFloatArray(point)); } public double getAnomalyScore(float[] point) { if (!isOutputReady()) { return 0.0; } IVisitorFactory<Double> visitorFactory = (tree, x) -> new AnomalyScoreVisitor(tree.projectToTree(x), tree.getMass()); BinaryOperator<Double> accumulator = Double::sum; Function<Double, Double> finisher = x -> x / numberOfTrees; return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Anomaly score evaluated sequentially with option of early stopping the early * stopping parameter precision gives an approximate solution in the range * (1-precision)*score(q)- precision, (1+precision)*score(q) + precision for the * score of a point q. In this function z is hardcoded to 0.1. If this function * is used, then not all the trees will be used in evaluation (but they have to * be updated anyways, because they may be used for the next q). The advantage * is that "almost certainly" anomalies/non-anomalies can be detected easily * with few trees. * * @param point input point q * @return anomaly score with early stopping with z=0.1 */ @Deprecated public double getApproximateAnomalyScore(double[] point) { return getApproximateAnomalyScore(toFloatArray(point)); } public double getApproximateAnomalyScore(float[] point) { if (!isOutputReady()) { return 0.0; } IVisitorFactory<Double> visitorFactory = (tree, x) -> new AnomalyScoreVisitor(tree.projectToTree(x), tree.getMass()); ConvergingAccumulator<Double> accumulator = new OneSidedConvergingDoubleAccumulator( DEFAULT_APPROXIMATE_ANOMALY_SCORE_HIGH_IS_CRITICAL, DEFAULT_APPROXIMATE_DYNAMIC_SCORE_PRECISION, DEFAULT_APPROXIMATE_DYNAMIC_SCORE_MIN_VALUES_ACCEPTED, numberOfTrees); Function<Double, Double> finisher = x -> x / accumulator.getValuesAccepted(); return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Compute an anomaly score attribution DiVector for the given point. The point * being scored is compared with the points in the sample to compute a measure * of how anomalous it is. The result DiVector will contain an anomaly score in * both the positive and negative directions for each dimension of the data. * <p> * See {@link AnomalyAttributionVisitor} for more details about the anomaly * score algorithm. * * @param point The point being scored. * @return an anomaly score for the given point. */ public DiVector getAnomalyAttribution(double[] point) { return getAnomalyAttribution(toFloatArray(point)); } public DiVector getAnomalyAttribution(float[] point) { // this will return the same (modulo floating point summation) L1Norm as // getAnomalyScore if (!isOutputReady()) { return new DiVector(dimensions); } IVisitorFactory<DiVector> visitorFactory = new VisitorFactory<>( (tree, y) -> new AnomalyAttributionVisitor(tree.projectToTree(y), tree.getMass()), (tree, x) -> x.lift(tree::liftFromTree)); BinaryOperator<DiVector> accumulator = DiVector::addToLeft; Function<DiVector, DiVector> finisher = x -> x.scale(1.0 / numberOfTrees); return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Sequential version of attribution corresponding to getAnomalyScoreSequential; * The high-low sum in the result should be the same as the scalar score * computed by {@link #getAnomalyScore(double[])}. * * @param point The point being scored. * @return anomaly attribution for the given point. */ public DiVector getApproximateAnomalyAttribution(double[] point) { return getApproximateAnomalyAttribution(toFloatArray(point)); } public DiVector getApproximateAnomalyAttribution(float[] point) { if (!isOutputReady()) { return new DiVector(dimensions); } IVisitorFactory<DiVector> visitorFactory = new VisitorFactory<>( (tree, y) -> new AnomalyAttributionVisitor(tree.projectToTree(y), tree.getMass()), (tree, x) -> x.lift(tree::liftFromTree)); ConvergingAccumulator<DiVector> accumulator = new OneSidedConvergingDiVectorAccumulator(dimensions, DEFAULT_APPROXIMATE_ANOMALY_SCORE_HIGH_IS_CRITICAL, DEFAULT_APPROXIMATE_DYNAMIC_SCORE_PRECISION, DEFAULT_APPROXIMATE_DYNAMIC_SCORE_MIN_VALUES_ACCEPTED, numberOfTrees); Function<DiVector, DiVector> finisher = x -> x.scale(1.0 / accumulator.getValuesAccepted()); return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Compute a density estimate at the given point. * <p> * See {@link SimpleInterpolationVisitor} and {@link DensityOutput} for more * details about the density computation. * * @param point The point where the density estimate is made. * @return A density estimate. */ @Deprecated public DensityOutput getSimpleDensity(double[] point) { return getSimpleDensity(toFloatArray(point)); } public DensityOutput getSimpleDensity(float[] point) { // density estimation should use sufficiently larger number of samples // and only return answers when full if (!samplersFull()) { return new DensityOutput(dimensions, sampleSize); } IVisitorFactory<InterpolationMeasure> visitorFactory = new VisitorFactory<>((tree, y) -> new SimpleInterpolationVisitor(tree.projectToTree(y), sampleSize, 1.0, centerOfMassEnabled), (tree, x) -> x.lift(tree::liftFromTree)); Collector<InterpolationMeasure, ?, InterpolationMeasure> collector = InterpolationMeasure.collector(dimensions, sampleSize, numberOfTrees); return new DensityOutput(traverseForest(transformToShingledPoint(point), visitorFactory, collector)); } /** * Given a point with missing values, return a new point with the missing values * imputed. Each tree in the forest individual produces an imputed value. For * 1-dimensional points, the median imputed value is returned. For points with * more than 1 dimension, the imputed point with the 25th percentile anomaly * score is returned. * * The first function exposes the distribution. * * @param point A point with missing values. * @param numberOfMissingValues The number of missing values in the point. * @param missingIndexes An array containing the indexes of the missing * values in the point. The length of the array * should be greater than or equal to the number of * missing values. * @param centrality a parameter that provides a central estimation * versus a more random estimation * @return A point with the missing values imputed. */ public List<ConditionalTreeSample> getConditionalField(float[] point, int numberOfMissingValues, int[] missingIndexes, double centrality) { checkArgument(numberOfMissingValues > 0, "numberOfMissingValues must be greater than 0"); checkNotNull(missingIndexes, "missingIndexes must not be null"); checkArgument(numberOfMissingValues <= missingIndexes.length, "numberOfMissingValues must be less than or equal to missingIndexes.length"); checkArgument(centrality >= 0 && centrality <= 1, "centrality needs to be in range [0,1]"); if (!isOutputReady()) { return new ArrayList<>(); } int[] liftedIndices = transformIndices(missingIndexes, point.length); IMultiVisitorFactory<ConditionalTreeSample> visitorFactory = (tree, y) -> new ImputeVisitor(y, tree.projectToTree(y), liftedIndices, tree.projectMissingIndices(liftedIndices), centrality, tree.getRandomSeed()); return traverseForestMulti(transformToShingledPoint(point), visitorFactory, ConditionalTreeSample.collector); } public SampleSummary getConditionalFieldSummary(float[] point, int numberOfMissingValues, int[] missingIndexes, double centrality) { checkArgument(numberOfMissingValues >= 0, "cannot be negative"); checkNotNull(missingIndexes, "missingIndexes must not be null"); checkArgument(numberOfMissingValues <= missingIndexes.length, "numberOfMissingValues must be less than or equal to missingIndexes.length"); checkArgument(centrality >= 0 && centrality <= 1, "centrality needs to be in range [0,1]"); checkArgument(point != null, " cannot be null"); if (!isOutputReady()) { return new SampleSummary(dimensions); } int[] liftedIndices = transformIndices(missingIndexes, point.length); ConditionalSampleSummarizer summarizer = new ConditionalSampleSummarizer(liftedIndices, transformToShingledPoint(point), centrality); return summarizer.summarize(getConditionalField(point, numberOfMissingValues, missingIndexes, centrality)); } public float[] imputeMissingValues(float[] point, int numberOfMissingValues, int[] missingIndexes) { return getConditionalFieldSummary(point, numberOfMissingValues, missingIndexes, 1.0).median; } /** * Given a point with missing values, return a new point with the missing values * imputed. Each tree in the forest individual produces an imputed value. For * 1-dimensional points, the median imputed value is returned. For points with * more than 1 dimension, the imputed point with the 25th percentile anomaly * score is returned. * * @param point A point with missing values. * @param numberOfMissingValues The number of missing values in the point. * @param missingIndexes An array containing the indexes of the missing * values in the point. The length of the array * should be greater than or equal to the number of * missing values. * @return A point with the missing values imputed. */ @Deprecated public double[] imputeMissingValues(double[] point, int numberOfMissingValues, int[] missingIndexes) { return toDoubleArray(imputeMissingValues(toFloatArray(point), numberOfMissingValues, missingIndexes)); } /** * Given an initial shingled point, extrapolate the stream into the future to * produce a forecast. This method is intended to be called when the input data * is being shingled, and it works by imputing forward one shingle block at a * time. * * @param point The starting point for extrapolation. * @param horizon The number of blocks to forecast. * @param blockSize The number of entries in a block. This should be the same * as the size of a single input to the shingle. * @param cyclic If true then the shingling is cyclic, otherwise it's a * sliding shingle. * @param shingleIndex If cyclic is true, then this should be the current index * in the shingle. That is, the index where the next point * added to the shingle would be written. If cyclic is false * then this value is not used. * @return a forecasted time series. */ @Deprecated public double[] extrapolateBasic(double[] point, int horizon, int blockSize, boolean cyclic, int shingleIndex) { return toDoubleArray(extrapolateBasic(toFloatArray(point), horizon, blockSize, cyclic, shingleIndex)); } public float[] extrapolateBasic(float[] point, int horizon, int blockSize, boolean cyclic, int shingleIndex) { return extrapolateWithRanges(point, horizon, blockSize, cyclic, shingleIndex, 1.0).values; } // the following is provided for maximum flexibilty from the calling entity; // but likely use is extrapolateFromShingle(), which abstracts away rotation // etc. public RangeVector extrapolateWithRanges(float[] point, int horizon, int blockSize, boolean cyclic, int shingleIndex, double centrality) { checkArgument(0 < blockSize && blockSize < dimensions, "blockSize must be between 0 and dimensions (exclusive)"); checkArgument(dimensions % blockSize == 0, "dimensions must be evenly divisible by blockSize"); checkArgument(0 <= shingleIndex && shingleIndex < dimensions / blockSize, "shingleIndex must be between 0 (inclusive) and dimensions / blockSize"); RangeVector result = new RangeVector(blockSize * horizon); int[] missingIndexes = new int[blockSize]; float[] queryPoint = Arrays.copyOf(point, dimensions); if (cyclic) { extrapolateBasicCyclic(result, horizon, blockSize, shingleIndex, queryPoint, missingIndexes, centrality); } else { extrapolateBasicSliding(result, horizon, blockSize, queryPoint, missingIndexes, centrality); } return result; } // external management of shingle; can function for both internal and external // shingling // however blocksize has to be externally managed public RangeVector extrapolateFromShingle(float[] shingle, int horizon, int blockSize, double centrality) { return extrapolateWithRanges(shingle, horizon, blockSize, isRotationEnabled(), ((int) nextSequenceIndex()) % shingleSize, centrality); } /** * Given an initial shingled point, extrapolate the stream into the future to * produce a forecast. This method is intended to be called when the input data * is being shingled, and it works by imputing forward one shingle block at a * time. If the shingle is cyclic, then this method uses 0 as the shingle index. * * @param point The starting point for extrapolation. * @param horizon The number of blocks to forecast. * @param blockSize The number of entries in a block. This should be the same as * the size of a single input to the shingle. * @param cyclic If true then the shingling is cyclic, otherwise it's a * sliding shingle. * @return a forecasted time series. */ @Deprecated public double[] extrapolateBasic(double[] point, int horizon, int blockSize, boolean cyclic) { return extrapolateBasic(point, horizon, blockSize, cyclic, 0); } public float[] extrapolateBasic(float[] point, int horizon, int blockSize, boolean cyclic) { return extrapolateBasic(point, horizon, blockSize, cyclic, 0); } /** * Given a shingle builder, extrapolate the stream into the future to produce a * forecast. This method assumes you are passing in the shingle builder used to * preprocess points before adding them to this forest. * * @param builder The shingle builder used to process points before adding them * to the forest. * @param horizon The number of blocks to forecast. * @return a forecasted time series. */ @Deprecated public double[] extrapolateBasic(ShingleBuilder builder, int horizon) { return extrapolateBasic(builder.getShingle(), horizon, builder.getInputPointSize(), builder.isCyclic(), builder.getShingleIndex()); } void extrapolateBasicSliding(RangeVector result, int horizon, int blockSize, float[] queryPoint, int[] missingIndexes, double centrality) { int resultIndex = 0; Arrays.fill(missingIndexes, 0); for (int y = 0; y < blockSize; y++) { missingIndexes[y] = dimensions - blockSize + y; } for (int k = 0; k < horizon; k++) { // shift all entries in the query point left by 1 block System.arraycopy(queryPoint, blockSize, queryPoint, 0, dimensions - blockSize); SampleSummary imputedSummary = getConditionalFieldSummary(queryPoint, blockSize, missingIndexes, centrality); for (int y = 0; y < blockSize; y++) { result.values[resultIndex] = queryPoint[dimensions - blockSize + y] = imputedSummary.median[dimensions - blockSize + y]; result.lower[resultIndex] = imputedSummary.lower[dimensions - blockSize + y]; result.upper[resultIndex] = imputedSummary.upper[dimensions - blockSize + y]; resultIndex++; } } } void extrapolateBasicCyclic(RangeVector result, int horizon, int blockSize, int shingleIndex, float[] queryPoint, int[] missingIndexes, double centrality) { int resultIndex = 0; int currentPosition = shingleIndex; Arrays.fill(missingIndexes, 0); for (int k = 0; k < horizon; k++) { for (int y = 0; y < blockSize; y++) { missingIndexes[y] = (currentPosition + y) % dimensions; } SampleSummary imputedSummary = getConditionalFieldSummary(queryPoint, blockSize, missingIndexes, centrality); for (int y = 0; y < blockSize; y++) { result.values[resultIndex] = queryPoint[(currentPosition + y) % dimensions] = imputedSummary.median[(currentPosition + y) % dimensions]; result.lower[resultIndex] = imputedSummary.lower[(currentPosition + y) % dimensions]; result.upper[resultIndex] = imputedSummary.upper[(currentPosition + y) % dimensions]; resultIndex++; } currentPosition = (currentPosition + blockSize) % dimensions; } } /** * Extrapolate the stream into the future to produce a forecast. This method is * intended to be called when the input data is being shingled internally, and * it works by imputing forward one shingle block at a time. * * @param horizon The number of blocks to forecast. * @return a forecasted time series. */ public double[] extrapolate(int horizon) { return toDoubleArray(extrapolateFromCurrentTime(horizon)); } public float[] extrapolateFromCurrentTime(int horizon) { checkArgument(internalShinglingEnabled, "incorrect use"); IPointStore<?, ?> store = stateCoordinator.getStore(); return extrapolateBasic(lastShingledPoint(), horizon, inputDimensions, store.isInternalRotationEnabled(), ((int) nextSequenceIndex()) % shingleSize); } /** * For each tree in the forest, follow the tree traversal path and return the * leaf node if the standard Euclidean distance between the query point and the * leaf point is smaller than the given threshold. Note that this will not * necessarily be the nearest point in the tree, because the traversal path is * determined by the random cuts in the tree. If the same leaf point is found in * multiple trees, those results will be combined into a single Neighbor in the * result. * * If sequence indexes are disabled for this forest, then the list of sequence * indexes will be empty in returned Neighbors. * * @param point A point whose neighbors we want to find. * @param distanceThreshold The maximum Euclidean distance for a point to be * considered a neighbor. * @return a list of Neighbors, ordered from closest to furthest. */ @Deprecated public List<Neighbor> getNearNeighborsInSample(double[] point, double distanceThreshold) { return getNearNeighborsInSample(toFloatArray(point), distanceThreshold); } public List<Neighbor> getNearNeighborsInSample(float[] point, double distanceThreshold) { checkNotNull(point, "point must not be null"); checkArgument(distanceThreshold > 0, "distanceThreshold must be greater than 0"); if (!isOutputReady()) { return Collections.emptyList(); } IVisitorFactory<Optional<Neighbor>> visitorFactory = (tree, x) -> new NearNeighborVisitor(x, distanceThreshold); return traverseForest(transformToShingledPoint(point), visitorFactory, Neighbor.collector()); } /** * For each tree in the forest, follow the tree traversal path and return the * leaf node. Note that this will not necessarily be the nearest point in the * tree, because the traversal path is determined by the random cuts in the * tree. If the same leaf point is found in multiple trees, those results will * be combined into a single Neighbor in the result. * * If sequence indexes are disabled for this forest, then sequenceIndexes will * be empty in the returned Neighbors. * * @param point A point whose neighbors we want to find. * @return a list of Neighbors, ordered from closest to furthest. */ @Deprecated public List<Neighbor> getNearNeighborsInSample(double[] point) { return getNearNeighborsInSample(toFloatArray(point)); } public List<Neighbor> getNearNeighborsInSample(float[] point) { return getNearNeighborsInSample(point, Double.POSITIVE_INFINITY); } /** * @return true if all samplers are ready to output results. */ public boolean isOutputReady() { return outputReady || (outputReady = stateCoordinator.getTotalUpdates() >= outputAfter && components.stream().allMatch(IComponentModel::isOutputReady)); } /** * @return true if all samplers in the forest are full. */ public boolean samplersFull() { return stateCoordinator.getTotalUpdates() >= sampleSize; } /** * Returns the total number updates to the forest. * * The count of updates is represented with long type and may overflow. * * @return the total number of updates to the forest. */ public long getTotalUpdates() { return stateCoordinator.getTotalUpdates(); } /** * an L1 clustering primitive that shows the aggregation of the points stored in * RCF the clustering uses multi-centroid clustering introduced in CURE * https://en.wikipedia.org/wiki/CURE_algorithm However CURE also shrunk the * well scattered points by a fraction alpha (there by creating new points); * while that concept is used herein, the (multi) summarization algorithm * changes the distance metric as opposed to creating new points since * continuity of values is not an useful assumption in context of RCFs. The * usage of distance metric is similar to the discussion in * https://en.wikipedia.org/wiki/Data_stream_clustering See the examples package * for an example of dynamic summarization. / * * @param maxAllowed maximum number of clusters one is willing to * see * @param shrinkage a parameter that controls between spherical * nature (=1) and MST (=0), this corresponds to * the parameter alpha in the description above * @param numberOfRepresentatives number of centroids used to represent a * cluster, this is the parameter c in the * description of CURE * @param separationRatio a parameter in [0,1] that controls how * zealously should the algorithm reduce the * number of clusters a default value of 0.8 is a * reasonable value for many settings. A value * close to 0 would tend to merge eveything into * a single cluster. The option is provided since * it can be of use in the future to produce * dendograms and similar information. * @param distance a distance function for points * @param previous a (possibly null) list of previous clustering * obtained. If the list is non-null then the * representatives of the previous cluster would * be added as zero weight points, ensuring that * the summarization is more smooth (in contrast * to two independent summarizations). The zero * weight points of the past can serve as * representatives of the current clustering. * @return a list of clusters */ public List<ICluster<float[]>> summarize(int maxAllowed, double shrinkage, int numberOfRepresentatives, double separationRatio, BiFunction<float[], float[], Double> distance, List<ICluster<float[]>> previous) { return stateCoordinator.getStore().summarize(maxAllowed, shrinkage, numberOfRepresentatives, separationRatio, distance, previous); } // same as above with default filled in public List<ICluster<float[]>> summarize(int maxAllowed, double shrinkage, int numberOfRepresentatives, List<ICluster<float[]>> previous) { return stateCoordinator.getStore().summarize(maxAllowed, shrinkage, numberOfRepresentatives, DEFAULT_SEPARATION_RATIO_FOR_MERGE, Summarizer::L1distance, previous); } public static class Builder<T extends Builder<T>> { // We use Optional types for optional primitive fields when it doesn't make // sense to use a constant default. private int dimensions; private int sampleSize = DEFAULT_SAMPLE_SIZE; private Optional<Integer> outputAfter = Optional.empty(); private int numberOfTrees = DEFAULT_NUMBER_OF_TREES; private Optional<Double> timeDecay = Optional.empty(); private Optional<Long> randomSeed = Optional.empty(); private boolean storeSequenceIndexesEnabled = DEFAULT_STORE_SEQUENCE_INDEXES_ENABLED; private boolean centerOfMassEnabled = DEFAULT_CENTER_OF_MASS_ENABLED; private boolean parallelExecutionEnabled = DEFAULT_PARALLEL_EXECUTION_ENABLED; private Optional<Integer> threadPoolSize = Optional.empty(); private boolean directLocationMapEnabled = DEFAULT_DIRECT_LOCATION_MAP; private double boundingBoxCacheFraction = DEFAULT_BOUNDING_BOX_CACHE_FRACTION; private int shingleSize = DEFAULT_SHINGLE_SIZE; protected boolean dynamicResizingEnabled = DEFAULT_DYNAMIC_RESIZING_ENABLED; private boolean internalShinglingEnabled = DEFAULT_INTERNAL_SHINGLING_ENABLED; protected boolean internalRotationEnabled = DEFAULT_INTERNAL_ROTATION_ENABLED; protected Optional<Integer> initialPointStoreSize = Optional.empty(); protected double initialAcceptFraction = DEFAULT_INITIAL_ACCEPT_FRACTION; public T dimensions(int dimensions) { this.dimensions = dimensions; return (T) this; } public T sampleSize(int sampleSize) { this.sampleSize = sampleSize; return (T) this; } public T outputAfter(int outputAfter) { this.outputAfter = Optional.of(outputAfter); return (T) this; } public T numberOfTrees(int numberOfTrees) { this.numberOfTrees = numberOfTrees; return (T) this; } public T shingleSize(int shingleSize) { this.shingleSize = shingleSize; return (T) this; } public T timeDecay(double timeDecay) { this.timeDecay = Optional.of(timeDecay); return (T) this; } public T randomSeed(long randomSeed) { this.randomSeed = Optional.of(randomSeed); return (T) this; } public T centerOfMassEnabled(boolean centerOfMassEnabled) { this.centerOfMassEnabled = centerOfMassEnabled; return (T) this; } public T parallelExecutionEnabled(boolean parallelExecutionEnabled) { this.parallelExecutionEnabled = parallelExecutionEnabled; return (T) this; } public T threadPoolSize(int threadPoolSize) { this.threadPoolSize = Optional.of(threadPoolSize); return (T) this; } public T initialPointStoreSize(int initialPointStoreSize) { this.initialPointStoreSize = Optional.of(initialPointStoreSize); return (T) this; } public T storeSequenceIndexesEnabled(boolean storeSequenceIndexesEnabled) { this.storeSequenceIndexesEnabled = storeSequenceIndexesEnabled; return (T) this; } @Deprecated public T compact(boolean compact) { return (T) this; } public T internalShinglingEnabled(boolean internalShinglingEnabled) { this.internalShinglingEnabled = internalShinglingEnabled; return (T) this; } public T internalRotationEnabled(boolean internalRotationEnabled) { this.internalRotationEnabled = internalRotationEnabled; return (T) this; } public T dynamicResizingEnabled(boolean dynamicResizingEnabled) { this.dynamicResizingEnabled = dynamicResizingEnabled; return (T) this; } @Deprecated public T precision(Precision precision) { return (T) this; } public T boundingBoxCacheFraction(double boundingBoxCacheFraction) { this.boundingBoxCacheFraction = boundingBoxCacheFraction; return (T) this; } public T initialAcceptFraction(double initialAcceptFraction) { this.initialAcceptFraction = initialAcceptFraction; return (T) this; } public RandomCutForest build() { return new RandomCutForest(this); } public Random getRandom() { // If a random seed was given, use it to create a new Random. Otherwise, call // the 0-argument constructor return randomSeed.map(Random::new).orElseGet(Random::new); } } /** * Score a point using the given scoring functions. * * @param point input point being scored * @param ignoreLeafMassThreshold said threshold * @param seen the function that applies if input is equal to * a previously seen sample in a leaf * @param unseen if the input does not have a match in the * leaves * @param damp damping function based on the duplicity of the * previously seen samples * @return anomaly score */ public double getDynamicScore(float[] point, int ignoreLeafMassThreshold, BiFunction<Double, Double, Double> seen, BiFunction<Double, Double, Double> unseen, BiFunction<Double, Double, Double> damp) { checkArgument(ignoreLeafMassThreshold >= 0, "ignoreLeafMassThreshold should be greater than or equal to 0"); if (!isOutputReady()) { return 0.0; } VisitorFactory<Double> visitorFactory = new VisitorFactory<>((tree, y) -> new DynamicScoreVisitor( tree.projectToTree(y), tree.getMass(), ignoreLeafMassThreshold, seen, unseen, damp)); BinaryOperator<Double> accumulator = Double::sum; Function<Double, Double> finisher = sum -> sum / numberOfTrees; return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Similar to above but now the scoring takes in a function of Bounding Box to * probabilities (vector over the dimensions); and produces a score af-if the * tree were built using that function (when in reality the tree is an RCF). * Changing the defaultRCFgVec function to some other function f() will provide * a mechanism of dynamic scoring for trees that are built using f() which is * the purpose of TransductiveScalarScore visitor. Note that the answer is an * MCMC simulation and is not normalized (because the scoring functions are * flexible and unknown) and over a small number of trees the errors can be * large specially if vecSep is very far from defaultRCFgVec * * Given the large number of possible sources of distortion, ignoreLeafThreshold * is not supported. * * @param point point to be scored * @param seen the score function for seen point * @param unseen score function for unseen points * @param damp dampening the score for duplicates * @param vecSep the function of (BoundingBox) -&gt; array of probabilities * @return the simuated score */ public double getDynamicSimulatedScore(float[] point, BiFunction<Double, Double, Double> seen, BiFunction<Double, Double, Double> unseen, BiFunction<Double, Double, Double> damp, Function<IBoundingBoxView, double[]> vecSep) { if (!isOutputReady()) { return 0.0; } VisitorFactory<Double> visitorFactory = new VisitorFactory<>( (tree, y) -> new SimulatedTransductiveScalarScoreVisitor(tree.projectToTree(y), tree.getMass(), seen, unseen, damp, CommonUtils::defaultRCFgVecFunction, vecSep)); BinaryOperator<Double> accumulator = Double::sum; Function<Double, Double> finisher = sum -> sum / numberOfTrees; return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Score a point using the given scoring functions. This method will * short-circuit before visiting all trees if the scores that are returned from * a subset of trees appears to be converging to a given value. See * {@link OneSidedConvergingDoubleAccumulator} for more about convergence. * * @param point input point * @param precision controls early convergence * @param highIsCritical this is true for the default scoring function. * If the user wishes to use a different scoring * function where anomaly scores are low values * (for example, height in tree) then this should * be set to false. * @param ignoreLeafMassThreshold said threshold * @param seen scoring function when the input matches some * tuple in the leaves * @param unseen scoring function when the input is not found * @param damp dampening function for duplicates which are * same as input (applies with seen) * @return the dynamic score under sequential early stopping */ public double getApproximateDynamicScore(float[] point, double precision, boolean highIsCritical, int ignoreLeafMassThreshold, BiFunction<Double, Double, Double> seen, BiFunction<Double, Double, Double> unseen, BiFunction<Double, Double, Double> damp) { checkArgument(ignoreLeafMassThreshold >= 0, "ignoreLeafMassThreshold should be greater than or equal to 0"); if (!isOutputReady()) { return 0.0; } VisitorFactory<Double> visitorFactory = new VisitorFactory<>((tree, y) -> new DynamicScoreVisitor( tree.projectToTree(y), tree.getMass(), ignoreLeafMassThreshold, seen, unseen, damp)); ConvergingAccumulator<Double> accumulator = new OneSidedConvergingDoubleAccumulator(highIsCritical, precision, DEFAULT_APPROXIMATE_DYNAMIC_SCORE_MIN_VALUES_ACCEPTED, numberOfTrees); Function<Double, Double> finisher = x -> x / accumulator.getValuesAccepted(); return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Same as above, but for dynamic scoring. See the params of * getDynamicScoreParallel * * @param point point to be scored * @param ignoreLeafMassThreshold said threshold * @param seen score function for seen points * @param unseen score function for unseen points * @param newDamp dampening function for duplicates in the seen * function * @return dynamic scoring attribution DiVector */ public DiVector getDynamicAttribution(float[] point, int ignoreLeafMassThreshold, BiFunction<Double, Double, Double> seen, BiFunction<Double, Double, Double> unseen, BiFunction<Double, Double, Double> newDamp) { if (!isOutputReady()) { return new DiVector(dimensions); } VisitorFactory<DiVector> visitorFactory = new VisitorFactory<>( (tree, y) -> new DynamicAttributionVisitor(tree.projectToTree(y), tree.getMass(), ignoreLeafMassThreshold, seen, unseen, newDamp), (tree, x) -> x.lift(tree::liftFromTree)); BinaryOperator<DiVector> accumulator = DiVector::addToLeft; Function<DiVector, DiVector> finisher = x -> x.scale(1.0 / numberOfTrees); return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } /** * Atrribution for dynamic sequential scoring; getL1Norm() should agree with * getDynamicScoringSequential * * @param point input * @param precision parameter to stop early stopping * @param highIsCritical are high values anomalous (otherwise low * values are anomalous) * @param ignoreLeafMassThreshold we ignore leaves with mass equal/below * * threshold * @param seen function for scoring points that have been * seen before * @param unseen function for scoring points not seen in tree * @param newDamp dampening function based on duplicates * @return attribution DiVector of the score */ public DiVector getApproximateDynamicAttribution(float[] point, double precision, boolean highIsCritical, int ignoreLeafMassThreshold, BiFunction<Double, Double, Double> seen, BiFunction<Double, Double, Double> unseen, BiFunction<Double, Double, Double> newDamp) { if (!isOutputReady()) { return new DiVector(dimensions); } VisitorFactory<DiVector> visitorFactory = new VisitorFactory<>((tree, y) -> new DynamicAttributionVisitor(y, tree.getMass(), ignoreLeafMassThreshold, seen, unseen, newDamp), (tree, x) -> x.lift(tree::liftFromTree)); ConvergingAccumulator<DiVector> accumulator = new OneSidedConvergingDiVectorAccumulator(dimensions, highIsCritical, precision, DEFAULT_APPROXIMATE_DYNAMIC_SCORE_MIN_VALUES_ACCEPTED, numberOfTrees); Function<DiVector, DiVector> finisher = vector -> vector.scale(1.0 / accumulator.getValuesAccepted()); return traverseForest(transformToShingledPoint(point), visitorFactory, accumulator, finisher); } }
441
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/ComponentList.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import java.util.ArrayList; import java.util.Collection; /** * A ComponentList is an ArrayList specialized to contain IComponentModel * instances. Executor classes operate on ComponentLists. * * @param <PointReference> The internal point representation expected by the * component models in this list. * @param <Point> The explicit data type of points being passed */ public class ComponentList<PointReference, Point> extends ArrayList<IComponentModel<PointReference, Point>> { public ComponentList() { super(); } public ComponentList(Collection<? extends IComponentModel<PointReference, Point>> collection) { super(collection); } public ComponentList(int initialCapacity) { super(initialCapacity); } }
442
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/IVisitorFactory.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import com.amazon.randomcutforest.tree.ITree; @FunctionalInterface public interface IVisitorFactory<R> { Visitor<R> newVisitor(ITree<?, ?> tree, float[] point); default R liftResult(ITree<?, ?> tree, R result) { return result; } }
443
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/MultiVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import com.amazon.randomcutforest.tree.INodeView; /** * This is the interface for a visitor which can be used with * {RandomCutTree::traversePathToLeafAndVisitNodesMulti}. In this traversal * method, we optionally choose to split the visitor into two copies when * visiting nodes. Each copy then visits one of the paths down from that node. * The results from both visitors are combined before returning back up the * tree. */ public interface MultiVisitor<R> extends Visitor<R> { /** * Returns true of the traversal method should split the visitor (i.e., create a * copy) at this node. * * @param node A node in the tree traversal * @return true if the traversal should split the visitor into two copies at * this node, false otherwise. */ boolean trigger(final INodeView node); /** * Return a copy of this visitor. The original visitor plus the copy will each * traverse one branch of the tree. * * @return a copy of this visitor */ MultiVisitor<R> newCopy(); /** * Combine two visitors. The state of the argument visitor should be combined * with the state of this instance. This method is called after both visitors * have traversed one branch of the tree. * * @param other A second visitor */ void combine(MultiVisitor<R> other); }
444
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/Visitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import com.amazon.randomcutforest.tree.INodeView; /** * This is the interface for a visitor which can be used to query a ITraversable * to produce a result. A visitor is submitted to * ITraversable#traverse(double[], Visitor), and during the traversal the * {@link #acceptLeaf} and {@link #accept} methods are invoked on the nodes in * the traversal path. * <p> * See ITraversable#traverse(double[], Visitor) for details about the traversal * path. */ public interface Visitor<R> { /** * Visit a node in the traversal path. * * @param node the node being visited * @param depthOfNode the depth of the node being visited */ void accept(INodeView node, int depthOfNode); /** * Visit the leaf node in the traversal path. By default, this method proxies to * {@link #accept(INodeView, int)}. * * @param leafNode the leaf node being visited * @param depthOfNode the depth of the leaf node */ default void acceptLeaf(INodeView leafNode, final int depthOfNode) { accept(leafNode, depthOfNode); } /** * At the end of the traversal, this method is called to obtain the result * computed by the visitor. * * @return the result value computed by the visitor. */ R getResult(); /** * This method short-circuits the evaluation of the Visitor at nodes on the * traversal path. By default, the accept (or acceptLeaf) method will be invoked * for each Node in the traversal path. But the NodeView has to prepare * information to support that visitor invocation. Before invocation, the value * of isConverged will be checked. If it is true, some of that preparation can * be skipped -- because the visitor would not be updated. This method can be * overwritten to optimize visitors that do not need to visit every node on the * root to leaf path before returning a value. * * Mote that this convergence applies to a single visitor computation and is * expected to be a speedup without any change in the value of the answer. This * is different from converging accumulator which corresponds to sequential * evaluation of different visitors and early stopping. **/ default boolean isConverged() { return false; } }
445
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/CommonUtils.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import java.util.Objects; import java.util.function.Supplier; import com.amazon.randomcutforest.tree.IBoundingBoxView; /** * A collection of common utility functions. */ public class CommonUtils { private CommonUtils() { } /** * Throws an {@link IllegalArgumentException} with the specified message if the * specified input is false. * * @param condition A condition to test. * @param message The error message to include in the * {@code IllegalArgumentException} if {@code condition} is * false. * @throws IllegalArgumentException if {@code condition} is false. */ public static void checkArgument(boolean condition, String message) { if (!condition) { throw new IllegalArgumentException(message); } } // a lazy equivalent of the above, which avoids parameter evaluation public static void checkArgument(boolean condition, Supplier<String> messageSupplier) { if (!condition) { throw new IllegalArgumentException(messageSupplier.get()); } } /** * Throws an {@link IllegalStateException} with the specified message if the * specified input is false. * * @param condition A condition to test. * @param message The error message to include in the * {@code IllegalStateException} if {@code condition} is false. * @throws IllegalStateException if {@code condition} is false. */ public static void checkState(boolean condition, String message) { if (!condition) { throw new IllegalStateException(message); } } /** * Throws an {@link IllegalStateException} with the specified message if the * specified input is false. This would eventually become asserts. * * @param condition A condition to test. * @param message The error message to include in the * {@code IllegalStateException} if {@code condition} is false. * @throws IllegalStateException if {@code condition} is false. */ public static void validateInternalState(boolean condition, String message) { if (!condition) { throw new IllegalStateException(message); } } /** * Throws a {@link NullPointerException} with the specified message if the * specified input is null. * * @param <T> An arbitrary type. * @param object An object reference to test for nullity. * @param message The error message to include in the * {@code NullPointerException} if {@code object} is null. * @return {@code object} if not null. * @throws NullPointerException if the supplied object is null. */ public static <T> T checkNotNull(T object, String message) { Objects.requireNonNull(object, message); return object; } /** * Compute the probability of separation for a bounding box adn a point. This * method considers the bounding box created by merging the query point into the * existing bounding box, and computes the probability that a random cut would * separate the query point from the merged bounding box. * * @param boundingBox is the bounding box used in RandomCutTree * @param queryPoint is the multidimensional point * @return the probability of separation choosing a random cut */ public static double getProbabilityOfSeparation(final IBoundingBoxView boundingBox, float[] queryPoint) { double sumOfNewRange = 0d; double sumOfDifferenceInRange = 0d; for (int i = 0; i < queryPoint.length; ++i) { double maxVal = boundingBox.getMaxValue(i); double minVal = boundingBox.getMinValue(i); double oldRange = maxVal - minVal; if (maxVal < queryPoint[i]) { maxVal = queryPoint[i]; } else if (minVal > queryPoint[i]) { minVal = queryPoint[i]; } else { sumOfNewRange += oldRange; continue; } double newRange = maxVal - minVal; sumOfNewRange += newRange; sumOfDifferenceInRange += (newRange - oldRange); } if (sumOfNewRange <= 0) { return 0; } else return sumOfDifferenceInRange / sumOfNewRange; } /** * The default anomaly scoring function for points that contained in a tree. * * @param depth The depth of the leaf node where this method is invoked * @param mass The number of times the point has been seen before * @return The score contribution from this previously-seen point */ public static double defaultScoreSeenFunction(double depth, double mass) { return 1.0 / (depth + Math.log(mass + 1.0) / Math.log(2.0)); } /** * The default anomaly scoring function for points not already contained in a * tree. * * @param depth The depth of the leaf node where this method is invoked * @param mass The number of times the point has been seen before * @return The score contribution from this point */ public static double defaultScoreUnseenFunction(double depth, double mass) { return 1.0 / (depth + 1); } public static double defaultDampFunction(double leafMass, double treeMass) { return 1.0 - leafMass / (2 * treeMass); } /** * Some algorithms which return a scalar value need to scale that value by tree * mass for consistency. This is the default method for computing the scale * factor in these cases. The function has to be associative in its first * argument (when the second is fixed) That is, fn (x1, y) + fn (x2, y) = fn (x1 * + x2, y) * * @param scalarValue The value being scaled * @param mass The mass of the tree where this method is invoked * @return The original value scaled appropriately for this tree */ public static double defaultScalarNormalizerFunction(double scalarValue, double mass) { return scalarValue * Math.log(mass + 1) / Math.log(2.0); } /** * The following function forms the core of RCFs, given a BoundingBox it * produces the probability of cutting in different dimensions. While this * function is absorbed in the logic of the different simpler scoring methods, * the scoring methods that are mode advanced (for example, trying to simulate * an Transductive Isolation Forest with streaming) require this function. A * different function can be used to simulate via non-RCFs. * * @param boundingBox bounding box of a set of points * @return array of probabilities of cutting in that specific dimension */ public static double[] defaultRCFgVecFunction(IBoundingBoxView boundingBox) { double[] answer = new double[boundingBox.getDimensions()]; for (int i = 0; i < boundingBox.getDimensions(); ++i) { double maxVal = boundingBox.getMaxValue(i); double minVal = boundingBox.getMinValue(i); double oldRange = maxVal - minVal; if (oldRange > 0) { answer[i] = oldRange; } } return answer; }; public static double[] toDoubleArray(float[] array) { checkNotNull(array, "array must not be null"); double[] result = new double[array.length]; for (int i = 0; i < array.length; i++) { result[i] = array[i]; } return result; } public static float[] toFloatArray(double[] array) { checkNotNull(array, "array must not be null"); float[] result = new float[array.length]; for (int i = 0; i < array.length; i++) { result[i] = (array[i] == 0) ? 0 : (float) array[i]; // eliminating -0.0 issues } return result; } public static int[] toIntArray(byte[] values) { checkNotNull(values, "array must not be null"); int[] result = new int[values.length]; for (int i = 0; i < values.length; i++) { result[i] = values[i] & 0xff; } return result; } public static int[] toIntArray(char[] values) { checkNotNull(values, "array must not be null"); int[] result = new int[values.length]; for (int i = 0; i < values.length; i++) { result[i] = values[i]; } return result; } public static char[] toCharArray(int[] values) { checkNotNull(values, "array must not be null"); char[] result = new char[values.length]; for (int i = 0; i < values.length; i++) { result[i] = (char) values[i]; } return result; } public static byte[] toByteArray(int[] values) { checkNotNull(values, "array must not be null"); byte[] result = new byte[values.length]; for (int i = 0; i < values.length; i++) { result[i] = (byte) values[i]; } return result; } }
446
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/IComponentModel.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import com.amazon.randomcutforest.config.IDynamicConfig; import com.amazon.randomcutforest.executor.ITraversable; import com.amazon.randomcutforest.executor.IUpdatable; /** * * @param <PointReference> The internal point representation expected by the * component models in this list. * @param <Point> The explicit data type of points being passed */ public interface IComponentModel<PointReference, Point> extends ITraversable, IUpdatable<PointReference>, IDynamicConfig { }
447
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/IMultiVisitorFactory.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import com.amazon.randomcutforest.tree.ITree; @FunctionalInterface public interface IMultiVisitorFactory<R> { MultiVisitor<R> newVisitor(ITree<?, ?> tree, float[] point); default R liftResult(ITree<?, ?> tree, R result) { return result; } }
448
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/MultiVisitorFactory.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest; import java.util.function.BiFunction; import com.amazon.randomcutforest.tree.ITree; /** * This is the interface for a visitor which can be used with * {RandomCutTree::traversePathToLeafAndVisitNodesMulti}. In this traversal * method, we optionally choose to split the visitor into two copies when * visiting nodes. Each copy then visits one of the paths down from that node. * The results from both visitors are combined before returning back up the * tree. */ public class MultiVisitorFactory<R> implements IMultiVisitorFactory<R> { private final BiFunction<ITree<?, ?>, float[], MultiVisitor<R>> newVisitor; private final BiFunction<ITree<?, ?>, R, R> liftResult; public MultiVisitorFactory(BiFunction<ITree<?, ?>, float[], MultiVisitor<R>> newVisitor, BiFunction<ITree<?, ?>, R, R> liftResult) { this.newVisitor = newVisitor; this.liftResult = liftResult; } public MultiVisitorFactory(BiFunction<ITree<?, ?>, float[], MultiVisitor<R>> newVisitor) { this(newVisitor, (tree, x) -> x); } @Override public MultiVisitor<R> newVisitor(ITree<?, ?> tree, float[] point) { return newVisitor.apply(tree, point); } @Override public R liftResult(ITree<?, ?> tree, R result) { return liftResult.apply(tree, result); } }
449
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/INodeView.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import java.util.HashMap; public interface INodeView { boolean isLeaf(); int getMass(); IBoundingBoxView getBoundingBox(); IBoundingBoxView getSiblingBoundingBox(float[] point); int getCutDimension(); double getCutValue(); float[] getLeafPoint(); default float[] getLiftedLeafPoint() { return getLeafPoint(); }; /** * for a leaf node, return the sequence indices corresponding leaf point. If * this method is invoked on a non-leaf node then it throws an * IllegalStateException. */ HashMap<Long, Integer> getSequenceIndexes(); /** * provides the probability of separation vis-a-vis the bounding box at the node * * @param point input piint being evaluated * @return the probability of separation */ double probailityOfSeparation(float[] point); /** * for a leaf node, return the index in the point store for the leaf point. If * this method is invoked on a non-leaf node then it throws an * IllegalStateException. */ int getLeafPointIndex(); }
450
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/NodeStoreMedium.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.toCharArray; import static com.amazon.randomcutforest.CommonUtils.toIntArray; import java.util.Arrays; import java.util.BitSet; import java.util.Stack; import com.amazon.randomcutforest.store.IndexIntervalManager; /** * A fixed-size buffer for storing interior tree nodes. An interior node is * defined by its location in the tree (parent and child nodes), its random cut, * and its bounding box. The NodeStore class uses arrays to store these field * values for a collection of nodes. An index in the store can be used to look * up the field values for a particular node. * * The internal nodes (handled by this store) corresponds to [0..capacity]. The * mass of the nodes is cyclic, i.e., mass % (capacity + 1) -- therefore, in * presence of duplicates there would be nodes which are free, and they would * have mass 0 == (capacity + 1). But those nodes would not be reachable by the * code below. * */ public class NodeStoreMedium extends AbstractNodeStore { private final char[] parentIndex; private final int[] leftIndex; private final int[] rightIndex; public final char[] cutDimension; private final char[] mass; public NodeStoreMedium(AbstractNodeStore.Builder builder) { super(builder); mass = new char[capacity]; Arrays.fill(mass, (char) 0); if (builder.storeParent) { parentIndex = new char[capacity]; Arrays.fill(parentIndex, (char) capacity); } else { parentIndex = null; } if (builder.leftIndex == null) { leftIndex = new int[capacity]; rightIndex = new int[capacity]; cutDimension = new char[capacity]; Arrays.fill(leftIndex, capacity); Arrays.fill(rightIndex, capacity); } else { leftIndex = Arrays.copyOf(builder.leftIndex, builder.leftIndex.length); rightIndex = Arrays.copyOf(builder.rightIndex, builder.rightIndex.length); cutDimension = toCharArray(builder.cutDimension); BitSet bits = new BitSet(capacity); if (builder.root != Null) { bits.set(builder.root); } for (int i = 0; i < leftIndex.length; i++) { if (isInternal(leftIndex[i])) { bits.set(leftIndex[i]); if (parentIndex != null) { parentIndex[leftIndex[i]] = (char) i; } } } for (int i = 0; i < rightIndex.length; i++) { if (isInternal(rightIndex[i])) { bits.set(rightIndex[i]); if (parentIndex != null) { parentIndex[rightIndex[i]] = (char) i; } } } freeNodeManager = new IndexIntervalManager(capacity, capacity, bits); // need to set up parents using the root } } @Override public int addNode(Stack<int[]> pathToRoot, float[] point, long sequenceIndex, int pointIndex, int childIndex, int childMassIfLeaf, int cutDimension, float cutValue, BoundingBox box) { int index = freeNodeManager.takeIndex(); this.cutValue[index] = cutValue; this.cutDimension[index] = (char) cutDimension; if (leftOf(cutValue, cutDimension, point)) { this.leftIndex[index] = (pointIndex + capacity + 1); this.rightIndex[index] = childIndex; } else { this.rightIndex[index] = (pointIndex + capacity + 1); this.leftIndex[index] = childIndex; } this.mass[index] = (char) ((((childMassIfLeaf > 0) ? childMassIfLeaf : getMass(childIndex)) + 1) % (capacity + 1)); int parentIndex = (pathToRoot.size() == 0) ? Null : pathToRoot.lastElement()[0]; if (this.parentIndex != null) { this.parentIndex[index] = (char) parentIndex; if (!isLeaf(childIndex)) { this.parentIndex[childIndex] = (char) (index); } } if (parentIndex != Null) { spliceEdge(parentIndex, childIndex, index); } return index; } @Override public void assignInPartialTree(int node, float[] point, int childReference) { if (leftOf(node, point)) { leftIndex[node] = childReference; } else { rightIndex[node] = childReference; } } public int getLeftIndex(int index) { return leftIndex[index]; } public int getRightIndex(int index) { return rightIndex[index]; } public int getParentIndex(int index) { checkArgument(parentIndex != null, "incorrect call"); return parentIndex[index]; } public void setRoot(int index) { if (!isLeaf(index) && parentIndex != null) { parentIndex[index] = (char) capacity; } } @Override protected void decreaseMassOfInternalNode(int node) { mass[node] = (char) ((mass[node] + capacity) % (capacity + 1)); // this cannot get to 0 } @Override protected void increaseMassOfInternalNode(int node) { mass[node] = (char) ((mass[node] + 1) % (capacity + 1)); // mass of root == 0; note capacity = number_of_leaves - 1 } public void deleteInternalNode(int index) { leftIndex[index] = capacity; rightIndex[index] = capacity; if (parentIndex != null) { parentIndex[index] = (char) capacity; } freeNodeManager.releaseIndex(index); } public int getMass(int index) { return mass[index] != 0 ? mass[index] : (capacity + 1); } public void spliceEdge(int parent, int node, int newNode) { assert (!isLeaf(newNode)); if (node == leftIndex[parent]) { leftIndex[parent] = newNode; } else { rightIndex[parent] = newNode; } if (parentIndex != null && isInternal(node)) { parentIndex[node] = (char) newNode; } } public void replaceParentBySibling(int grandParent, int parent, int node) { int sibling = getSibling(node, parent); if (parent == leftIndex[grandParent]) { leftIndex[grandParent] = sibling; } else { rightIndex[grandParent] = sibling; } if (parentIndex != null && isInternal(sibling)) { parentIndex[sibling] = (char) grandParent; } } public int getCutDimension(int index) { return cutDimension[index]; } public int[] getCutDimension() { return toIntArray(cutDimension); } public int[] getLeftIndex() { return Arrays.copyOf(leftIndex, leftIndex.length); } public int[] getRightIndex() { return Arrays.copyOf(rightIndex, rightIndex.length); } }
451
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/Cut.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; /** * A Cut represents a division of space into two half-spaces. Cuts are used to * define the tree structure in {@link RandomCutTree}, and they determine the * standard tree traversal path defined in {@link RandomCutTree#traverse}. */ public class Cut { private final int dimension; private final double value; /** * Create a new Cut with the given dimension and value. * * @param dimension The 0-based index of the dimension that the cut is made in. * @param value The spatial value of the cut. */ public Cut(int dimension, double value) { this.dimension = dimension; this.value = value; } /** * For the given point, this method compares the value of that point in the cut * dimension to the cut value. If the point's value in the cut dimension is less * than or equal to the cut value this method returns true, otherwise it returns * false. The name of this method is a mnemonic: if we are working in a * one-dimensional space, then this method will return 'true' if the point value * is to the left of the cut value on the standard number line. * * @param point A point that we are testing in relation to the cut * @param cut A Cut instance. * @return true if the value of the point coordinate corresponding to the cut * dimension is less than or equal to the cut value, false otherwise. */ public static boolean isLeftOf(double[] point, Cut cut) { return point[cut.getDimension()] <= cut.getValue(); } /** * Return the index of the dimension that this cut was made in. * * @return the 0-based index of the dimension that this cut was made in. */ public int getDimension() { return dimension; } /** * Return the value of the cut. This value separates space into two half-spaces: * the set of points whose coordinate in the cut dimension is less than the cut * value, and the set of points whose coordinate in the cut dimension is greater * than the cut value. * * @return the value of the cut. */ public double getValue() { return value; } @Override public String toString() { return String.format("Cut(%d, %f)", dimension, value); } }
452
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/BoundingBox.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.checkState; import java.util.Arrays; /** * A single precision implementation of AbstractBoundingBox which also satisfies * the interface for Visitor classes */ public class BoundingBox implements IBoundingBoxView { /** * An array containing the minimum value corresponding to each dimension. */ protected final float[] minValues; /** * An array containing the maximum value corresponding to each dimensions */ protected final float[] maxValues; /** * The sum of side lengths defined by this bounding box. */ protected double rangeSum; public BoundingBox(float[] point) { minValues = maxValues = point; // a copy in not needed because mergedBox would create a copy // addPoint, addBox would also create copies rangeSum = 0.0; } /** * Create a new BoundingBox with the given minimum values and maximum values. * * @param minValues The minimum values for each coordinate. * @param maxValues The maximum values for each coordinate */ public BoundingBox(final float[] minValues, final float[] maxValues, double sum) { this.minValues = minValues; this.maxValues = maxValues; rangeSum = sum; } public BoundingBox(final float[] first, final float[] second) { checkArgument(first.length == second.length, " incorrect lengths in box"); minValues = new float[first.length]; maxValues = new float[first.length]; rangeSum = 0; for (int i = 0; i < minValues.length; ++i) { minValues[i] = Math.min(first[i], second[i]); maxValues[i] = Math.max(first[i], second[i]); rangeSum += maxValues[i] - minValues[i]; } } public BoundingBox copy() { return new BoundingBox(Arrays.copyOf(minValues, minValues.length), Arrays.copyOf(maxValues, maxValues.length), rangeSum); } public BoundingBox getMergedBox(IBoundingBoxView otherBox) { float[] minValuesMerged = new float[minValues.length]; float[] maxValuesMerged = new float[minValues.length]; double sum = 0.0; for (int i = 0; i < minValues.length; ++i) { minValuesMerged[i] = Math.min(minValues[i], (float) otherBox.getMinValue(i)); maxValuesMerged[i] = Math.max(maxValues[i], (float) otherBox.getMaxValue(i)); sum += maxValuesMerged[i] - minValuesMerged[i]; } return new BoundingBox(minValuesMerged, maxValuesMerged, sum); } public double probabilityOfCut(float[] point) { double range = 0; for (int i = 0; i < point.length; i++) { range += Math.max(minValues[i] - point[i], 0); } for (int i = 0; i < point.length; i++) { range += Math.max(point[i] - maxValues[i], 0); } if (range == 0) { return 0; } else if (rangeSum == 0) { return 1; } else { return range / (range + rangeSum); } } public BoundingBox getMergedBox(float[] point) { checkArgument(point.length == minValues.length, "incorrect length"); return copy().addPoint(point); } public float[] getMaxValues() { return maxValues; } public float[] getMinValues() { return minValues; } public BoundingBox addPoint(float[] point) { checkArgument(minValues.length == point.length, "incorrect length"); checkArgument(minValues != maxValues, "not a mutable box"); rangeSum = 0; for (int i = 0; i < point.length; ++i) { minValues[i] = Math.min(minValues[i], point[i]); } for (int i = 0; i < point.length; ++i) { maxValues[i] = Math.max(maxValues[i], point[i]); } for (int i = 0; i < point.length; ++i) { rangeSum += maxValues[i] - minValues[i]; } return this; } public BoundingBox addBox(BoundingBox otherBox) { checkState(minValues != maxValues, "not a mutable box"); rangeSum = 0; for (int i = 0; i < minValues.length; ++i) { minValues[i] = Math.min(minValues[i], otherBox.minValues[i]); } for (int i = 0; i < minValues.length; ++i) { maxValues[i] = Math.max(maxValues[i], otherBox.maxValues[i]); } for (int i = 0; i < minValues.length; ++i) { rangeSum += maxValues[i] - minValues[i]; } return this; } public int getDimensions() { return minValues.length; } /** * @return the sum of side lengths for this BoundingBox. */ public double getRangeSum() { return rangeSum; } /** * Gets the max value of the specified dimension. * * @param dimension the dimension for which we need the max value * @return the max value of the specified dimension */ public double getMaxValue(final int dimension) { return maxValues[dimension]; } /** * Gets the min value of the specified dimension. * * @param dimension the dimension for which we need the min value * @return the min value of the specified dimension */ public double getMinValue(final int dimension) { return minValues[dimension]; } /** * Returns true if the given point is contained in this bounding box. This is * equivalent to the point being a member of the set defined by this bounding * box. * * @param point with which we're performing the comparison * @return whether the point is contained by the bounding box */ public boolean contains(float[] point) { checkArgument(point.length == minValues.length, " incorrect lengths"); for (int i = 0; i < minValues.length; i++) { if (minValues[i] > point[i] || maxValues[i] < point[i]) { return false; } } return true; } public boolean contains(BoundingBox otherBox) { checkArgument(otherBox.minValues.length == minValues.length, " incorrect lengths"); return contains(otherBox.minValues) && contains(otherBox.maxValues); } public double getRange(final int dimension) { return maxValues[dimension] - minValues[dimension]; } @Override public String toString() { return String.format("BoundingBox(%s, %s)", Arrays.toString(minValues), Arrays.toString(maxValues)); } /** * Two bounding boxes are considered equal if they have the same dimensions and * all their min values and max values are the same. Min and max values are * compared as primitive doubles using ==, so two bounding boxes are not equal * if their min and max values are merely very close. * * @param other An object to test for equality * @return true if other is a bounding box with the same min and max values */ @Override public boolean equals(Object other) { if (!(other instanceof BoundingBox)) { return false; } BoundingBox otherBox = (BoundingBox) other; return Arrays.equals(minValues, otherBox.minValues) && Arrays.equals(maxValues, otherBox.maxValues); } }
453
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/ITree.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import com.amazon.randomcutforest.config.IDynamicConfig; import com.amazon.randomcutforest.executor.ITraversable; /** * A tree that can potentially interact with a coordinator * * @param <PointReference> The internal point representation expected by the * component models in this list. * @param <Point> The explicit data type of points being passed */ public interface ITree<PointReference, Point> extends ITraversable, IDynamicConfig { int getMass(); float[] projectToTree(float[] point); float[] liftFromTree(float[] result); double[] liftFromTree(double[] result); int[] projectMissingIndices(int[] list); PointReference addPoint(PointReference point, long sequenceIndex); void addPointToPartialTree(PointReference point, long sequenceIndex); void validateAndReconstruct(); PointReference deletePoint(PointReference point, long sequenceIndex); default long getRandomSeed() { return 0L; } }
454
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/HyperTree.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.tree.AbstractNodeStore.Null; import java.util.ArrayList; import java.util.Arrays; import java.util.List; import java.util.Random; import java.util.function.Function; public class HyperTree extends RandomCutTree { private final Function<IBoundingBoxView, double[]> gVecBuild; public Function<IBoundingBoxView, double[]> getgVec() { return gVecBuild; } public static Builder builder() { return new Builder(); } protected HyperTree(HyperTree.Builder builder) { super(builder); this.gVecBuild = builder.gVec; } public void makeTree(List<Integer> list, int seed) { // this function allows a public call, which may be useful someday if (list.size() > 0 && list.size() < numberOfLeaves + 1) { int[] leftIndex = new int[numberOfLeaves - 1]; int[] rightIndex = new int[numberOfLeaves - 1]; Arrays.fill(leftIndex, numberOfLeaves - 1); Arrays.fill(rightIndex, numberOfLeaves - 1); int[] cutDimension = new int[numberOfLeaves - 1]; float[] cutValue = new float[numberOfLeaves - 1]; root = makeTreeInt(list, seed, 0, this.gVecBuild, leftIndex, rightIndex, cutDimension, cutValue); nodeStore = AbstractNodeStore.builder().dimension(dimension).capacity(numberOfLeaves - 1) .leftIndex(leftIndex).rightIndex(rightIndex).cutDimension(cutDimension).cutValues(cutValue).build(); // the cuts are specififed; now build tree for (int i = 0; i < list.size(); i++) { addPointToPartialTree(list.get(i), 0L); } } else { root = Null; } } private int makeTreeInt(List<Integer> pointList, int seed, int firstFree, Function<IBoundingBoxView, double[]> vecBuild, int[] left, int[] right, int[] cutDimension, float[] cutValue) { if (pointList.size() == 0) return Null; BoundingBox thisBox = new BoundingBox(pointStoreView.getNumericVector(pointList.get(0))); for (int i = 1; i < pointList.size(); i++) { thisBox = (BoundingBox) thisBox.getMergedBox(pointStoreView.getNumericVector(pointList.get(i))); } if (thisBox.getRangeSum() <= 0) { return pointList.get(0) + nodeStore.getCapacity() + 1; } Random ring = new Random(seed); int leftSeed = ring.nextInt(); int rightSeed = ring.nextInt(); Cut cut = getCut(thisBox, ring, vecBuild); List<Integer> leftList = new ArrayList<>(); List<Integer> rightList = new ArrayList<>(); for (int j = 0; j < pointList.size(); j++) { if (nodeStore.leftOf((float) cut.getValue(), cut.getDimension(), pointStoreView.getNumericVector(pointList.get(j)))) { leftList.add(pointList.get(j)); } else rightList.add(pointList.get(j)); } int leftIndex = makeTreeInt(leftList, leftSeed, firstFree + 1, vecBuild, left, right, cutDimension, cutValue); int rightIndex = makeTreeInt(rightList, rightSeed, firstFree + leftList.size(), vecBuild, left, right, cutDimension, cutValue); left[firstFree] = Math.min(leftIndex, numberOfLeaves - 1); right[firstFree] = Math.min(rightIndex, numberOfLeaves - 1); cutDimension[firstFree] = cut.getDimension(); cutValue[firstFree] = (float) cut.getValue(); return firstFree; } private Cut getCut(IBoundingBoxView bb, Random ring, Function<IBoundingBoxView, double[]> vecSeparation) { Random rng = new Random(ring.nextInt()); double cutf = rng.nextDouble(); double dimf = rng.nextDouble(); int td = -1; double rangeSum = 0; double[] vector = vecSeparation.apply(bb); for (int i = 0; i < bb.getDimensions(); i++) { vector[i] = (float) vector[i]; rangeSum += vector[i]; } double breakPoint = dimf * rangeSum; float cutValue = 0; for (int i = 0; i < bb.getDimensions(); i++) { double range = vector[i]; if (range > 0) { if ((breakPoint > 0) && (breakPoint <= range)) { td = i; cutValue = (float) (bb.getMinValue(td) + bb.getRange(td) * cutf); if (cutValue == bb.getMaxValue(td)) { cutValue = (float) bb.getMinValue(td); } } breakPoint -= range; } } checkArgument(td != -1, "Pivot selection failed."); return new Cut(td, cutValue); } public static class Builder extends RandomCutTree.Builder<Builder> { private Function<IBoundingBoxView, double[]> gVec; public Builder buildGVec(Function<IBoundingBoxView, double[]> gVec) { this.gVec = gVec; return this; } public HyperTree build() { return new HyperTree(this); } } }
455
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/NodeView.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkState; import java.util.HashMap; import com.amazon.randomcutforest.store.IPointStoreView; public class NodeView implements INodeView { public static double SWITCH_FRACTION = 0.499; RandomCutTree tree; int currentNodeOffset; float[] leafPoint; BoundingBox currentBox; public NodeView(RandomCutTree tree, IPointStoreView<float[]> pointStoreView, int root) { this.currentNodeOffset = root; this.tree = tree; } public int getMass() { return tree.getMass(currentNodeOffset); } public IBoundingBoxView getBoundingBox() { if (currentBox == null) { return tree.getBox(currentNodeOffset); } return currentBox; } public IBoundingBoxView getSiblingBoundingBox(float[] point) { return (toLeft(point)) ? tree.getBox(tree.nodeStore.getRightIndex(currentNodeOffset)) : tree.getBox(tree.nodeStore.getLeftIndex(currentNodeOffset)); } public int getCutDimension() { return tree.nodeStore.getCutDimension(currentNodeOffset); } @Override public double getCutValue() { return tree.nodeStore.getCutValue(currentNodeOffset); } public float[] getLeafPoint() { return leafPoint; } public HashMap<Long, Integer> getSequenceIndexes() { checkState(isLeaf(), "can only be invoked for a leaf"); if (tree.storeSequenceIndexesEnabled) { return tree.getSequenceMap(tree.getPointIndex(currentNodeOffset)); } else { return new HashMap<>(); } } @Override public double probailityOfSeparation(float[] point) { return tree.probabilityOfCut(currentNodeOffset, point, currentBox); } @Override public int getLeafPointIndex() { return tree.getPointIndex(currentNodeOffset); } public boolean isLeaf() { return tree.nodeStore.isLeaf(currentNodeOffset); } protected void setCurrentNode(int newNode, int index, boolean setBox) { currentNodeOffset = newNode; leafPoint = tree.pointStoreView.getNumericVector(index); if (setBox && tree.boundingBoxCacheFraction < SWITCH_FRACTION) { currentBox = new BoundingBox(leafPoint, leafPoint); } } protected void setCurrentNodeOnly(int newNode) { currentNodeOffset = newNode; } public void updateToParent(int parent, int currentSibling, boolean updateBox) { currentNodeOffset = parent; if (updateBox && tree.boundingBoxCacheFraction < SWITCH_FRACTION) { tree.growNodeBox(currentBox, tree.pointStoreView, parent, currentSibling); } } // this function exists for matching the behavior of RCF2.0 and will be replaced // this function explicitly uses the encoding of the new nodestore protected boolean toLeft(float[] point) { return point[tree.nodeStore.getCutDimension(currentNodeOffset)] <= tree.nodeStore .getCutValue(currentNodeOffset); } }
456
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/IBoundingBoxView.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; public interface IBoundingBoxView { double getRangeSum(); int getDimensions(); double getRange(int i); double getMinValue(int i); double getMaxValue(int i); // duplicates IBoundingBoxView copy(); // below keeps the older box unchanged IBoundingBoxView getMergedBox(float[] point); // merges and keeps the older box unchaged IBoundingBoxView getMergedBox(IBoundingBoxView otherBox); }
457
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/AbstractNodeStore.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import java.util.Stack; import com.amazon.randomcutforest.store.IndexIntervalManager; /** * A fixed-size buffer for storing interior tree nodes. An interior node is * defined by its location in the tree (parent and child nodes), its random cut, * and its bounding box. The NodeStore class uses arrays to store these field * values for a collection of nodes. An index in the store can be used to look * up the field values for a particular node. * * The internal nodes (handled by this store) corresponds to * [0..upperRangeLimit] * * If we think of an array of Node objects as being row-oriented (where each row * is a Node), then this class is analogous to a column-oriented database of * Nodes. * */ public abstract class AbstractNodeStore { public static int Null = -1; public static boolean DEFAULT_STORE_PARENT = false; /** * the number of internal nodes; the nodes will range from 0..capacity-1 the * value capacity would correspond to "not yet set" the values Y= capacity+1+X * correspond to pointstore index X note that capacity + 1 + X = * number_of_leaves + X */ protected final int capacity; protected final float[] cutValue; protected IndexIntervalManager freeNodeManager; public AbstractNodeStore(AbstractNodeStore.Builder<?> builder) { this.capacity = builder.capacity; if ((builder.leftIndex == null)) { freeNodeManager = new IndexIntervalManager(capacity); } cutValue = (builder.cutValues != null) ? builder.cutValues : new float[capacity]; } protected abstract int addNode(Stack<int[]> pathToRoot, float[] point, long sendex, int pointIndex, int childIndex, int childMassIfLeaf, int cutDimension, float cutValue, BoundingBox box); public boolean isLeaf(int index) { return index > capacity; } public boolean isInternal(int index) { return index < capacity && index >= 0; } public abstract void assignInPartialTree(int savedParent, float[] point, int childReference); public abstract int getLeftIndex(int index); public abstract int getRightIndex(int index); public abstract int getParentIndex(int index); public abstract void setRoot(int index); protected abstract void decreaseMassOfInternalNode(int node); protected abstract void increaseMassOfInternalNode(int node); protected void manageInternalNodesPartial(Stack<int[]> path) { while (!path.isEmpty()) { int index = path.pop()[0]; increaseMassOfInternalNode(index); } } public Stack<int[]> getPath(int root, float[] point, boolean verbose) { int node = root; Stack<int[]> answer = new Stack<>(); answer.push(new int[] { root, capacity }); while (isInternal(node)) { double y = getCutValue(node); if (leftOf(node, point)) { answer.push(new int[] { getLeftIndex(node), getRightIndex(node) }); node = getLeftIndex(node); } else { // this would push potential Null, of node == capacity // that would be used for tree reconstruction answer.push(new int[] { getRightIndex(node), getLeftIndex(node) }); node = getRightIndex(node); } } return answer; } public abstract void deleteInternalNode(int index); public abstract int getMass(int index); protected boolean leftOf(float cutValue, int cutDimension, float[] point) { return point[cutDimension] <= cutValue; } public boolean leftOf(int node, float[] point) { int cutDimension = getCutDimension(node); return leftOf(cutValue[node], cutDimension, point); } public int getSibling(int node, int parent) { int sibling = getLeftIndex(parent); if (node == sibling) { sibling = getRightIndex(parent); } return sibling; } public abstract void spliceEdge(int parent, int node, int newNode); public abstract void replaceParentBySibling(int grandParent, int parent, int node); public abstract int getCutDimension(int index); public double getCutValue(int index) { return cutValue[index]; } protected boolean toLeft(float[] point, int currentNodeOffset) { return point[getCutDimension(currentNodeOffset)] <= cutValue[currentNodeOffset]; } public abstract int[] getCutDimension(); public abstract int[] getRightIndex(); public abstract int[] getLeftIndex(); public float[] getCutValues() { return cutValue; } public int getCapacity() { return capacity; } public int size() { return capacity - freeNodeManager.size(); } /** * a builder */ public static class Builder<T extends Builder<T>> { protected int capacity; protected int[] leftIndex; protected int[] rightIndex; protected int[] cutDimension; protected float[] cutValues; protected boolean storeParent = DEFAULT_STORE_PARENT; protected int dimension; protected int root; // maximum number of points in the store public T capacity(int capacity) { this.capacity = capacity; return (T) this; } public T dimension(int dimension) { this.dimension = dimension; return (T) this; } public T useRoot(int root) { this.root = root; return (T) this; } public T leftIndex(int[] leftIndex) { this.leftIndex = leftIndex; return (T) this; } public T rightIndex(int[] rightIndex) { this.rightIndex = rightIndex; return (T) this; } public T cutDimension(int[] cutDimension) { this.cutDimension = cutDimension; return (T) this; } public T cutValues(float[] cutValues) { this.cutValues = cutValues; return (T) this; } public T storeParent(boolean storeParent) { this.storeParent = storeParent; return (T) this; } public AbstractNodeStore build() { if (leftIndex == null) { checkArgument(rightIndex == null, " incorrect option of right indices"); checkArgument(cutValues == null, "incorrect option of cut values"); checkArgument(cutDimension == null, " incorrect option of cut dimensions"); } else { checkArgument(rightIndex.length == capacity, " incorrect length of right indices"); checkArgument(cutValues.length == capacity, "incorrect length of cut values"); checkArgument(cutDimension.length == capacity, " incorrect length of cut dimensions"); } // capacity is numbner of internal nodes if (capacity < 256 && dimension <= 256) { return new NodeStoreSmall(this); } else if (capacity < Character.MAX_VALUE && dimension <= Character.MAX_VALUE) { return new NodeStoreMedium(this); } else { return new NodeStoreLarge(this); } } } public static Builder builder() { return new Builder(); } }
458
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/RandomCutTree.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.checkNotNull; import static com.amazon.randomcutforest.CommonUtils.checkState; import static com.amazon.randomcutforest.tree.AbstractNodeStore.DEFAULT_STORE_PARENT; import static com.amazon.randomcutforest.tree.AbstractNodeStore.Null; import static java.lang.Math.max; import java.util.ArrayList; import java.util.Arrays; import java.util.HashMap; import java.util.List; import java.util.Optional; import java.util.Random; import java.util.Stack; import com.amazon.randomcutforest.IMultiVisitorFactory; import com.amazon.randomcutforest.IVisitorFactory; import com.amazon.randomcutforest.MultiVisitor; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.Visitor; import com.amazon.randomcutforest.config.Config; import com.amazon.randomcutforest.store.IPointStoreView; /** * A Compact Random Cut Tree is a tree data structure whose leaves represent * points inserted into the tree and whose interior nodes represent regions of * space defined by Bounding Boxes and Cuts. New nodes and leaves are added to * the tree by making random cuts. * * The offsets are encoded as follows: an offset greater or equal maxSize * corresponds to a leaf node of offset (offset - maxSize) otherwise the offset * corresponds to an internal node * * The main use of this class is to be updated with points sampled from a * stream, and to define traversal methods. Users can then implement a * {@link Visitor} which can be submitted to a traversal method in order to * compute a statistic from the tree. */ public class RandomCutTree implements ITree<Integer, float[]> { /** * The index value used to represent the absence of a node. For example, when * the tree is created the root node index will be NULL. After a point is added * and a root node is created, the root node's parent will be NULL, and so on. */ private Random testRandom; protected boolean storeSequenceIndexesEnabled; protected boolean centerOfMassEnabled; private long randomSeed; protected int root; protected IPointStoreView<float[]> pointStoreView; protected int numberOfLeaves; protected AbstractNodeStore nodeStore; protected double boundingBoxCacheFraction; protected int outputAfter; protected int dimension; protected final HashMap<Integer, Integer> leafMass; protected double[] rangeSumData; protected float[] boundingBoxData; protected float[] pointSum; protected HashMap<Integer, List<Long>> sequenceMap; protected RandomCutTree(Builder<?> builder) { pointStoreView = builder.pointStoreView; numberOfLeaves = builder.capacity; randomSeed = builder.randomSeed; testRandom = builder.random; outputAfter = builder.outputAfter.orElse(max(1, numberOfLeaves / 4)); dimension = (builder.dimension != 0) ? builder.dimension : pointStoreView.getDimensions(); nodeStore = (builder.nodeStore != null) ? builder.nodeStore : AbstractNodeStore.builder().capacity(numberOfLeaves - 1).storeParent(builder.storeParent) .dimension(dimension).build(); this.boundingBoxCacheFraction = builder.boundingBoxCacheFraction; this.storeSequenceIndexesEnabled = builder.storeSequenceIndexesEnabled; this.centerOfMassEnabled = builder.centerOfMassEnabled; this.root = builder.root; leafMass = new HashMap<>(); int cache_limit = (int) Math.floor(boundingBoxCacheFraction * (numberOfLeaves - 1)); rangeSumData = new double[cache_limit]; boundingBoxData = new float[2 * dimension * cache_limit]; if (this.centerOfMassEnabled) { pointSum = new float[(numberOfLeaves - 1) * dimension]; } if (this.storeSequenceIndexesEnabled) { sequenceMap = new HashMap<>(); } } @Override public <T> void setConfig(String name, T value, Class<T> clazz) { if (Config.BOUNDING_BOX_CACHE_FRACTION.equals(name)) { checkArgument(Double.class.isAssignableFrom(clazz), () -> String.format("Setting '%s' must be a double value", name)); setBoundingBoxCacheFraction((Double) value); } else { throw new IllegalArgumentException("Unsupported configuration setting: " + name); } } @Override public <T> T getConfig(String name, Class<T> clazz) { checkNotNull(clazz, "clazz must not be null"); if (Config.BOUNDING_BOX_CACHE_FRACTION.equals(name)) { checkArgument(clazz.isAssignableFrom(Double.class), () -> String.format("Setting '%s' must be a double value", name)); return clazz.cast(boundingBoxCacheFraction); } else { throw new IllegalArgumentException("Unsupported configuration setting: " + name); } } // dynamically change the fraction of the new nodes which caches their bounding // boxes // 0 would mean less space usage, but slower throughput // 1 would imply larger space but better throughput public void setBoundingBoxCacheFraction(double fraction) { checkArgument(0 <= fraction && fraction <= 1, "incorrect parameter"); boundingBoxCacheFraction = fraction; resizeCache(fraction); } /** * Return a new {@link Cut}, which is chosen uniformly over the space of * possible cuts for a bounding box and its union with a point. The cut must * exist unless the union box is a single point. There are floating point issues * -- even though the original values are in float anf the calculations are in * double, which can show up with large number of dimensions (each trigerring an * addition/substraction). * * @param factor A random cut * @param point the point whose union is taken with the box * @param box A bounding box that we want to find a random cut for. * @return A new Cut corresponding to a random cut in the bounding box. */ protected Cut randomCut(double factor, float[] point, BoundingBox box) { double range = 0.0; for (int i = 0; i < point.length; i++) { float minValue = (float) box.getMinValue(i); float maxValue = (float) box.getMaxValue(i); if (point[i] < minValue) { minValue = point[i]; } else if (point[i] > maxValue) { maxValue = point[i]; } range += maxValue - minValue; } checkArgument(range > 0, () -> " the union is a single point " + Arrays.toString(point) + "or the box is inappropriate, box" + box.toString() + "factor =" + factor); double breakPoint = factor * range; for (int i = 0; i < box.getDimensions(); i++) { float minValue = (float) box.getMinValue(i); float maxValue = (float) box.getMaxValue(i); if (point[i] < minValue) { minValue = point[i]; } else if (point[i] > maxValue) { maxValue = point[i]; } double gap = maxValue - minValue; if (breakPoint <= gap && gap > 0) { float cutValue = (float) (minValue + breakPoint); // Random cuts have to take a value in the half-open interval [minValue, // maxValue) to ensure that a // Node has a valid left child and right child. if (cutValue >= maxValue) { cutValue = Math.nextAfter((float) maxValue, minValue); } return new Cut(i, cutValue); } breakPoint -= gap; } // if we are here then factor is likely almost 1 and we have floating point // issues // we will randomize between the first and the last non-zero ranges and choose // the // same cutValue as using nextAfter above -- we will use the factor as a seed // and // not be optimizing this sequel (either in execution or code) to ensure easier // debugging // this should be an anomaly - no pun intended. Random rng = new Random((long) (factor * Long.MAX_VALUE / 2)); if (rng.nextDouble() < 0.5) { for (int i = 0; i < box.getDimensions(); i++) { float minValue = (float) box.getMinValue(i); float maxValue = (float) box.getMaxValue(i); if (point[i] < minValue) { minValue = point[i]; } else if (point[i] > maxValue) { maxValue = point[i]; } if (maxValue > minValue) { double cutValue = Math.nextAfter((float) maxValue, minValue); return new Cut(i, cutValue); } } } else { for (int i = box.getDimensions() - 1; i >= 0; i--) { float minValue = (float) box.getMinValue(i); float maxValue = (float) box.getMaxValue(i); if (point[i] < minValue) { minValue = point[i]; } else if (point[i] > maxValue) { maxValue = point[i]; } if (maxValue > minValue) { double cutValue = Math.nextAfter((float) maxValue, minValue); return new Cut(i, cutValue); } } } throw new IllegalStateException("The break point did not lie inside the expected range; factor " + factor + ", point " + Arrays.toString(point) + " box " + box.toString()); } /** * the following function adds a point to the tree * * @param pointIndex the number corresponding to the point * @param sequenceIndex sequence index of the point * @return the value of the point index where the point was added; this is * pointIndex if there are no duplicates; otherwise it is the value of * the point being duplicated. */ public Integer addPoint(Integer pointIndex, long sequenceIndex) { if (root == Null) { root = convertToLeaf(pointIndex); addLeaf(pointIndex, sequenceIndex); return pointIndex; } else { float[] point = projectToTree(pointStoreView.getNumericVector(pointIndex)); checkArgument(point.length == dimension, () -> " mismatch in dimensions for " + pointIndex); Stack<int[]> pathToRoot = nodeStore.getPath(root, point, false); int[] first = pathToRoot.pop(); int leafNode = first[0]; int savedParent = (pathToRoot.size() == 0) ? Null : pathToRoot.lastElement()[0]; int leafSavedSibling = first[1]; int sibling = leafSavedSibling; int leafPointIndex = getPointIndex(leafNode); float[] oldPoint = projectToTree(pointStoreView.getNumericVector(leafPointIndex)); checkArgument(oldPoint.length == dimension, () -> " mismatch in dimensions for " + pointIndex); Stack<int[]> parentPath = new Stack<>(); if (Arrays.equals(point, oldPoint)) { increaseLeafMass(leafNode); manageAncestorsAdd(pathToRoot, point); addLeaf(leafPointIndex, sequenceIndex); return leafPointIndex; } else { int node = leafNode; int savedNode = node; int parent = savedParent; float savedCutValue = (float) 0.0; BoundingBox currentBox = new BoundingBox(oldPoint, oldPoint); BoundingBox savedBox = currentBox.copy(); int savedDim = Integer.MAX_VALUE; Random rng; if (testRandom == null) { rng = new Random(randomSeed); randomSeed = rng.nextLong(); } else { rng = testRandom; } while (true) { double factor = rng.nextDouble(); Cut cut = randomCut(factor, point, currentBox); int dim = cut.getDimension(); float value = (float) cut.getValue(); boolean separation = ((point[dim] <= value && value < currentBox.getMinValue(dim) || point[dim] > value && value >= currentBox.getMaxValue(dim))); if (separation) { savedCutValue = value; savedDim = dim; savedParent = parent; savedNode = node; savedBox = currentBox.copy(); parentPath.clear(); } else { parentPath.push(new int[] { node, sibling }); } if (currentBox.contains(point) || parent == Null) { break; } else { growNodeBox(currentBox, pointStoreView, parent, sibling); int[] next = pathToRoot.pop(); node = next[0]; sibling = next[1]; if (pathToRoot.size() != 0) { parent = pathToRoot.lastElement()[0]; } else { parent = Null; } } } if (savedParent != Null) { while (!parentPath.isEmpty()) { pathToRoot.push(parentPath.pop()); } } int childMassIfLeaf = isLeaf(savedNode) ? getLeafMass(savedNode) : 0; int mergedNode = nodeStore.addNode(pathToRoot, point, sequenceIndex, pointIndex, savedNode, childMassIfLeaf, savedDim, savedCutValue, savedBox); addLeaf(pointIndex, sequenceIndex); addBox(mergedNode, point, savedBox); manageAncestorsAdd(pathToRoot, point); if (pointSum != null) { recomputePointSum(mergedNode); } if (savedParent == Null) { root = mergedNode; } } return pointIndex; } } protected void manageAncestorsAdd(Stack<int[]> path, float[] point) { while (!path.isEmpty()) { int index = path.pop()[0]; nodeStore.increaseMassOfInternalNode(index); if (pointSum != null) { recomputePointSum(index); } if (boundingBoxCacheFraction > 0.0) { checkContainsAndRebuildBox(index, point, pointStoreView); addPointInPlace(index, point); } } } /** * the following is the same as in addPoint() except this function is used to * rebuild the tree structure. This function does not create auxiliary arrays, * which should be performed using validateAndReconstruct() * * @param pointIndex index of point (in point store) * @param sequenceIndex sequence index (stored in sampler) */ public void addPointToPartialTree(Integer pointIndex, long sequenceIndex) { checkArgument(root != Null, " a null root is not a partial tree"); float[] point = projectToTree(pointStoreView.getNumericVector(pointIndex)); checkArgument(point.length == dimension, () -> " incorrect projection at index " + pointIndex); Stack<int[]> pathToRoot = nodeStore.getPath(root, point, false); int[] first = pathToRoot.pop(); int leafNode = first[0]; int savedParent = (pathToRoot.size() == 0) ? Null : pathToRoot.lastElement()[0]; if (!isLeaf(leafNode)) { if (savedParent == Null) { root = convertToLeaf(pointIndex); } else { nodeStore.assignInPartialTree(savedParent, point, convertToLeaf(pointIndex)); nodeStore.manageInternalNodesPartial(pathToRoot); addLeaf(pointIndex, sequenceIndex); } return; } int leafPointIndex = getPointIndex(leafNode); float[] oldPoint = projectToTree(pointStoreView.getNumericVector(leafPointIndex)); checkArgument(oldPoint.length == dimension && Arrays.equals(point, oldPoint), () -> "incorrect state on adding " + pointIndex); increaseLeafMass(leafNode); nodeStore.manageInternalNodesPartial(pathToRoot); addLeaf(leafPointIndex, sequenceIndex); return; } public Integer deletePoint(Integer pointIndex, long sequenceIndex) { checkArgument(root != Null, " deleting from an empty tree"); float[] point = projectToTree(pointStoreView.getNumericVector(pointIndex)); checkArgument(point.length == dimension, () -> " incorrect projection at index " + pointIndex); Stack<int[]> pathToRoot = nodeStore.getPath(root, point, false); int[] first = pathToRoot.pop(); int leafSavedSibling = first[1]; int leafNode = first[0]; int leafPointIndex = getPointIndex(leafNode); checkArgument(leafPointIndex == pointIndex, () -> " deleting wrong node " + leafPointIndex + " instead of " + pointIndex); removeLeaf(leafPointIndex, sequenceIndex); if (decreaseLeafMass(leafNode) == 0) { if (pathToRoot.size() == 0) { root = Null; } else { int parent = pathToRoot.pop()[0]; if (pathToRoot.size() == 0) { root = leafSavedSibling; } else { int grandParent = pathToRoot.lastElement()[0]; nodeStore.replaceParentBySibling(grandParent, parent, leafNode); manageAncestorsDelete(pathToRoot, point); } nodeStore.deleteInternalNode(parent); if (pointSum != null) { invalidatePointSum(parent); } int idx = translate(parent); if (idx != Integer.MAX_VALUE) { rangeSumData[idx] = 0.0; } } } else { manageAncestorsDelete(pathToRoot, point); } return leafPointIndex; } protected void manageAncestorsDelete(Stack<int[]> path, float[] point) { boolean resolved = false; while (!path.isEmpty()) { int index = path.pop()[0]; nodeStore.decreaseMassOfInternalNode(index); if (pointSum != null) { recomputePointSum(index); } if (boundingBoxCacheFraction > 0.0 && !resolved) { resolved = checkContainsAndRebuildBox(index, point, pointStoreView); } } } //// leaf, nonleaf representations public boolean isLeaf(int index) { // note that numberOfLeaves - 1 corresponds to an unspefied leaf in partial tree // 0 .. numberOfLeaves - 2 corresponds to internal nodes return index >= numberOfLeaves; } public boolean isInternal(int index) { // note that numberOfLeaves - 1 corresponds to an unspefied leaf in partial tree // 0 .. numberOfLeaves - 2 corresponds to internal nodes return index < numberOfLeaves - 1 && index >= 0; } public int convertToLeaf(int pointIndex) { return pointIndex + numberOfLeaves; } public int getPointIndex(int index) { checkArgument(index >= numberOfLeaves, () -> " does not have a point associated " + index); return index - numberOfLeaves; } public int getLeftChild(int index) { checkArgument(isInternal(index), () -> "incorrect call to get left Index " + index); return nodeStore.getLeftIndex(index); } public int getRightChild(int index) { checkArgument(isInternal(index), () -> "incorrect call to get right child " + index); return nodeStore.getRightIndex(index); } public int getCutDimension(int index) { checkArgument(isInternal(index), () -> "incorrect call to get cut dimension " + index); return nodeStore.getCutDimension(index); } public double getCutValue(int index) { checkArgument(isInternal(index), () -> "incorrect call to get cut value " + index); return nodeStore.getCutValue(index); } ///// mass assignments; separating leafs and internal nodes protected int getMass(int index) { return (isLeaf(index)) ? getLeafMass(index) : nodeStore.getMass(index); } protected int getLeafMass(int index) { int y = (index - numberOfLeaves); Integer value = leafMass.get(y); return (value != null) ? value + 1 : 1; } protected void increaseLeafMass(int index) { int y = (index - numberOfLeaves); leafMass.merge(y, 1, Integer::sum); } protected int decreaseLeafMass(int index) { int y = (index - numberOfLeaves); Integer value = leafMass.remove(y); if (value != null) { if (value > 1) { leafMass.put(y, (value - 1)); return value; } else { return 1; } } else { return 0; } } @Override public int getMass() { return root == Null ? 0 : isLeaf(root) ? getLeafMass(root) : nodeStore.getMass(root); } /////// Bounding box public void resizeCache(double fraction) { if (fraction == 0) { rangeSumData = null; boundingBoxData = null; } else { int limit = (int) Math.floor(fraction * (numberOfLeaves - 1)); rangeSumData = (rangeSumData == null) ? new double[limit] : Arrays.copyOf(rangeSumData, limit); boundingBoxData = (boundingBoxData == null) ? new float[limit * 2 * dimension] : Arrays.copyOf(boundingBoxData, limit * 2 * dimension); } boundingBoxCacheFraction = fraction; } protected int translate(int index) { if (rangeSumData == null || rangeSumData.length <= index) { return Integer.MAX_VALUE; } else { return index; } } void copyBoxToData(int idx, BoundingBox box) { int base = 2 * idx * dimension; int mid = base + dimension; System.arraycopy(box.getMinValues(), 0, boundingBoxData, base, dimension); System.arraycopy(box.getMaxValues(), 0, boundingBoxData, mid, dimension); rangeSumData[idx] = box.getRangeSum(); } void addPointInPlace(int index, float[] point) { int idx = translate(index); if (idx != Integer.MAX_VALUE) { int base = 2 * idx * dimension; int mid = base + dimension; double rangeSum = 0; for (int i = 0; i < dimension; i++) { boundingBoxData[base + i] = Math.min(boundingBoxData[base + i], point[i]); } for (int i = 0; i < dimension; i++) { boundingBoxData[mid + i] = max(boundingBoxData[mid + i], point[i]); } for (int i = 0; i < dimension; i++) { rangeSum += boundingBoxData[mid + i] - boundingBoxData[base + i]; } rangeSumData[idx] = rangeSum; } } public BoundingBox getBox(int index) { if (isLeaf(index)) { float[] point = projectToTree(pointStoreView.getNumericVector(getPointIndex(index))); checkArgument(point.length == dimension, () -> "failure in projection at index " + index); return new BoundingBox(point, point); } else { checkState(isInternal(index), " incomplete state"); int idx = translate(index); if (idx != Integer.MAX_VALUE) { if (rangeSumData[idx] != 0) { // return non-trivial boxes return getBoxFromData(idx); } else { BoundingBox box = reconstructBox(index, pointStoreView); copyBoxToData(idx, box); return box; } } return reconstructBox(index, pointStoreView); } } BoundingBox reconstructBox(int index, IPointStoreView<float[]> pointStoreView) { BoundingBox mutatedBoundingBox = getBox(nodeStore.getLeftIndex(index)); growNodeBox(mutatedBoundingBox, pointStoreView, index, nodeStore.getRightIndex(index)); return mutatedBoundingBox; } boolean checkStrictlyContains(int index, float[] point) { int idx = translate(index); if (idx != Integer.MAX_VALUE) { int base = 2 * idx * dimension; int mid = base + dimension; boolean isInside = true; for (int i = 0; i < dimension && isInside; i++) { if (point[i] >= boundingBoxData[mid + i] || boundingBoxData[base + i] >= point[i]) { isInside = false; } } return isInside; } return false; } boolean checkContainsAndRebuildBox(int index, float[] point, IPointStoreView<float[]> pointStoreView) { int idx = translate(index); if (idx != Integer.MAX_VALUE) { if (!checkStrictlyContains(index, point)) { BoundingBox mutatedBoundingBox = reconstructBox(index, pointStoreView); copyBoxToData(idx, mutatedBoundingBox); return false; } return true; } return false; } BoundingBox getBoxFromData(int idx) { int base = 2 * idx * dimension; int mid = base + dimension; return new BoundingBox(Arrays.copyOfRange(boundingBoxData, base, base + dimension), Arrays.copyOfRange(boundingBoxData, mid, mid + dimension)); } void addBox(int index, float[] point, BoundingBox box) { int idx = translate(index); if (idx != Integer.MAX_VALUE) { // always add irrespective of rangesum copyBoxToData(idx, box); addPointInPlace(index, point); } } void growNodeBox(BoundingBox box, IPointStoreView<float[]> pointStoreView, int node, int sibling) { if (isLeaf(sibling)) { float[] point = projectToTree(pointStoreView.getNumericVector(getPointIndex(sibling))); checkArgument(point.length == dimension, () -> " incorrect projection at index " + sibling); box.addPoint(point); } else { if (!isInternal(sibling)) { throw new IllegalStateException(" incomplete state " + sibling); } int siblingIdx = translate(sibling); if (siblingIdx != Integer.MAX_VALUE) { if (rangeSumData[siblingIdx] != 0) { box.addBox(getBoxFromData(siblingIdx)); } else { BoundingBox newBox = getBox(siblingIdx); copyBoxToData(siblingIdx, newBox); box.addBox(newBox); } return; } growNodeBox(box, pointStoreView, sibling, nodeStore.getLeftIndex(sibling)); growNodeBox(box, pointStoreView, sibling, nodeStore.getRightIndex(sibling)); return; } } public double probabilityOfCut(int node, float[] point, BoundingBox otherBox) { int nodeIdx = translate(node); if (nodeIdx != Integer.MAX_VALUE && rangeSumData[nodeIdx] != 0) { int base = 2 * nodeIdx * dimension; int mid = base + dimension; double minsum = 0; double maxsum = 0; for (int i = 0; i < dimension; i++) { minsum += max(boundingBoxData[base + i] - point[i], 0); } for (int i = 0; i < dimension; i++) { maxsum += max(point[i] - boundingBoxData[mid + i], 0); } double sum = maxsum + minsum; if (sum == 0.0) { return 0.0; } return sum / (rangeSumData[nodeIdx] + sum); } else if (otherBox != null) { return otherBox.probabilityOfCut(point); } else { BoundingBox box = getBox(node); return box.probabilityOfCut(point); } } /// additional information at nodes public float[] getPointSum(int index) { checkArgument(centerOfMassEnabled, " enable center of mass"); if (isLeaf(index)) { float[] point = projectToTree(pointStoreView.getNumericVector(getPointIndex(index))); checkArgument(point.length == dimension, () -> " incorrect projection"); int mass = getMass(index); for (int i = 0; i < point.length; i++) { point[i] *= mass; } return point; } else { return Arrays.copyOfRange(pointSum, index * dimension, (index + 1) * dimension); } } public void invalidatePointSum(int index) { for (int i = 0; i < dimension; i++) { pointSum[index * dimension + i] = 0; } } public void recomputePointSum(int index) { float[] left = getPointSum(nodeStore.getLeftIndex(index)); float[] right = getPointSum(nodeStore.getRightIndex(index)); for (int i = 0; i < dimension; i++) { pointSum[index * dimension + i] = left[i] + right[i]; } } public HashMap<Long, Integer> getSequenceMap(int index) { HashMap<Long, Integer> hashMap = new HashMap<>(); List<Long> list = getSequenceList(index); for (Long e : list) { hashMap.merge(e, 1, Integer::sum); } return hashMap; } public List<Long> getSequenceList(int index) { return sequenceMap.get(index); } protected void addLeaf(int pointIndex, long sequenceIndex) { if (storeSequenceIndexesEnabled) { List<Long> leafList = sequenceMap.remove(pointIndex); if (leafList == null) { leafList = new ArrayList<>(1); } leafList.add(sequenceIndex); sequenceMap.put(pointIndex, leafList); } } public void removeLeaf(int leafPointIndex, long sequenceIndex) { if (storeSequenceIndexesEnabled) { List<Long> leafList = sequenceMap.remove(leafPointIndex); checkArgument(leafList != null, " leaf index not found in tree"); checkArgument(leafList.remove(sequenceIndex), " sequence index not found in leaf"); if (!leafList.isEmpty()) { sequenceMap.put(leafPointIndex, leafList); } } } //// validations public void validateAndReconstruct() { if (root != Null) { validateAndReconstruct(root); } } /** * This function is supposed to validate the integrity of the tree and rebuild * internal data structures. At this moment the only internal structure is the * pointsum. * * @param index the node of a tree * @return a bounding box of the points */ public BoundingBox validateAndReconstruct(int index) { if (isLeaf(index)) { return getBox(index); } else { checkState(isInternal(index), "illegal state"); BoundingBox leftBox = validateAndReconstruct(getLeftChild(index)); BoundingBox rightBox = validateAndReconstruct(getRightChild(index)); if (leftBox.maxValues[getCutDimension(index)] > getCutValue(index) || rightBox.minValues[getCutDimension(index)] <= getCutValue(index)) { throw new IllegalStateException(" incorrect bounding state at index " + index + " cut value " + getCutValue(index) + "cut dimension " + getCutDimension(index) + " left Box " + leftBox.toString() + " right box " + rightBox.toString()); } if (centerOfMassEnabled) { recomputePointSum(index); } rightBox.addBox(leftBox); int idx = translate(index); if (idx != Integer.MAX_VALUE) { // always add irrespective of rangesum copyBoxToData(idx, rightBox); } return rightBox; } } //// traversals /** * Starting from the root, traverse the canonical path to a leaf node and visit * the nodes along the path. The canonical path is determined by the input * point: at each interior node, we select the child node by comparing the * node's {@link Cut} to the corresponding coordinate value in the input point. * The method recursively traverses to the leaf node first and then invokes the * visitor on each node in reverse order. That is, if the path to the leaf node * determined by the input point is root, node1, node2, ..., node(N-1), nodeN, * leaf; then we will first invoke visitor::acceptLeaf on the leaf node, and * then we will invoke visitor::accept on the remaining nodes in the following * order: nodeN, node(N-1), ..., node2, node1, and root. * * @param point A point which determines the traversal path from the * root to a leaf node. * @param visitorFactory A visitor that will be invoked for each node on the * path. * @param <R> The return type of the Visitor. * @return the value of {@link Visitor#getResult()}} after the traversal. */ @Override public <R> R traverse(float[] point, IVisitorFactory<R> visitorFactory) { checkArgument(root != Null, "this tree doesn't contain any nodes"); checkNotNull(point, "point must not be null"); checkNotNull(visitorFactory, "visitor must not be null"); Visitor<R> visitor = visitorFactory.newVisitor(this, point); NodeView currentNodeView = new NodeView(this, pointStoreView, root); traversePathToLeafAndVisitNodes(point, visitor, currentNodeView, root, 0); return visitorFactory.liftResult(this, visitor.getResult()); } protected <R> void traversePathToLeafAndVisitNodes(float[] point, Visitor<R> visitor, NodeView currentNodeView, int node, int depthOfNode) { if (isLeaf(node)) { currentNodeView.setCurrentNode(node, getPointIndex(node), true); visitor.acceptLeaf(currentNodeView, depthOfNode); } else { checkState(isInternal(node), " incomplete state "); if (nodeStore.toLeft(point, node)) { traversePathToLeafAndVisitNodes(point, visitor, currentNodeView, nodeStore.getLeftIndex(node), depthOfNode + 1); currentNodeView.updateToParent(node, nodeStore.getRightIndex(node), !visitor.isConverged()); } else { traversePathToLeafAndVisitNodes(point, visitor, currentNodeView, nodeStore.getRightIndex(node), depthOfNode + 1); currentNodeView.updateToParent(node, nodeStore.getLeftIndex(node), !visitor.isConverged()); } visitor.accept(currentNodeView, depthOfNode); } } /** * This is a traversal method which follows the standard traversal path (defined * in {@link #traverse(float[], IVisitorFactory)}) but at Node in checks to see * whether the visitor should split. If a split is triggered, then independent * copies of the visitor are sent down each branch of the tree and then merged * before propagating the result. * * @param point A point which determines the traversal path from the * root to a leaf node. * @param visitorFactory A visitor that will be invoked for each node on the * path. * @param <R> The return type of the Visitor. * @return the value of {@link Visitor#getResult()}} after the traversal. */ @Override public <R> R traverseMulti(float[] point, IMultiVisitorFactory<R> visitorFactory) { checkArgument(root != Null, "this tree doesn't contain any nodes"); checkNotNull(point, "point must not be null"); checkNotNull(visitorFactory, "visitor must not be null"); MultiVisitor<R> visitor = visitorFactory.newVisitor(this, point); NodeView currentNodeView = new NodeView(this, pointStoreView, root); traverseTreeMulti(point, visitor, currentNodeView, root, 0); return visitorFactory.liftResult(this, visitor.getResult()); } protected <R> void traverseTreeMulti(float[] point, MultiVisitor<R> visitor, NodeView currentNodeView, int node, int depthOfNode) { if (isLeaf(node)) { currentNodeView.setCurrentNode(node, getPointIndex(node), false); visitor.acceptLeaf(currentNodeView, depthOfNode); } else { checkState(isInternal(node), " incomplete state"); currentNodeView.setCurrentNodeOnly(node); if (visitor.trigger(currentNodeView)) { traverseTreeMulti(point, visitor, currentNodeView, nodeStore.getLeftIndex(node), depthOfNode + 1); MultiVisitor<R> newVisitor = visitor.newCopy(); currentNodeView.setCurrentNodeOnly(nodeStore.getRightIndex(node)); traverseTreeMulti(point, newVisitor, currentNodeView, nodeStore.getRightIndex(node), depthOfNode + 1); currentNodeView.updateToParent(node, nodeStore.getLeftIndex(node), false); visitor.combine(newVisitor); } else if (nodeStore.toLeft(point, node)) { traverseTreeMulti(point, visitor, currentNodeView, nodeStore.getLeftIndex(node), depthOfNode + 1); currentNodeView.updateToParent(node, nodeStore.getRightIndex(node), false); } else { traverseTreeMulti(point, visitor, currentNodeView, nodeStore.getRightIndex(node), depthOfNode + 1); currentNodeView.updateToParent(node, nodeStore.getLeftIndex(node), false); } visitor.accept(currentNodeView, depthOfNode); } } public int getNumberOfLeaves() { return numberOfLeaves; } public boolean isCenterOfMassEnabled() { return centerOfMassEnabled; } public boolean isStoreSequenceIndexesEnabled() { return storeSequenceIndexesEnabled; } public double getBoundingBoxCacheFraction() { return boundingBoxCacheFraction; } public int getDimension() { return dimension; } /** * * @return the root of the tree */ public Integer getRoot() { return (int) root; } /** * returns the number of samples that needs to be seen before returning * meaningful inference */ public int getOutputAfter() { return outputAfter; } @Override public boolean isOutputReady() { return getMass() >= outputAfter; } public float[] projectToTree(float[] point) { return Arrays.copyOf(point, point.length); } public float[] liftFromTree(float[] result) { return Arrays.copyOf(result, result.length); } public double[] liftFromTree(double[] result) { return Arrays.copyOf(result, result.length); } public int[] projectMissingIndices(int[] list) { return Arrays.copyOf(list, list.length); } public long getRandomSeed() { return randomSeed; } public AbstractNodeStore getNodeStore() { return nodeStore; } public static class Builder<T extends Builder<T>> { protected boolean storeSequenceIndexesEnabled = RandomCutForest.DEFAULT_STORE_SEQUENCE_INDEXES_ENABLED; protected boolean centerOfMassEnabled = RandomCutForest.DEFAULT_CENTER_OF_MASS_ENABLED; protected double boundingBoxCacheFraction = RandomCutForest.DEFAULT_BOUNDING_BOX_CACHE_FRACTION; protected long randomSeed = new Random().nextLong(); protected Random random = null; protected int capacity = RandomCutForest.DEFAULT_SAMPLE_SIZE; protected Optional<Integer> outputAfter = Optional.empty(); protected int dimension; protected IPointStoreView<float[]> pointStoreView; protected AbstractNodeStore nodeStore; protected int root = Null; protected boolean storeParent = DEFAULT_STORE_PARENT; public T capacity(int capacity) { this.capacity = capacity; return (T) this; } public T boundingBoxCacheFraction(double boundingBoxCacheFraction) { this.boundingBoxCacheFraction = boundingBoxCacheFraction; return (T) this; } public T pointStoreView(IPointStoreView<float[]> pointStoreView) { this.pointStoreView = pointStoreView; return (T) this; } public T nodeStore(AbstractNodeStore nodeStore) { this.nodeStore = nodeStore; return (T) this; } public T randomSeed(long randomSeed) { this.randomSeed = randomSeed; return (T) this; } public T random(Random random) { this.random = random; return (T) this; } public T outputAfter(int outputAfter) { this.outputAfter = Optional.of(outputAfter); return (T) this; } public T dimension(int dimension) { this.dimension = dimension; return (T) this; } public T setRoot(int root) { this.root = root; return (T) this; } public T storeParent(boolean storeParent) { this.storeParent = storeParent; return (T) this; } public T storeSequenceIndexesEnabled(boolean storeSequenceIndexesEnabled) { this.storeSequenceIndexesEnabled = storeSequenceIndexesEnabled; return (T) this; } public T centerOfMassEnabled(boolean centerOfMassEnabled) { this.centerOfMassEnabled = centerOfMassEnabled; return (T) this; } public RandomCutTree build() { return new RandomCutTree(this); } } public static Builder builder() { return new Builder(); } }
459
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/NodeStoreLarge.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import java.util.Arrays; import java.util.BitSet; import java.util.Stack; import com.amazon.randomcutforest.store.IndexIntervalManager; /** * A fixed-size buffer for storing interior tree nodes. An interior node is * defined by its location in the tree (parent and child nodes), its random cut, * and its bounding box. The NodeStore class uses arrays to store these field * values for a collection of nodes. An index in the store can be used to look * up the field values for a particular node. * * The internal nodes (handled by this store) corresponds to * [0..upperRangeLimit] * * If we think of an array of Node objects as being row-oriented (where each row * is a Node), then this class is analogous to a column-oriented database of * Nodes. * */ public class NodeStoreLarge extends AbstractNodeStore { private final int[] parentIndex; private final int[] leftIndex; private final int[] rightIndex; public final int[] cutDimension; private final int[] mass; public NodeStoreLarge(AbstractNodeStore.Builder builder) { super(builder); mass = new int[capacity]; Arrays.fill(mass, 0); if (builder.storeParent) { parentIndex = new int[capacity]; Arrays.fill(parentIndex, capacity); } else { parentIndex = null; } if (builder.leftIndex == null) { leftIndex = new int[capacity]; rightIndex = new int[capacity]; cutDimension = new int[capacity]; Arrays.fill(leftIndex, capacity); Arrays.fill(rightIndex, capacity); } else { leftIndex = Arrays.copyOf(builder.leftIndex, builder.leftIndex.length); rightIndex = Arrays.copyOf(builder.rightIndex, builder.rightIndex.length); cutDimension = Arrays.copyOf(builder.cutDimension, builder.cutDimension.length); BitSet bits = new BitSet(capacity); if (builder.root != Null) { bits.set(builder.root); } for (int i = 0; i < leftIndex.length; i++) { if (isInternal(leftIndex[i])) { bits.set(leftIndex[i]); if (parentIndex != null) { parentIndex[leftIndex[i]] = i; } } } for (int i = 0; i < rightIndex.length; i++) { if (isInternal(rightIndex[i])) { bits.set(rightIndex[i]); if (parentIndex != null) { parentIndex[rightIndex[i]] = i; } } } freeNodeManager = new IndexIntervalManager(capacity, capacity, bits); } } @Override public int addNode(Stack<int[]> pathToRoot, float[] point, long sequenceIndex, int pointIndex, int childIndex, int childMassIfLeaf, int cutDimension, float cutValue, BoundingBox box) { int index = freeNodeManager.takeIndex(); this.cutValue[index] = cutValue; this.cutDimension[index] = (byte) cutDimension; if (leftOf(cutValue, cutDimension, point)) { this.leftIndex[index] = (pointIndex + capacity + 1); this.rightIndex[index] = childIndex; } else { this.rightIndex[index] = (pointIndex + capacity + 1); this.leftIndex[index] = childIndex; } this.mass[index] = (((childMassIfLeaf > 0) ? childMassIfLeaf : getMass(childIndex)) + 1) % (capacity + 1); int parentIndex = (pathToRoot.size() == 0) ? Null : pathToRoot.lastElement()[0]; if (this.parentIndex != null) { this.parentIndex[index] = parentIndex; if (!isLeaf(childIndex)) { this.parentIndex[childIndex] = (index); } } if (parentIndex != Null) { spliceEdge(parentIndex, childIndex, index); } return index; } public int getLeftIndex(int index) { return leftIndex[index]; } public int getRightIndex(int index) { return rightIndex[index]; } public void setRoot(int index) { if (!isLeaf(index) && parentIndex != null) { parentIndex[index] = capacity; } } @Override protected void decreaseMassOfInternalNode(int node) { mass[node] = (mass[node] + capacity) % (capacity + 1); } @Override protected void increaseMassOfInternalNode(int node) { mass[node] = (mass[node] + 1) % (capacity + 1); } public void deleteInternalNode(int index) { leftIndex[index] = capacity; rightIndex[index] = capacity; if (parentIndex != null) { parentIndex[index] = capacity; } freeNodeManager.releaseIndex(index); } public int getMass(int index) { return mass[index] != 0 ? mass[index] : (capacity + 1); } @Override public void assignInPartialTree(int node, float[] point, int childReference) { if (leftOf(node, point)) { leftIndex[node] = childReference; } else { rightIndex[node] = childReference; } } public void spliceEdge(int parent, int node, int newNode) { assert (!isLeaf(newNode)); if (node == leftIndex[parent]) { leftIndex[parent] = newNode; } else { rightIndex[parent] = newNode; } if (parentIndex != null && isInternal(node)) { parentIndex[node] = newNode; } } public void replaceParentBySibling(int grandParent, int parent, int node) { int sibling = getSibling(node, parent); if (parent == leftIndex[grandParent]) { leftIndex[grandParent] = sibling; } else { rightIndex[grandParent] = sibling; } if (parentIndex != null && isInternal(sibling)) { parentIndex[sibling] = grandParent; } } public int getCutDimension(int index) { return cutDimension[index]; } public int[] getCutDimension() { return Arrays.copyOf(cutDimension, cutDimension.length); } public int[] getLeftIndex() { return Arrays.copyOf(leftIndex, leftIndex.length); } public int[] getRightIndex() { return Arrays.copyOf(rightIndex, rightIndex.length); } public int getParentIndex(int index) { checkArgument(parentIndex != null, "incorrect call"); return parentIndex[index]; } }
460
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/tree/NodeStoreSmall.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.tree; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.toByteArray; import static com.amazon.randomcutforest.CommonUtils.toCharArray; import static com.amazon.randomcutforest.CommonUtils.toIntArray; import java.util.Arrays; import java.util.BitSet; import java.util.Stack; import com.amazon.randomcutforest.store.IndexIntervalManager; /** * A fixed-size buffer for storing interior tree nodes. An interior node is * defined by its location in the tree (parent and child nodes), its random cut, * and its bounding box. The NodeStore class uses arrays to store these field * values for a collection of nodes. An index in the store can be used to look * up the field values for a particular node. * * The internal nodes (handled by this store) corresponds to * [0..upperRangeLimit] * * If we think of an array of Node objects as being row-oriented (where each row * is a Node), then this class is analogous to a column-oriented database of * Nodes. */ public class NodeStoreSmall extends AbstractNodeStore { private final byte[] parentIndex; private final char[] leftIndex; private final char[] rightIndex; public final byte[] cutDimension; private final byte[] mass; public NodeStoreSmall(AbstractNodeStore.Builder builder) { super(builder); mass = new byte[capacity]; Arrays.fill(mass, (byte) 0); if (builder.storeParent) { parentIndex = new byte[capacity]; Arrays.fill(parentIndex, (byte) capacity); } else { parentIndex = null; } if (builder.leftIndex == null) { leftIndex = new char[capacity]; rightIndex = new char[capacity]; cutDimension = new byte[capacity]; Arrays.fill(leftIndex, (char) capacity); Arrays.fill(rightIndex, (char) capacity); } else { checkArgument(builder.leftIndex.length == capacity, " incorrect length"); checkArgument(builder.rightIndex.length == capacity, " incorrect length"); leftIndex = toCharArray(builder.leftIndex); rightIndex = toCharArray(builder.rightIndex); cutDimension = toByteArray(builder.cutDimension); BitSet bits = new BitSet(capacity); if (builder.root != Null) { bits.set(builder.root); } for (int i = 0; i < leftIndex.length; i++) { if (isInternal(leftIndex[i])) { bits.set(leftIndex[i]); if (parentIndex != null) { parentIndex[leftIndex[i]] = (byte) i; } } } for (int i = 0; i < rightIndex.length; i++) { if (isInternal(rightIndex[i])) { bits.set(rightIndex[i]); if (parentIndex != null) { parentIndex[rightIndex[i]] = (byte) i; } } } freeNodeManager = new IndexIntervalManager(capacity, capacity, bits); // need to set up parents using the root } } @Override public int addNode(Stack<int[]> pathToRoot, float[] point, long sequenceIndex, int pointIndex, int childIndex, int childMassIfLeaf, int cutDimension, float cutValue, BoundingBox box) { int index = freeNodeManager.takeIndex(); this.cutValue[index] = cutValue; this.cutDimension[index] = (byte) cutDimension; if (leftOf(cutValue, cutDimension, point)) { this.leftIndex[index] = (char) (pointIndex + capacity + 1); this.rightIndex[index] = (char) childIndex; } else { this.rightIndex[index] = (char) (pointIndex + capacity + 1); this.leftIndex[index] = (char) childIndex; } this.mass[index] = (byte) ((((childMassIfLeaf > 0) ? childMassIfLeaf : getMass(childIndex)) + 1) % (capacity + 1)); int parentIndex = (pathToRoot.size() == 0) ? Null : pathToRoot.lastElement()[0]; if (this.parentIndex != null) { this.parentIndex[index] = (byte) parentIndex; if (!isLeaf(childIndex)) { this.parentIndex[childIndex] = (byte) (index); } } if (parentIndex != Null) { spliceEdge(parentIndex, childIndex, index); } return index; } @Override public void assignInPartialTree(int node, float[] point, int childReference) { if (leftOf(node, point)) { leftIndex[node] = (char) childReference; } else { rightIndex[node] = (char) childReference; } } public int getLeftIndex(int index) { return leftIndex[index]; } public int getRightIndex(int index) { return rightIndex[index]; } public int getParentIndex(int index) { checkArgument(parentIndex != null, "incorrect call"); return parentIndex[index]; } public void setRoot(int index) { if (!isLeaf(index) && parentIndex != null) { parentIndex[index] = (byte) capacity; } } @Override protected void decreaseMassOfInternalNode(int node) { mass[node] = (byte) (((mass[node] & 0xff) + capacity) % (capacity + 1)); // this cannot get to 0 } @Override protected void increaseMassOfInternalNode(int node) { mass[node] = (byte) (((mass[node] & 0xff) + 1) % (capacity + 1)); // mass of root == 0; note capacity = number_of_leaves - 1 } public void deleteInternalNode(int index) { leftIndex[index] = (char) capacity; rightIndex[index] = (char) capacity; if (parentIndex != null) { parentIndex[index] = (byte) capacity; } freeNodeManager.releaseIndex(index); } public int getMass(int index) { return mass[index] != 0 ? (mass[index] & 0xff) : (capacity + 1); } public void spliceEdge(int parent, int node, int newNode) { assert (!isLeaf(newNode)); if (node == leftIndex[parent]) { leftIndex[parent] = (char) newNode; } else { rightIndex[parent] = (char) newNode; } if (parentIndex != null && isInternal(node)) { parentIndex[node] = (byte) newNode; } } public void replaceParentBySibling(int grandParent, int parent, int node) { int sibling = getSibling(node, parent); if (parent == leftIndex[grandParent]) { leftIndex[grandParent] = (char) sibling; } else { rightIndex[grandParent] = (char) sibling; } if (parentIndex != null && isInternal(sibling)) { parentIndex[sibling] = (byte) grandParent; } } public int getCutDimension(int index) { return cutDimension[index] & 0xff; } public int[] getCutDimension() { return toIntArray(cutDimension); } public int[] getLeftIndex() { return toIntArray(leftIndex); } public int[] getRightIndex() { return toIntArray(rightIndex); } }
461
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/AnomalyScoreVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import com.amazon.randomcutforest.CommonUtils; /** * This visitor computes a scalar anomaly score for a specified point. The basic * score computation is defined by {@link AbstractScalarScoreVisitor}, and this * class overrides the scoring functions so that input points that are more * likely to separated from in-sample points by a random cut receive a higher * anomaly score. * * While this basic algorithm produces good results when all the points in the * sample are distinct, it can produce unexpected results when a significant * portion of the points in the sample are duplicates. Therefore this class * supports different optional features for modifying the score produced when * the point being scored is equal to the leaf node in the traversal. */ public class AnomalyScoreVisitor extends AbstractScalarScoreVisitor { /** * Construct a new ScalarScoreVisitor * * @param pointToScore The point whose anomaly score we are computing * @param treeMass The total mass of the RandomCutTree that is scoring the * point */ public AnomalyScoreVisitor(float[] pointToScore, int treeMass) { super(pointToScore, treeMass); } /** * Construct a new ScalarScoreVisitor * * @param pointToScore The point whose anomaly score we are computing * @param treeMass The total mass of the RandomCutTree that is * scoring the point * @param ignoreLeafMassThreshold Is the maximum mass of the leaf which can be * ignored */ public AnomalyScoreVisitor(float[] pointToScore, int treeMass, int ignoreLeafMassThreshold) { super(pointToScore, treeMass, ignoreLeafMassThreshold); } @Override protected double scoreSeen(int depth, int mass) { return CommonUtils.defaultScoreSeenFunction(depth, mass); } @Override protected double scoreUnseen(int depth, int mass) { return CommonUtils.defaultScoreUnseenFunction(depth, mass); } @Override protected double damp(int leafMass, int treeMass) { return CommonUtils.defaultDampFunction(leafMass, treeMass); } }
462
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/TransductiveScalarScoreVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import java.util.function.BiFunction; import java.util.function.Function; import com.amazon.randomcutforest.tree.IBoundingBoxView; import com.amazon.randomcutforest.tree.INodeView; public class TransductiveScalarScoreVisitor extends DynamicScoreVisitor { /* * the goal of this visitor is to allow tranductive inference; where during * scoring we make adjustments so that it appears (to the best of simulation * ability) that the tree was built using the knowledge of the point being * scored * */ protected final Function<IBoundingBoxView, double[]> vecSepScore; /** * Construct a new SimulatedTransductiveScalarScoreVisitor * * @param pointToScore The point whose anomaly score we are computing * @param treeMass The total mass of the RandomCutTree that is scoring the * point * @param scoreSeen is the part of the score function when the point has been * seen * @param scoreUnseen is the part of the score when the point has not been seen * @param damp corresponds to the dampening of the effect of the seen * points * @param vecSep A function that provides the probabilities of choosing * different dimensions given a BoundingBox when the tree * was built. This must be the same as the probabilies of * Transductive inference during scoring. For extenstions * where these are different, see * SimulatedTransductiveScalarScoreVisitor * * Note that scores are not normalized because the function * ranges are unknown as is the case with * DynamicScoreVisitor */ public TransductiveScalarScoreVisitor(float[] pointToScore, int treeMass, BiFunction<Double, Double, Double> scoreSeen, BiFunction<Double, Double, Double> scoreUnseen, BiFunction<Double, Double, Double> damp, Function<IBoundingBoxView, double[]> vecSep) { super(pointToScore, treeMass, 0, scoreSeen, scoreUnseen, damp); this.vecSepScore = vecSep; // build function is the same as scoring function } /** * Update the anomaly score based on the next step of the tree traversal. * * @param node The current node in the tree traversal * @param depthOfNode The depth of the current node in the tree */ @Override public void accept(INodeView node, int depthOfNode) { if (pointInsideBox) { return; } // note that score was unchanged before the return // this is only reasonable if the scoring was done using the same // probability function used to build the trees. double probabilityOfSeparation = getProbabilityOfSeparation(node.getBoundingBox()); double weight = getWeight(node.getCutDimension(), vecSepScore, node.getBoundingBox()); if (probabilityOfSeparation == 0) { pointInsideBox = true; return; } score = probabilityOfSeparation * scoreUnseen(depthOfNode, node.getMass()) + weight * score; } /** * Compute the probability that a random cut would separate the point from the * rest of the bounding box. This method is intended to compute the probability * for a non-leaf Node, and will throw an exception if a leaf-node bounding box * is detected. * * @param boundingBox The bounding box that we are computing the probability of * separation from. * @return is the probability */ @Override protected double getProbabilityOfSeparation(final IBoundingBoxView boundingBox) { double sumOfDenominator = 0d; double sumOfNumerator = 0d; double[] vec = vecSepScore.apply(boundingBox.getMergedBox(pointToScore)); for (int i = 0; i < pointToScore.length; ++i) { double maxVal = boundingBox.getMaxValue(i); double minVal = boundingBox.getMinValue(i); double oldRange = maxVal - minVal; sumOfDenominator += vec[i]; if (!coordInsideBox[i]) { if (maxVal < pointToScore[i]) { maxVal = pointToScore[i]; } else if (minVal > pointToScore[i]) { minVal = pointToScore[i]; } double newRange = maxVal - minVal; if (newRange > oldRange) { sumOfNumerator += vec[i] * (newRange - oldRange) / newRange; } else coordInsideBox[i] = true; } } if (sumOfDenominator <= 0) { // Sum of range across dimensions should only be 0 at leaf nodes as non-leaf // nodes always contain // more than one distinct point throw new IllegalStateException("Incorrect State"); } return sumOfNumerator / sumOfDenominator; // for RCFs vec[i] = newRange (for dimension i) and therefore the // sumOfNumerator is the sum of the difference (after and before // merging the point to the box) of ranges // sum of denominator is the sum the ranges in each dimension } // for this visitor class the assumption is that the trees are built using the // same probabilities as are used in scoring. In the application herein // vecSepBuild // is the same as vecSepScore as in the accept(node) above; however the function // is // written in the more general form so that it can be used for the Simulated // version as well without any changes. protected double getWeight(int dim, Function<IBoundingBoxView, double[]> vecSepBuild, final IBoundingBoxView boundingBox) { double[] vecSmall = vecSepBuild.apply(boundingBox); // the smaller box was built! IBoundingBoxView largeBox = boundingBox.getMergedBox(pointToScore); double[] vecLarge = vecSepScore.apply(largeBox); // the larger box is only scored! double sumSmall = 0; double sumLarge = 0; for (int i = 0; i < pointToScore.length; i++) { sumSmall += vecSmall[i]; sumLarge += vecLarge[i]; } return (boundingBox.getRange(dim) / largeBox.getRange(dim)) * (sumSmall / sumLarge) * (vecLarge[dim] / vecSmall[dim]); // this can be larger than 1 // For RCFs vecLarge[dim] = largeBox.getRange(dim) and // vecSmall[dim] = smallBox.getRange(dim) // sumSmall/sumLarge is the probability of non-separation } }
463
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/AbstractScalarScoreVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import java.util.Arrays; import com.amazon.randomcutforest.CommonUtils; import com.amazon.randomcutforest.Visitor; import com.amazon.randomcutforest.tree.IBoundingBoxView; import com.amazon.randomcutforest.tree.INodeView; /** * This abstract visitor encodes a standard method for computing a scalar result * value. The basic computation is as follows: * * <ol> * <li>After following the traversal path to a leaf, compute a base score at the * leaf node.</li> * <li>For each node in the traversal path from the leaf to the root, compute * the probability that a random cut would separate the query point from the * node. The updated score uses this probability to create a weighted * combination between the current score and a score contribution from the * current node.</li> * </ol> * <p> * While this basic algorithm produces good results when all the points in the * sample are distinct, it can produce unexpected results when a significant * portion of the points in the sample are duplicates. Therefore this class * supports different optional features for modifying the score produced when * the point being scored is equal to the leaf node in the traversal. */ public abstract class AbstractScalarScoreVisitor implements Visitor<Double> { public static final int DEFAULT_IGNORE_LEAF_MASS_THRESHOLD = 0; /** * The point whose anomaly score is being computed. */ protected final float[] pointToScore; /** * The mass of the tree being visited. This value is used to normalize the final * result. */ protected final int treeMass; /** * This flag is set to 'true' if the point being scored is found to be contained * by a bounding box in the traversal path, allowing us to short-circuit further * computation. */ protected boolean pointInsideBox; /** * Similar to pointInsideBox, the array coordInsideBox keeps track of whether * each coordinate is contained in the corresponding bounding box projection for * a bounding box in the traversal path. This field is used to skip unnecessary * steps in the probability computation. */ protected boolean[] coordInsideBox; /** * shadowbox used in attribution and ignoring the leaf to simulate a deletion */ protected IBoundingBoxView shadowBox = null; /** * The function used to compute the base score in the case where the point being * scored is equal to the leaf point (provided the ignoreLeafEquals and * ignoreLeafMassThreshold variables indicate that we should use this method). * * Function arguments: leaf depth, leaf mass */ protected double score; /** * If true, then the scoreUnseen method will be used to score a point equal to a * leaf point in {@link #acceptLeaf(INodeView, int)}. */ protected boolean ignoreLeafEquals; /** * If the point being scored is equal to the leaf point but the leaf mass is * smaller than this value, then the scoreUnseen method will be used to score * the point in {@link #accept(INodeView, int)}. */ protected int ignoreLeafMassThreshold; /** * Construct a new ScalarScoreVisitor * * @param pointToScore The point whose anomaly score we are computing * @param treeMass The total mass of the RandomCutTree that is * scoring the point * @param ignoreLeafMassThreshold Is the maximum mass of the leaf which can be * ignored */ public AbstractScalarScoreVisitor(float[] pointToScore, int treeMass, int ignoreLeafMassThreshold) { this.pointToScore = Arrays.copyOf(pointToScore, pointToScore.length); this.treeMass = treeMass; pointInsideBox = false; score = 0.0; this.ignoreLeafEquals = (ignoreLeafMassThreshold > DEFAULT_IGNORE_LEAF_MASS_THRESHOLD); this.ignoreLeafMassThreshold = ignoreLeafMassThreshold; // will be initialized to an array of false values coordInsideBox = new boolean[pointToScore.length]; } /** * Construct a new AbstractScalarScoreVisitor using default leaf options. * * @param pointToScore The point whose anomaly score we are computing * @param treeMass The total mass of the RandomCutTree that is scoring the * point */ public AbstractScalarScoreVisitor(float[] pointToScore, int treeMass) { this(pointToScore, treeMass, DEFAULT_IGNORE_LEAF_MASS_THRESHOLD); } /** * @return The score computed up until this point. */ @Override public Double getResult() { return CommonUtils.defaultScalarNormalizerFunction(score, treeMass); } /** * Update the anomaly score based on the next step of the tree traversal. * * @param node The current node in the tree traversal * @param depthOfNode The depth of the current node in the tree */ @Override public void accept(INodeView node, int depthOfNode) { if (pointInsideBox) { return; } double probabilityOfSeparation; if (!ignoreLeafEquals) { probabilityOfSeparation = node.probailityOfSeparation(pointToScore); if (probabilityOfSeparation <= 0) { pointInsideBox = true; return; } } else { shadowBox = shadowBox == null ? node.getSiblingBoundingBox(pointToScore) : shadowBox.getMergedBox(node.getSiblingBoundingBox(pointToScore)); probabilityOfSeparation = (shadowBox.getRangeSum() <= 0) ? 1.0 : getProbabilityOfSeparation(shadowBox); } score = probabilityOfSeparation * scoreUnseen(depthOfNode, node.getMass()) + (1 - probabilityOfSeparation) * score; } /** * Update the anomaly score with the given leaf node. * * @param leafNode The leaf node that was reached by traversing the tree * @param depthOfNode The depth of the leaf node */ @Override public void acceptLeaf(INodeView leafNode, int depthOfNode) { if (Arrays.equals(leafNode.getLeafPoint(), pointToScore) && (!ignoreLeafEquals || (leafNode.getMass() > ignoreLeafMassThreshold))) { pointInsideBox = true; score = damp(leafNode.getMass(), treeMass) * scoreSeen(depthOfNode, leafNode.getMass()); } else { score = scoreUnseen(depthOfNode, leafNode.getMass()); } } /** * A scoring function which is applied when the leaf node visited is equal to * the point being scored. * * @param depth The depth of the node being visited * @param mass The mass of the node being visited * @return an anomaly score contribution for a given node */ protected abstract double scoreSeen(int depth, int mass); /** * A scoring function which is applied when the leaf node visited is not equal * to the point being scored. This function is also used to compute the * contribution to the anomaly score from non-leaf nodes. * * @param depth The depth of the node being visited. * @param mass The mass of the node being visited. * @return an anomaly score contribution for a given node. */ protected abstract double scoreUnseen(int depth, int mass); /** * This function produces a scaling factor which can be used to reduce the * influence of leaf nodes with mass greater than 1. * * @param leafMass The mass of the leaf node visited * @param treeMass The mass of the tree being visited * @return a scaling factor to apply to the result from * {@link #scoreSeen(int, int)}. */ protected abstract double damp(int leafMass, int treeMass); /** * Compute the probability that a random cut would separate the point from the * rest of the bounding box. This method is intended to compute the probability * for a non-leaf Node, and will throw an exception if a leaf-node bounding box * is detected. * * @param boundingBox The bounding box that we are computing the probability of * separation from. * @return is the probability */ protected double getProbabilityOfSeparation(final IBoundingBoxView boundingBox) { double sumOfNewRange = 0d; double sumOfDifferenceInRange = 0d; for (int i = 0; i < pointToScore.length; ++i) { double maxVal = boundingBox.getMaxValue(i); double minVal = boundingBox.getMinValue(i); double oldRange = maxVal - minVal; if (!coordInsideBox[i]) { if (maxVal < pointToScore[i]) { maxVal = pointToScore[i]; } else if (minVal > pointToScore[i]) { minVal = pointToScore[i]; } else if (!ignoreLeafEquals) { // optimization turned on for ignoreLeafEquals==false sumOfNewRange += oldRange; coordInsideBox[i] = true; continue; } double newRange = maxVal - minVal; sumOfNewRange += newRange; sumOfDifferenceInRange += (newRange - oldRange); } else { sumOfNewRange += oldRange; } } if (sumOfNewRange <= 0) { // Sum of range across dimensions should only be 0 at leaf nodes as non-leaf // nodes always contain // more than one distinct point throw new IllegalStateException("Sum of new range of merged box in scoring function is smaller than 0 " + "for a non-leaf node. The sum of range of new bounding box is: " + sumOfNewRange); } return sumOfDifferenceInRange / sumOfNewRange; } public boolean isConverged() { return pointInsideBox; } }
464
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/AbstractAttributionVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import java.util.Arrays; import com.amazon.randomcutforest.CommonUtils; import com.amazon.randomcutforest.Visitor; import com.amazon.randomcutforest.returntypes.DiVector; import com.amazon.randomcutforest.tree.IBoundingBoxView; import com.amazon.randomcutforest.tree.INodeView; /** * Attribution exposes the attribution of scores produced by ScalarScoreVisitor * corresponding to different attributes. It allows a boolean * ignoreClosestCandidate; which when true will compute the attribution as it * that near neighbor was not present in RCF. This is turned on by default for * duplicate points seen by the forest, so that the attribution does not change * is a sequence of duplicate points are seen. For non-duplicate points, if the * boolean turned on, reduces effects of masking (when anomalous points are * included in the forest -- which will be true with a few samples or when the * samples are not refreshed appropriately). It is worth remembering that * disallowing anomalous points from being included in the forest explicitly * will render the algorithm incapable of adjusting to a new normal -- which is * a strength of this algorithm. **/ public abstract class AbstractAttributionVisitor implements Visitor<DiVector> { public static final int DEFAULT_IGNORE_LEAF_MASS_THRESHOLD = 0; protected final double[] differenceInRangeVector; protected final float[] pointToScore; protected final int treeMass; protected final DiVector directionalAttribution; protected boolean hitDuplicates; protected double savedScore; protected double sumOfNewRange; protected double sumOfDifferenceInRange; protected boolean ignoreLeaf; protected int ignoreLeafMassThreshold; /** * A flag that states whether the point to score is known to be contained inside * the bounding box of Nodes being accepted. Assumes nodes are accepted in * leaf-to-root order. */ protected boolean pointInsideBox; /** * An array that keeps track of whether each margin of the point being scored is * outside inside the box considered during the recursive call to compute the * score. Assumes nodes are accepted in leaf-to-root order. */ protected boolean[] coordInsideBox; protected IBoundingBoxView shadowBox; public AbstractAttributionVisitor(float[] pointToScore, int treeMass, int ignoreLeafMassThreshold) { this.pointToScore = Arrays.copyOf(pointToScore, pointToScore.length); this.treeMass = treeMass; this.ignoreLeaf = ignoreLeafMassThreshold > DEFAULT_IGNORE_LEAF_MASS_THRESHOLD; this.ignoreLeafMassThreshold = ignoreLeafMassThreshold; hitDuplicates = false; pointInsideBox = false; savedScore = 0; directionalAttribution = new DiVector(pointToScore.length); shadowBox = null; coordInsideBox = new boolean[pointToScore.length]; // array is twice as long as pointToScore because we store // positive and negative differences separately differenceInRangeVector = new double[2 * pointToScore.length]; } public AbstractAttributionVisitor(float[] pointToScore, int treeMass) { this(pointToScore, treeMass, DEFAULT_IGNORE_LEAF_MASS_THRESHOLD); } /** * Take the normalization function applied to the corresponding scoring visitor * and apply that to each coordinate of the DiVector to modify the data in * place. The function has to be associative in its first parameter; that is, fn * (x1, y) + fn (x2, y) = fn (x1 + x2, y) * * @return The modified data. */ @Override public DiVector getResult() { DiVector result = new DiVector(directionalAttribution); result.componentwiseTransform(x -> CommonUtils.defaultScalarNormalizerFunction(x, treeMass)); return result; } /** * Update the anomaly score based on the next step of the tree traversal. * * @param node The current node in the tree traversal * @param depthOfNode The depth of the current node in the tree */ @Override public void accept(INodeView node, int depthOfNode) { if (pointInsideBox) { return; } IBoundingBoxView smallBox; if (hitDuplicates || ignoreLeaf) { // use the sibling bounding box to represent counterfactual "what if point & the // candidate near neighbor // had not been inserted in the tree" shadowBox = shadowBox == null ? node.getSiblingBoundingBox(pointToScore) : shadowBox.getMergedBox(node.getSiblingBoundingBox(pointToScore)); smallBox = shadowBox; } else { smallBox = node.getBoundingBox(); } IBoundingBoxView largeBox = smallBox.getMergedBox(pointToScore); updateRangesForScoring(smallBox, largeBox); double probOfCut = sumOfDifferenceInRange / sumOfNewRange; // if leaves were ignored we need to keep accounting for the score if (ignoreLeaf) { savedScore = probOfCut * scoreUnseen(depthOfNode, node.getMass()) + (1 - probOfCut) * savedScore; } if (probOfCut <= 0) { pointInsideBox = true; } else { double newScore = scoreUnseen(depthOfNode, node.getMass()); for (int i = 0; i < pointToScore.length; i++) { double probOfCutInSpikeDirection = differenceInRangeVector[2 * i] / sumOfNewRange; directionalAttribution.high[i] = probOfCutInSpikeDirection * newScore + (1 - probOfCut) * directionalAttribution.high[i]; double probOfCutInDipDirection = differenceInRangeVector[2 * i + 1] / sumOfNewRange; directionalAttribution.low[i] = probOfCutInDipDirection * newScore + (1 - probOfCut) * directionalAttribution.low[i]; } } boolean capture = (pointInsideBox || depthOfNode == 0); if ((hitDuplicates || ignoreLeaf) && capture) { // final rescaling; this ensures agreement with the ScalarScoreVector // the scoreUnseen/scoreSeen should be the same as scoring; other uses need // caution. directionalAttribution.renormalize(savedScore); } } @Override public void acceptLeaf(INodeView leafNode, int depthOfNode) { updateRangesForScoring(leafNode.getBoundingBox(), leafNode.getBoundingBox().getMergedBox(pointToScore)); // newrange == 0 corresponds to equality of points and is fater than // Array.equals if (sumOfNewRange <= 0) { hitDuplicates = true; } if ((hitDuplicates) && ((!ignoreLeaf) || (leafNode.getMass() > ignoreLeafMassThreshold))) { savedScore = damp(leafNode.getMass(), treeMass) * scoreSeen(depthOfNode, leafNode.getMass()); } else { savedScore = scoreUnseen(depthOfNode, leafNode.getMass()); } if ((hitDuplicates) || ((ignoreLeaf) && (leafNode.getMass() <= ignoreLeafMassThreshold))) { Arrays.fill(directionalAttribution.high, savedScore / (2 * pointToScore.length)); Arrays.fill(directionalAttribution.low, savedScore / (2 * pointToScore.length)); /* in this case do not have a better option than an equal attribution */ Arrays.fill(coordInsideBox, false); } else { for (int i = 0; i < pointToScore.length; i++) { directionalAttribution.high[i] = savedScore * differenceInRangeVector[2 * i] / sumOfNewRange; directionalAttribution.low[i] = savedScore * differenceInRangeVector[2 * i + 1] / sumOfNewRange; } } } /** * A scoring function which is applied when the leaf node visited is equal to * the point being scored. * * @param depth The depth of the node being visited * @param mass The mass of the node being visited * @return an anomaly score contribution for a given node */ protected abstract double scoreSeen(int depth, int mass); /** * A scoring function which is applied when the leaf node visited is not equal * to the point being scored. This function is also used to compute the * contribution to the anomaly score from non-leaf nodes. * * @param depth The depth of the node being visited. * @param mass The mass of the node being visited. * @return an anomaly score contribution for a given node. */ protected abstract double scoreUnseen(int depth, int mass); /** * This function produces a scaling factor which can be used to reduce the * influence of leaf nodes with mass greater than 1. * * @param leafMass The mass of the leaf node visited * @param treeMass The mass of the tree being visited * @return a scaling factor to apply to the result from * {@link #scoreSeen(int, int)}. */ protected abstract double damp(int leafMass, int treeMass); /** * When updating the score for a node, we compare the node's bounding box to the * merged bounding box that would be created by adding the point to be scored. * This method updates local instance variables sumOfDifferenceInRange and * differenceInRange vector to reflect the total difference in side length and * the difference in side length in each dimension, respectively. * * @param smallBox The bounding box corresponding to a Node being visited. * @param largeBox The merged bounding box containing smallBox and the point * being scored. */ protected void updateRangesForScoring(IBoundingBoxView smallBox, IBoundingBoxView largeBox) { sumOfDifferenceInRange = 0.0; sumOfNewRange = 0.0; Arrays.fill(differenceInRangeVector, 0.0); for (int i = 0; i < pointToScore.length; i++) { sumOfNewRange += largeBox.getRange(i); // optimization turned off for ignoreLeaf if (coordInsideBox[i] && !ignoreLeaf) { continue; } double maxGap = Math.max(largeBox.getMaxValue(i) - smallBox.getMaxValue(i), 0.0); double minGap = Math.max(smallBox.getMinValue(i) - largeBox.getMinValue(i), 0.0); if (maxGap + minGap > 0.0) { sumOfDifferenceInRange += (minGap + maxGap); differenceInRangeVector[2 * i] = maxGap; differenceInRangeVector[2 * i + 1] = minGap; } else { coordInsideBox[i] = true; } } } @Override public boolean isConverged() { return pointInsideBox; } }
465
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/DynamicAttributionVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import java.util.function.BiFunction; import com.amazon.randomcutforest.returntypes.DiVector; public class DynamicAttributionVisitor extends AbstractAttributionVisitor { /** * The function used to compute the base score in the case where the point being * scored is equal to the leaf point (provided the ignoreLeafEquals and * ignoreLeafMassThreshold variables indicate that we should use this method). * <p> * Function arguments: leaf depth, leaf mass */ private final BiFunction<Double, Double, Double> scoreSeen; /** * A damping function used to dilute the impact of a point with a large number * of duplicates on the base score. * <p> * Function arguments: leaf mass, tree mass */ private final BiFunction<Double, Double, Double> damp; /** * The scoring function to use when the point being scored is not equal to the * leaf point, or when the points are equal but the ignoreLeafEquals or * ignoreLeafMassThreshold variable indicates that we should use the scoreUnseen * method. * <p> * Function arguments: leaf depth, leaf mass */ private final BiFunction<Double, Double, Double> scoreUnseen; /** * * @param point to be scored * @param treeMass mass of the tree * @param ignoreLeafMassThreshold threshold of mass for leaves to be ignored * @param scoreSeen part of the score when point has been seen * @param scoreUnseen part of the score for unseen point * @param damp dampening function for seen points */ public DynamicAttributionVisitor(float[] point, int treeMass, int ignoreLeafMassThreshold, BiFunction<Double, Double, Double> scoreSeen, BiFunction<Double, Double, Double> scoreUnseen, BiFunction<Double, Double, Double> damp) { super(point, treeMass, ignoreLeafMassThreshold); this.scoreSeen = scoreSeen; this.scoreUnseen = scoreUnseen; this.damp = damp; } @Override protected double scoreSeen(int depth, int leafMass) { return scoreSeen.apply((double) depth, (double) leafMass); } @Override protected double scoreUnseen(int depth, int leafMass) { return scoreUnseen.apply((double) depth, (double) leafMass); } @Override protected double damp(int leafMass, int treeMass) { return damp.apply((double) leafMass, (double) treeMass); } // turning off normalization @Override public DiVector getResult() { return new DiVector(directionalAttribution); } }
466
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/DynamicScoreVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import java.util.function.BiFunction; public class DynamicScoreVisitor extends AbstractScalarScoreVisitor { /** * The function used to compute the base score in the case where the point being * scored is equal to the leaf point (provided the ignoreLeafEquals and * ignoreLeafMassThreshold variables indicate that we should use this method). * <p> * Function arguments: leaf depth, leaf mass */ protected final BiFunction<Double, Double, Double> scoreSeen; /** * A damping function used to dilute the impact of a point with a large number * of duplicates on the base score. * <p> * Function arguments: leaf mass, tree mass */ protected final BiFunction<Double, Double, Double> damp; /** * The scoring function to use when the point being scored is not equal to the * leaf point, or when the points are equal but the ignoreLeafEquals or * ignoreLeafMassThreshold variable indicates that we should use the scoreUnseen * method. * <p> * Function arguments: leaf depth, leaf mass */ protected final BiFunction<Double, Double, Double> scoreUnseen; /** * Constructor * * @param point being scored * @param treeMass mass of the tree * @param ignoreLeafMassThreshold the threshold for ignoring leaf nodes * @param scoreSeen the part of score function for previously seen * values * @param scoreUnseen part of the score for unseen values * @param damp dampening function for seen points */ public DynamicScoreVisitor(float[] point, int treeMass, int ignoreLeafMassThreshold, BiFunction<Double, Double, Double> scoreSeen, BiFunction<Double, Double, Double> scoreUnseen, BiFunction<Double, Double, Double> damp) { super(point, treeMass, ignoreLeafMassThreshold); this.scoreSeen = scoreSeen; this.scoreUnseen = scoreUnseen; this.damp = damp; } @Override protected double scoreSeen(int depth, int leafMass) { return scoreSeen.apply((double) depth, (double) leafMass); } @Override protected double scoreUnseen(int depth, int leafMass) { return scoreUnseen.apply((double) depth, (double) leafMass); } @Override protected double damp(int leafMass, int treeMass) { return damp.apply((double) leafMass, (double) treeMass); } /** * normalization is turned off for dynamic scoring because the function ranges * are unknown */ @Override public Double getResult() { return score; } }
467
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/SimulatedTransductiveScalarScoreVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import java.util.function.BiFunction; import java.util.function.Function; import com.amazon.randomcutforest.tree.IBoundingBoxView; import com.amazon.randomcutforest.tree.INodeView; public class SimulatedTransductiveScalarScoreVisitor extends TransductiveScalarScoreVisitor { private final Function<IBoundingBoxView, double[]> vecSepBuild; /** * Construct a new SimulatedTransductiveScalarScoreVisitor * * @param pointToScore The point whose anomaly score we are computing * @param treeMass The total mass of the RandomCutTree that is scoring the * point * @param scoreSeen is the part of the score function when the point has been * seen * @param scoreUnseen is the part of the score when the point has not been seen * @param damp corresponds to the dampening of the effect of the seen * points * @param vecSepBuild A function that provides the probabilities of choosing * different dimensions given a BoundingBox when the tree * was built. * @param vecSepScore A function that corresponds to importance of dimensions * during scoring */ public SimulatedTransductiveScalarScoreVisitor(float[] pointToScore, int treeMass, BiFunction<Double, Double, Double> scoreSeen, BiFunction<Double, Double, Double> scoreUnseen, BiFunction<Double, Double, Double> damp, Function<IBoundingBoxView, double[]> vecSepBuild, Function<IBoundingBoxView, double[]> vecSepScore) { super(pointToScore, treeMass, scoreSeen, scoreUnseen, damp, vecSepScore); this.vecSepBuild = vecSepBuild; } /** * Update the anomaly score based on the next step of the tree traversal. * * @param node The current node in the tree traversal * @param depthOfNode The depth of the current node in the tree */ @Override public void accept(INodeView node, int depthOfNode) { double weight = getWeight(node.getCutDimension(), vecSepBuild, node.getBoundingBox()); if (pointInsideBox) { score *= weight; return; } double probabilityOfSeparation = getProbabilityOfSeparation(node.getBoundingBox()); if (probabilityOfSeparation == 0) { pointInsideBox = true; } score = probabilityOfSeparation * scoreUnseen(depthOfNode, node.getMass()) + weight * score; } // The above function differs from TransductiveScalarScoreVisitor only in the // weight // computation and when the weight function is used. @Override public boolean isConverged() { return false; } }
468
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/anomalydetection/AnomalyAttributionVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.anomalydetection; import com.amazon.randomcutforest.CommonUtils; /** * Attribution exposes the attribution of scores produced by ScalarScoreVisitor * corresponding to different attributes. It allows a boolean * ignoreClosestCandidate; which when true will compute the attribution as it * that near neighbor was not present in RCF. This is turned on by default for * duplicate points seen by the forest, so that the attribution does not change * is a sequence of duplicate points are seen. For non-duplicate points, if the * boolean turned on, reduces effects of masking (when anomalous points are * included in the forest (which will be true with a few samples or when the * samples are not refreshed appropriately). It is worth remembering that * disallowing anomalous points from being included in the forest forest * explicitly will render the algorithm incapable of adjusting to a new normal * -- which is a strength of this algorithm. **/ public class AnomalyAttributionVisitor extends AbstractAttributionVisitor { public AnomalyAttributionVisitor(float[] pointToScore, int treeMass, int ignoreThreshold) { super(pointToScore, treeMass, ignoreThreshold); } public AnomalyAttributionVisitor(float[] pointToScore, int treeMass) { super(pointToScore, treeMass); } @Override protected double scoreSeen(int depth, int mass) { return CommonUtils.defaultScoreSeenFunction(depth, mass); } @Override protected double scoreUnseen(int depth, int mass) { return CommonUtils.defaultScoreUnseenFunction(depth, mass); } @Override protected double damp(int leafMass, int treeMass) { return CommonUtils.defaultDampFunction(leafMass, treeMass); } }
469
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/inspect/NearNeighborVisitor.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.inspect; import java.util.ArrayList; import java.util.List; import java.util.Optional; import com.amazon.randomcutforest.Visitor; import com.amazon.randomcutforest.returntypes.Neighbor; import com.amazon.randomcutforest.tree.INodeView; /** * A visitor that returns the leaf node in a traversal if the distance between * the leaf point and the query point is less than a given threshold. */ public class NearNeighborVisitor implements Visitor<Optional<Neighbor>> { private final float[] queryPoint; private final double distanceThreshold; private Neighbor neighbor; /** * Create a NearNeighborVisitor for the given query point. * * @param queryPoint The point whose neighbors we are looking for. * @param distanceThreshold Leaf points whose distance from the query point is * less than this value are considered near neighbors. */ public NearNeighborVisitor(float[] queryPoint, double distanceThreshold) { this.queryPoint = queryPoint; this.distanceThreshold = distanceThreshold; neighbor = null; } /** * Create a NearNeighborVisitor which always returns the leaf point in the * traversal. The distance threshold is set to positive infinity. * * @param queryPoint The point whose neighbors we are looking for. */ public NearNeighborVisitor(float[] queryPoint) { this(queryPoint, Double.POSITIVE_INFINITY); } /** * Near neighbors are identified in the {@link #acceptLeaf} method, hence this * method does nothing. * * @param node the node being visited * @param depthOfNode the depth of the node being visited */ @Override public void accept(INodeView node, int depthOfNode) { } /** * Check to see whether the Euclidean distance between the leaf point and the * query point is less than the distance threshold. If it is, then this visitor * will return an {@link java.util.Optional} containing this leaf point * (converted to a {@link Neighbor} object). Otherwise, this visitor will return * an empty Optional. * * @param leafNode the leaf node being visited * @param depthOfNode the depth of the leaf node */ @Override public void acceptLeaf(INodeView leafNode, int depthOfNode) { float[] leafPoint = leafNode.getLiftedLeafPoint(); double distanceSquared = 0.0; for (int i = 0; i < leafPoint.length; i++) { double diff = queryPoint[i] - leafPoint[i]; distanceSquared += diff * diff; } if (Math.sqrt(distanceSquared) < distanceThreshold) { List<Long> sequenceIndexes = new ArrayList<>(leafNode.getSequenceIndexes().keySet()); neighbor = new Neighbor(leafPoint, Math.sqrt(distanceSquared), sequenceIndexes); } } /** * @return an {@link Optional} containing the leaf point (converted to a * {@link Neighbor} if the Euclidean distance between the leaf point and * the query point is less than the distance threshold. Otherwise return * an empty Optional. */ @Override public Optional<Neighbor> getResult() { return Optional.ofNullable(neighbor); } @Override public boolean isConverged() { return true; } }
470
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/util/ArrayPacking.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.checkNotNull; import static java.lang.Math.min; import java.nio.ByteBuffer; import java.util.Arrays; public class ArrayPacking { /** * For a given base value, return the smallest int value {@code p} so that * {@code base^(p + 1) >= Integer.MAX_VALUE}. If * {@code base >= Integer.MAX_VALUE}, return 1. * * @param base Compute the approximate log of {@code Integer.MAX_VALUE} in this * base. * @return the largest int value {@code p} so that * {@code base^p >= Integer.MAX_VALUE} or 1 if * {@code base >= Integer.MAX_VALUE}. */ public static int logMax(long base) { checkArgument(base > 1, "Absolute value of base must be greater than 1"); int pack = 0; long num = base; while (num < Integer.MAX_VALUE) { num = num * base; ++pack; } return Math.max(pack, 1); // pack can be 0 for max - min being more than Integer.MaxValue } /** * Pack an array of ints. If {@code compress} is true, then this method will * apply arithmetic compression to the inputs, otherwise it returns a copy of * the input. * * @param inputArray An array of ints to pack. * @param compress A flag indicating whether to apply arithmetic compression. * @return an array of packed ints. */ public static int[] pack(int[] inputArray, boolean compress) { return pack(inputArray, inputArray.length, compress); } /** * Pack an array of ints. If {@code compress} is true, then this method will * apply arithmetic compression to the inputs, otherwise it returns a copy of * the input. * * @param inputArray An array of ints to pack. * @param length The length of the output array. Only the first * {@code length} values in {@code inputArray} will be packed. * @param compress A flag indicating whether to apply arithmetic compression. * @return an array of packed ints. */ public static int[] pack(int[] inputArray, int length, boolean compress) { checkNotNull(inputArray, "inputArray must not be null"); checkArgument(0 <= length && length <= inputArray.length, "length must be between 0 and inputArray.length (inclusive)"); if (!compress || length < 3) { return Arrays.copyOf(inputArray, length); } int min = inputArray[0]; int max = inputArray[0]; for (int i = 1; i < length; i++) { min = min(min, inputArray[i]); max = Math.max(max, inputArray[i]); } long base = (long) max - min + 1; if (base == 1) { return new int[] { min, max, length }; } else { int packNum = logMax(base); int[] output = new int[3 + (int) Math.ceil(1.0 * length / packNum)]; output[0] = min; output[1] = max; output[2] = length; int len = 0; int used = 0; while (len < length) { long code = 0; int reach = min(len + packNum - 1, length - 1); for (int i = reach; i >= len; i--) { code = base * code + (inputArray[i] - min); } output[3 + used++] = (int) code; len += packNum; } // uncomment for debug; should be always true // checkArgument(used + 3 == output.length, "incorrect state"); return output; } } /** * Pack an array of shorts. If {@code compress} is true, then this method will * apply arithmetic compression to the inputs, otherwise it returns a copy of * the input. * * @param inputArray An array of ints to pack. * @param compress A flag indicating whether to apply arithmetic compression. * @return an array of packed ints. */ public static int[] pack(short[] inputArray, boolean compress) { return pack(inputArray, inputArray.length, compress); } /** * Pack an array of shorts. If {@code compress} is true, then this method will * apply arithmetic compression to the inputs, otherwise it returns a copy of * the input. * * @param inputArray An array of ints to pack. * @param length The length of the output array. Only the first * {@code length} values in {@code inputArray} will be packed. * @param compress A flag indicating whether to apply arithmetic compression. * @return an array of packed ints. */ public static int[] pack(short[] inputArray, int length, boolean compress) { checkNotNull(inputArray, "inputArray must not be null"); checkArgument(0 <= length && length <= inputArray.length, "length must be between 0 and inputArray.length (inclusive)"); if (!compress || length < 3) { int[] ret = new int[length]; for (int i = 0; i < length; i++) { ret[i] = inputArray[i]; } return ret; } int min = inputArray[0]; int max = inputArray[0]; for (int i = 1; i < length; i++) { min = min(min, inputArray[i]); max = Math.max(max, inputArray[i]); } long base = (long) max - min + 1; if (base == 1) { return new int[] { min, max, length }; } else { int packNum = logMax(base); int[] output = new int[3 + (int) Math.ceil(1.0 * length / packNum)]; output[0] = min; output[1] = max; output[2] = length; int len = 0; int used = 0; while (len < length) { long code = 0; int reach = min(len + packNum - 1, length - 1); for (int i = reach; i >= len; i--) { code = base * code + (inputArray[i] - min); } output[3 + used++] = (int) code; len += packNum; } // uncomment for debug; should be always true // checkArgument(used + 3 == output.length, "incorrect state"); return output; } } /** * Unpack an array previously created by {@link #pack(int[], int, boolean)}. * * @param packedArray An array previously created by * {@link #pack(int[], int, boolean)}. * @param decompress A flag indicating whether the packed array was created * with arithmetic compression enabled. * @return the array of unpacked ints. */ public static int[] unpackInts(int[] packedArray, boolean decompress) { checkNotNull(packedArray, " array unpacking invoked on null arrays"); if (!decompress) { return Arrays.copyOf(packedArray, packedArray.length); } return (packedArray.length < 3) ? unpackInts(packedArray, packedArray.length, decompress) : unpackInts(packedArray, packedArray[2], decompress); } /** * Unpack an array previously created by {@link #pack(int[], int, boolean)}. * * @param packedArray An array previously created by * {@link #pack(int[], int, boolean)}. * @param length The desired length of the output array. If this number is * different from the length of the array that was originally * packed, then the result will be truncated or padded with * zeros as needed. * @param decompress A flag indicating whether the packed array was created * with arithmetic compression enabled. * @return the array of unpacked ints. */ public static int[] unpackInts(int[] packedArray, int length, boolean decompress) { checkNotNull(packedArray, " array unpacking invoked on null arrays"); checkArgument(length >= 0, "incorrect length parameter"); if (packedArray.length < 3 || !decompress) { return Arrays.copyOf(packedArray, length); } int min = packedArray[0]; int max = packedArray[1]; int[] output = new int[length]; if (min == max) { if (packedArray[2] >= length) { Arrays.fill(output, min); } else { for (int i = 0; i < packedArray[2]; i++) { output[i] = min; } } } else { long base = ((long) max - min + 1); int packNum = logMax(base); int count = 0; for (int i = 3; i < packedArray.length; i++) { long code = packedArray[i]; for (int j = 0; j < packNum && count < min(packedArray[2], length); j++) { output[count++] = (int) (min + code % base); code = (int) (code / base); } } } return output; } private static short[] copyToShort(int[] array, int length) { short[] ret = new short[length]; for (int i = 0; i < Math.min(length, array.length); i++) { ret[i] = (short) array[i]; } return ret; } /** * Unpack an array previously created by {@link #pack(short[], int, boolean)}. * * @param packedArray An array previously created by * {@link #pack(short[], int, boolean)}. * @param decompress A flag indicating whether the packed array was created * with arithmetic compression enabled. * @return the array of unpacked shorts. */ public static short[] unpackShorts(int[] packedArray, boolean decompress) { checkNotNull(packedArray, " array unpacking invoked on null arrays"); if (!decompress) { return copyToShort(packedArray, packedArray.length); } return (packedArray.length < 3) ? unpackShorts(packedArray, packedArray.length, decompress) : unpackShorts(packedArray, packedArray[2], decompress); } /** * Unpack an array previously created by {@link #pack(short[], int, boolean)}. * * @param packedArray An array previously created by * {@link #pack(short[], int, boolean)}. * @param length The desired length of the output array. If this number is * different from the length of the array that was originally * packed, then the result will be truncated or padded with * zeros as needed. * @param decompress A flag indicating whether the packed array was created * with arithmetic compression enabled. * @return the array of unpacked ints. */ public static short[] unpackShorts(int[] packedArray, int length, boolean decompress) { checkNotNull(packedArray, " array unpacking invoked on null arrays"); checkArgument(length >= 0, "incorrect length parameter"); if (packedArray.length < 3 || !decompress) { return copyToShort(packedArray, length); } int min = packedArray[0]; int max = packedArray[1]; short[] output = new short[length]; if (min == max) { if (packedArray[2] >= length) { Arrays.fill(output, (short) min); } else { for (int i = 0; i < packedArray[2]; i++) { output[i] = (short) min; } } } else { long base = ((long) max - min + 1); int packNum = logMax(base); int count = 0; for (int i = 3; i < packedArray.length; i++) { long code = packedArray[i]; for (int j = 0; j < packNum && count < min(packedArray[2], length); j++) { output[count++] = (short) (min + code % base); code = (int) (code / base); } } } return output; } /** * Pack an array of doubles into an array of bytes. * * @param array An array of doubles. * @return An array of bytes representing the original array of doubles. */ public static byte[] pack(double[] array) { checkNotNull(array, "array must not be null"); return pack(array, array.length); } /** * Pack an array of doubles into an array of bytes. * * @param array An array of doubles. * @param length The number of doubles in the input array to pack into the * resulting byte array. * @return An array of bytes representing the original array of doubles. */ public static byte[] pack(double[] array, int length) { checkNotNull(array, "array must not be null"); checkArgument(0 <= length, "incorrect length parameter"); checkArgument(length <= array.length, "length must be between 0 and inputArray.length (inclusive)"); ByteBuffer buf = ByteBuffer.allocate(length * Double.BYTES); for (int i = 0; i < length; i++) { buf.putDouble(array[i]); } return buf.array(); } /** * Pack an array of floats into an array of bytes. * * @param array An array of floats. * @return An array of bytes representing the original array of floats. */ public static byte[] pack(float[] array) { checkNotNull(array, "array must not be null"); return pack(array, array.length); } /** * Pack an array of floats into an array of bytes. * * @param array An array of floats. * @param length The number of doubles in the input array to pack into the * resulting byte array. * @return An array of bytes representing the original array of floats. */ public static byte[] pack(float[] array, int length) { checkArgument(0 <= length, "incorrect length parameter"); checkArgument(length <= array.length, "length must be between 0 and inputArray.length (inclusive)"); ByteBuffer buf = ByteBuffer.allocate(length * Float.BYTES); for (int i = 0; i < length; i++) { buf.putFloat(array[i]); } return buf.array(); } /** * Unpack an array of bytes as an array of doubles. * * @param bytes An array of bytes. * @return an array of doubles obtained by marshalling consecutive bytes in the * input array into doubles. */ public static double[] unpackDoubles(byte[] bytes) { checkNotNull(bytes, "bytes must not be null"); return unpackDoubles(bytes, bytes.length / Double.BYTES); } /** * Unpack an array of bytes as an array of doubles. * * @param bytes An array of bytes. * @param length The desired length of the resulting double array. The input * will be truncated or padded with zeros as needed. * @return an array of doubles obtained by marshalling consecutive bytes in the * input array into doubles. */ public static double[] unpackDoubles(byte[] bytes, int length) { checkNotNull(bytes, "bytes must not be null"); checkArgument(length >= 0, "length must be greater than or equal to 0"); checkArgument(bytes.length % Double.BYTES == 0, "bytes.length must be divisible by Double.BYTES"); ByteBuffer buf = ByteBuffer.wrap(bytes); double[] result = new double[length]; int m = Math.min(length, bytes.length / Double.BYTES); for (int i = 0; i < m; i++) { result[i] = buf.getDouble(); } return result; } /** * Unpack an array of bytes as an array of floats. * * @param bytes An array of bytes. * @return an array of floats obtained by marshalling consecutive bytes in the * input array into floats. */ public static float[] unpackFloats(byte[] bytes) { checkNotNull(bytes, "bytes must not be null"); return unpackFloats(bytes, bytes.length / Float.BYTES); } /** * Unpack an array of bytes as an array of floats. * * @param bytes An array of bytes. * @param length The desired length of the resulting float array. The input will * be truncated or padded with zeros as needed. * @return an array of doubles obtained by marshalling consecutive bytes in the * input array into floats. */ public static float[] unpackFloats(byte[] bytes, int length) { checkNotNull(bytes, "bytes must not be null"); checkArgument(length >= 0, "length must be greater than or equal to 0"); checkArgument(bytes.length % Float.BYTES == 0, "bytes.length must be divisible by Float.BYTES"); ByteBuffer buf = ByteBuffer.wrap(bytes); float[] result = new float[length]; int m = Math.min(length, bytes.length / Float.BYTES); for (int i = 0; i < m; i++) { result[i] = buf.getFloat(); } return result; } }
471
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/util/Weighted.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import java.util.ArrayList; import java.util.List; import java.util.Random; /** * a container class that manages weights * * @param <Q> */ public class Weighted<Q> { public Q index; public float weight; public Weighted(Q object, float weight) { this.index = object; this.weight = weight; } /** * a generic MonteCarlo sampler that creates an Arraylist of WeightedIndexes * * @param input input list of weighted objects * @param seed random seed for repreoducibility * @param forceSampleFraction add the items which have weight over this fraction * @param scale scale that multiples the weights of the remainder. * Note that elements that are sampled are rescaled * to have ensured that the total weight (after * removal of heavy items) remains the same in * expectation * @param <Q> a generic index type, typically float[] in the * current usage * @return a randomly sampled arraylist (which can be the same list) of length * about LengthBound */ public static <Q> List<Weighted<Q>> createSample(List<Weighted<Q>> input, long seed, int lengthBound, double forceSampleFraction, double scale) { if (input.size() < lengthBound) { return input; } ArrayList<Weighted<Q>> samples = new ArrayList<>(); Random rng = new Random(seed); double totalWeight = input.stream().map(x -> (double) x.weight).reduce(Double::sum).get(); double remainder = totalWeight; if (forceSampleFraction > 0) { remainder = input.stream().map(e -> { if (e.weight > totalWeight * forceSampleFraction) { samples.add(new Weighted<>(e.index, e.weight)); return 0.0; } else { return (double) e.weight; } }).reduce(Double::sum).get(); } float factor = (float) (lengthBound * 1.0 / input.size()); float newScale = (float) (scale * (remainder / totalWeight) / factor); input.stream().forEach(e -> { if ((e.weight <= totalWeight * forceSampleFraction) && (rng.nextDouble() < factor)) { samples.add(new Weighted<>(e.index, e.weight * newScale)); } }); return samples; } /** * an utility routine to pick the element such that the prefix sum including * that element exceeds a weight (or is the last element) * * @param points a list of weighted objects * @param wt a parameter determining the cumulative weight * @return the position of the item satisfying the prefix condition or the last * element */ public static <Q> Weighted<Q> prefixPick(List<Weighted<Q>> points, double wt) { checkArgument(points.size() > 0, "cannot pick from an empty list"); double running = wt; Weighted<Q> saved = points.get(0); for (Weighted<Q> point : points) { if (running - point.weight <= 0.0) { return point; } running -= point.weight; saved = point; } return saved; } }
472
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/util/ArrayUtils.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import java.util.Arrays; /** * A utility class for data arrays. */ public class ArrayUtils { /** * Returns a clean deep copy of the point. Current clean-ups include changing * negative zero -0.0 to positive zero 0.0. * * @param point The original data point. * @return a clean deep copy of the original point. */ public static double[] cleanCopy(double[] point) { double[] pointCopy = Arrays.copyOf(point, point.length); for (int i = 0; i < point.length; i++) { if (pointCopy[i] == 0.0) { pointCopy[i] = 0.0; } } return pointCopy; } public static float[] cleanCopy(float[] point) { float[] pointCopy = Arrays.copyOf(point, point.length); for (int i = 0; i < point.length; i++) { if (pointCopy[i] == 0.0) { pointCopy[i] = 0.0f; } } return pointCopy; } }
473
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/util/ShingleBuilder.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.util; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.checkNotNull; /** * A utility class for creating shingled points, which are also referred to as * shingles. A shingle consists of multiple points appended together. If * individual points have n dimensions, and we include k points in a shingle, * then the shingle will have size n * m. * * There are two strategies for shingling: sliding and cyclic. In a sliding * shingle, new points are appended to the end of the shingle, and old points * are removed from the front. For example, if we have a shingle size of 4 which * currently contains the points a, b, c, and d, then we can represent the * shingle as abcd. The following schematic shows how the shingle is updated as * we add new points e and f. * * <pre> * abcd =&gt; bcde * bcde =&gt; cdef * </pre> * * With cycling shingling, when a new point is added to a shingle it overwrites * the oldest point in the shingle. Using the same setup as above, a cyclic * shingle would be updated as follows: * * <pre> * abcd =&gt; ebcd * ebcd =&gt; efcd * </pre> */ public class ShingleBuilder { /** * Number of dimensions of each point in the shingle. */ private final int dimensions; /** * Number of points in the shingle. */ private final int shingleSize; /** * A buffer containing points recently added to the shingle. */ private final double[][] recentPoints; /** * A flag indicating whether we should use a cyclic shift or a linear shift when * creating shingles. */ private final boolean cyclic; /** * The index where the next point will be copied to. This is equal to the index * of the oldest point currently in the shingle. */ private int shingleIndex; /** * A flag indicating whether the shingle has been completely filled once. */ private boolean full; /** * Create a new ShingleBuilder with the given dimensions and shingle size. * * @param dimensions The number of dimensions in the input points. * @param shingleSize The number of points to store in a shingle. * @param cyclic If true, the shingle will use cyclic updates. If false, it * will use sliding updates. */ public ShingleBuilder(int dimensions, int shingleSize, boolean cyclic) { checkArgument(dimensions > 0, "dimensions must be greater than 0"); checkArgument(shingleSize > 0, "shingleSize must be greater than 0"); this.dimensions = dimensions; this.shingleSize = shingleSize; this.cyclic = cyclic; recentPoints = new double[shingleSize][dimensions]; shingleIndex = 0; full = false; } /** * Create a ShingleBuilder with the given dimensions and shingleSize. The * resulting builder uses sliding updates. * * @param dimensions The number of dimensions in the input points. * @param shingleSize The number of points to store in a shingle. */ public ShingleBuilder(int dimensions, int shingleSize) { this(dimensions, shingleSize, false); } /** * @return true if the shingle has been completely filled once, false otherwise. */ public boolean isFull() { return full; } /** * @return the number of dimensions in input points. */ public int getInputPointSize() { return dimensions; } /** * @return the number of dimensions in a shingled point. */ public int getShingledPointSize() { return dimensions * shingleSize; } /** * @return true if this ShingleBuilder uses cyclic updates, false otherwise. */ public boolean isCyclic() { return cyclic; } /** * Return the index where the next input point will be stored in the internal * shingle buffer. If the ShingleBuilder uses cyclic updates, this value * indicates the current point in the cycle. * * @return the index where the next input point will be stored in the internal * shingle buffer. */ public int getShingleIndex() { return shingleIndex; } /** * Add a new point to this shingle. The point values are copied. * * @param point The new point to be added to the shingle. */ public void addPoint(double[] point) { checkNotNull(point, "point must not be null"); checkArgument(point.length == dimensions, String.format("point.length must equal %d", dimensions)); System.arraycopy(point, 0, recentPoints[shingleIndex], 0, dimensions); shingleIndex = (shingleIndex + 1) % shingleSize; if (!full && shingleIndex == 0) { full = true; } } /** * @return the current shingled point. */ public double[] getShingle() { double[] shingle = new double[shingleSize * dimensions]; getShingle(shingle); return shingle; } /** * Write the current shingled point into the supplied buffer. * * @param shingle A buffer where the shingled point will be written. */ public void getShingle(double[] shingle) { checkNotNull(shingle, "shingle must not be null"); checkArgument(shingle.length == dimensions * shingleSize, "shingle.length must be dimensions * shingleSize"); int beginIndex = cyclic ? 0 : shingleIndex; for (int i = 0; i < shingleSize; i++) { System.arraycopy(recentPoints[(beginIndex + i) % shingleSize], 0, shingle, i * dimensions, dimensions); } } }
474
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/Precision.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; /** * Options for floating-point precision. */ public enum Precision { /** * Single-precision (32 bit) floating point numbers. */ FLOAT_32, /** * Double-precision (64 bit) floating point numbers. */ FLOAT_64; }
475
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/TransformMethod.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; /** * Options for internally transforming data in RCF These are built for * convenience. Domain knowledge before feeding data into RCF(any tool) will * often have the best benefit! These apply to the basic data and not * timestamps, time is (hopefully) always moving forward and is measured shifted * (from a running mean), with an option of normalization. */ public enum TransformMethod { /** * the best transformation for data! */ NONE, /** * standard column normalization using fixed weights */ WEIGHTED, /** * subtract a moving average -- the average would be computed using the same * discount factor as the time decay of the RCF samplers. */ SUBTRACT_MA, /** * divide by standard deviation, after subtracting MA */ NORMALIZE, /** * difference from previous */ DIFFERENCE, /** * divide by standard deviation of difference, after differencing (again * subtract MA) */ NORMALIZE_DIFFERENCE; }
476
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/ImputationMethod.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; /** * Options for filling in missing values */ public enum ImputationMethod { /** * use all 0's */ ZERO, /** * use a fixed set of specified values (same as input dimension) */ FIXED_VALUES, /** * last known value in each input dimension */ PREVIOUS, /** * next seen value in each input dimension */ NEXT, /** * linear interpolation */ LINEAR, /** * use the RCF imputation; but would often require a minimum number of * observations and would use defaults (often LINEAR) till that point */ RCF; }
477
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/ScoringStrategy.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; /** * Options for using RCF, specially with thresholds */ public enum ScoringStrategy { /** * default behavior to be optimized; currently EXPECTED_INVERSE_DEPTH */ EXPECTED_INVERSE_DEPTH, /** * This is the same as STANDARD mode where the scoring function is switched to * distances between the vectors. Since RCFs build a multiresolution tree, and * in the aggregate, preserves distances to some approximation, this provides an * alternate anomaly detection mechanism which can be useful for shingleSize = 1 * and (dynamic) population analysis via RCFs. Specifically it switches the * scoring to be based on the distance computation in the Density Estimation * (interpolation). This allows for a direct comparison of clustering based * outlier detection and RCFs over numeric vectors. All transformations * available to the STANDARD mode in the ThresholdedRCF are available for this * mode as well; this does not affect RandomCutForest core in any way. For * timeseries analysis the STANDARD mode is recommended, but this does provide * another option in combination with the TransformMethods. */ DISTANCE, /** * RCFs are an updatable data structure that can support multiple difference * inference methods. Given the longstanding interest in ensembles of different * models, this strategy uses the multiple inference capabilities to increase * precision. It does not escape our attention that multi-mode allows the * functionality of multi-models yet use a significantly smaller state/memory * footprint since all the modes use RCF. The different modes are probed with * computational efficiency in mind. */ MULTI_MODE, /** * Same as above, except optimized for increasing recall. */ MULTI_MODE_RECALL; }
478
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/CorrectionMode.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; /** * Options for using RCF, specially with thresholds */ public enum CorrectionMode { /** * default behavior, no correction */ NONE, /** * due to transforms, or due to input noise */ NOISE, /** * elimination due to multi mode operation */ MULTI_MODE, /** * effect of an anomaly in shingle */ ANOMALY_IN_SHINGLE, /** * conditional forecast, using conditional fields */ CONDITIONAL_FORECAST, /** * forecasted value was not very different */ FORECAST, /** * data drifts and level shifts, will not be corrected unless level shifts are * turned on */ DATA_DRIFT }
479
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/Config.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; public class Config { public static final String BOUNDING_BOX_CACHE_FRACTION = "bounding_box_cache_fraction"; public static final String TIME_DECAY = "time_decay"; }
480
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/ForestMode.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; /** * Options for using RCF, specially with thresholds */ public enum ForestMode { /** * a standard mode that uses shingling and most known applications; it uses the * last K data points where K=1 would correspond to non time series (population) * analysis */ STANDARD, /** * time stamp is added automatically to data to correlate within RCF itself; * this is useful for event streaams and for modeling sparse events. Option is * provided to normalize the time gaps. */ TIME_AUGMENTED, /** * uses various Fill-In strageies for data with gaps but not really sparse. Must * have shingleSize greater than 1, typically larger shingle size is better, and * so is fewer input dimensions */ STREAMING_IMPUTE; }
481
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/config/IDynamicConfig.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.config; /** * This interface is used by model classes to configure model parameters by * name. This is intended primarily for settings that a user may want to change * at runtime. */ public interface IDynamicConfig { <T> void setConfig(String name, T value, Class<T> clazz); default void setConfig(String name, short value) { setConfig(name, value, Short.class); } default void setConfig(String name, int value) { setConfig(name, value, Integer.class); } default void setConfig(String name, long value) { setConfig(name, value, Long.class); } default void setConfig(String name, float value) { setConfig(name, value, Float.class); } default void setConfig(String name, double value) { setConfig(name, value, Double.class); } default void setConfig(String name, boolean value) { setConfig(name, value, Boolean.class); } <T> T getConfig(String name, Class<T> clazz); default Object getConfig(String name) { return getConfig(name, Object.class); } }
482
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/summarization/Center.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.summarization; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static java.lang.Math.exp; import java.util.ArrayList; import java.util.Arrays; import java.util.List; import java.util.function.BiFunction; import java.util.function.Function; import com.amazon.randomcutforest.util.Weighted; /** * the following class abstracts a single centroid representation of a group of * points */ public class Center implements ICluster<float[]> { float[] representative; double weight; ArrayList<Weighted<Integer>> assignedPoints; double sumOfRadius; double previousWeight = 0; double previousSumOFRadius = 0; Center(float[] coordinate, float weight) { // explicitly copied because array elements will change this.representative = Arrays.copyOf(coordinate, coordinate.length); this.weight = weight; this.assignedPoints = new ArrayList<>(); } public static Center initialize(float[] coordinate, float weight) { return new Center(coordinate, weight); } // adds a point; only the index to keep space bounds lower // note that the weight may not be the entire weight of a point in case of a // "soft" assignment public void addPoint(int index, float weight, double dist, float[] point, BiFunction<float[], float[], Double> distance) { assignedPoints.add(new Weighted<>(index, weight)); this.weight += weight; this.sumOfRadius += weight * dist; } // the following sets up reassignment of the coordinate based on the points // assigned to the center public void reset() { assignedPoints = new ArrayList<>(); previousWeight = weight; weight = 0; previousSumOFRadius = sumOfRadius; } public double averageRadius() { return (weight > 0) ? sumOfRadius / weight : 0; } // average radius computation, provides an extent measure public double extentMeasure() { return (weight > 0) ? sumOfRadius / weight : 0; } public double getWeight() { return weight; } // a standard reassignment using the median values and NOT the mean; the mean is // unlikely to // provide robust convergence public double recompute(Function<Integer, float[]> getPoint, boolean approx, BiFunction<float[], float[], Double> distance) { if (assignedPoints.size() == 0 || weight == 0.0) { Arrays.fill(representative, 0); // zero out values return 0; } previousSumOFRadius = sumOfRadius; sumOfRadius = 0; for (int i = 0; i < representative.length; i++) { int index = i; // the following would be significantly slow unless points are backed by arrays assignedPoints .sort((o1, o2) -> Double.compare(getPoint.apply(o1.index)[index], getPoint.apply(o2.index)[index])); double runningWeight = weight / 2; int position = 0; while (runningWeight >= 0 && position < assignedPoints.size()) { if (runningWeight > assignedPoints.get(position).weight) { runningWeight -= assignedPoints.get(position).weight; ++position; } else { break; } } if (position == assignedPoints.size()) { position--; } representative[index] = getPoint.apply(assignedPoints.get(position).index)[index]; } for (int j = 0; j < assignedPoints.size(); j++) { double addTerm = distance.apply(representative, getPoint.apply(assignedPoints.get(j).index)) * assignedPoints.get(j).weight; checkArgument(addTerm >= 0, "distances or weights cannot be negative"); sumOfRadius += addTerm; } return (previousSumOFRadius - sumOfRadius); } @Override public List<Weighted<Integer>> getAssignedPoints() { return assignedPoints; } // merges a center into another // this can be followed by a reassignment step; however the merger uses a // sigmoid based weightage // for robustness public void absorb(ICluster<float[]> other, BiFunction<float[], float[], Double> distance) { List<Weighted<float[]>> representatives = other.getRepresentatives(); float[] closest = representatives.get(0).index; double dist = Double.MAX_VALUE; for (Weighted<float[]> e : representatives) { double t = distance.apply(e.index, representative); checkArgument(t >= 0, "distances cannot be negative"); if (t < dist) { dist = t; closest = e.index; } } double otherWeight = other.getWeight(); double expRatio = exp(2 * (weight - otherWeight) / (weight + otherWeight)); double factor = expRatio / (1.0 + expRatio); for (int i = 0; i < representative.length; i++) { representative[i] = (float) (factor * representative[i] + (1 - factor) * closest[i]); } // distance is (approximately) the reverse of the ratio // this computation is meant to be approximate sumOfRadius += (weight * (1.0 - factor) + otherWeight * factor) * dist; weight += otherWeight; assignedPoints.addAll(other.getAssignedPoints()); other.reset(); } public double distance(float[] point, BiFunction<float[], float[], Double> distance) { double t = distance.apply(point, representative); checkArgument(t >= 0, "distance cannot be negative"); return t; } @Override public double distance(ICluster<float[]> other, BiFunction<float[], float[], Double> distance) { return other.distance(representative, distance); } @Override public float[] primaryRepresentative(BiFunction<float[], float[], Double> distance) { return Arrays.copyOf(representative, representative.length); } @Override public List<Weighted<float[]>> getRepresentatives() { ArrayList<Weighted<float[]>> answer = new ArrayList<>(); answer.add(new Weighted<>(representative, (float) weight)); return answer; } }
483
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/summarization/GenericMultiCenter.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.summarization; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static java.lang.Math.min; import java.util.ArrayList; import java.util.List; import java.util.function.BiFunction; import java.util.function.Function; import com.amazon.randomcutforest.util.Weighted; /** * the following class abstracts a single centroid representation of a group of * points. The class is modeled after the well scattered representatives used in * CURE https://en.wikipedia.org/wiki/CURE_algorithm * * The number of representatives (refered as c in above) determines the possible * shapes that can be represented. Setting c=1 corresponds to stnadard centroid * based clustering * * The parameter shrinkage is slightly different from its usage in CURE, * although the idea of its use is similar. The main reason is that CURE was * designed for geometric spaces, and RCFSummarize is designed to support * arbitrary distance based clustering; once the user provides a distance * function from (R, R) into double based on ideas of STREAM * https://en.wikipedia.org/wiki/Data_stream_clustering In CURE, shrinkage was * used to create representatives close to the center of a cluster which is * impossible for generic types R. Instead shrinkage value in [0,1] corresponds * to morphing the distance function to "pretend" as if the distance is to the * primary representative of the cluster. * * This generic version does not store any assigned points. As a result the size * is bounded and these clusters are ideal for streaming algorithms where * resource usage would not increase with more data. */ public class GenericMultiCenter<R> implements ICluster<R> { public static int DEFAULT_NUMBER_OF_REPRESENTATIVES = 5; public static double DEFAULT_SHRINKAGE = 0.0; int numberOfRepresentatives = DEFAULT_NUMBER_OF_REPRESENTATIVES; double shrinkage = DEFAULT_SHRINKAGE; ArrayList<Weighted<R>> representatives; double weight; double sumOfRadius; double previousWeight = 0; double previousSumOFRadius = 0; GenericMultiCenter(R coordinate, float weight, double shrinkage, int numberOfRepresentatives) { // explicitly copied because array elements will change this.representatives = new ArrayList<>(); this.representatives.add(new Weighted<>(coordinate, weight)); this.weight = weight; this.numberOfRepresentatives = numberOfRepresentatives; this.shrinkage = shrinkage; } public static <R> GenericMultiCenter<R> initialize(R coordinate, float weight, double shrinkage, int numberOfRepresentatives) { checkArgument(shrinkage >= 0 && shrinkage <= 1.0, " parameter has to be in [0,1]"); checkArgument(numberOfRepresentatives > 0 && numberOfRepresentatives <= 100, " the number of representatives has to be in (0,100]"); return new GenericMultiCenter<>(coordinate, weight, shrinkage, numberOfRepresentatives); } // adds a point; only the index to keep space bounds lower // note that the weight may not be the entire weight of a point in case of a // "soft" assignment public void addPoint(int index, float weight, double dist, R point, BiFunction<R, R, Double> distance) { // accounting for the closest representative, if there are more than one Weighted<R> closest = representatives.get(0); if (representatives.size() > 1) { double newDist = distance.apply(point, representatives.get(0).index); for (int i = 1; i < representatives.size(); i++) { double t = distance.apply(point, representatives.get(i).index); if (t < newDist) { newDist = t; closest = representatives.get(i); } } } closest.weight += weight; this.weight += weight; this.sumOfRadius += weight * dist; } // the following sets up reassignment of the coordinate based on the points // assigned to the center public void reset() { previousWeight = weight; weight = 0; for (int i = 0; i < representatives.size(); i++) { representatives.get(i).weight = 0; } previousSumOFRadius = sumOfRadius; sumOfRadius = 0; } public double averageRadius() { return (weight > 0) ? sumOfRadius / weight : 0; } // forces a nearest neighbor merge public double extentMeasure() { return (weight > 0) ? 0.5 * sumOfRadius / (numberOfRepresentatives * weight) : 0; } public double getWeight() { return weight; } // reassignment may not be meaningful for generic types, without additional // information public double recompute(Function<Integer, R> getPoint, boolean flag, BiFunction<R, R, Double> distanceFunction) { return 0; } // merges a center into another public void absorb(ICluster<R> other, BiFunction<R, R, Double> distance) { List<Weighted<R>> savedRepresentatives = this.representatives; savedRepresentatives.addAll(other.getRepresentatives()); this.representatives = new ArrayList<>(); int maxIndex = 0; float weight = savedRepresentatives.get(0).weight; for (int i = 1; i < savedRepresentatives.size(); i++) { if (weight < savedRepresentatives.get(i).weight) { weight = savedRepresentatives.get(i).weight; maxIndex = i; } } this.representatives.add(savedRepresentatives.get(maxIndex)); savedRepresentatives.remove(maxIndex); sumOfRadius += other.extentMeasure() * other.getWeight(); this.weight += other.getWeight(); /** * create a list of representatives based on the farthest point method, which * correspond to a well scattered set. See * https://en.wikipedia.org/wiki/CURE_algorithm */ while (savedRepresentatives.size() > 0 && this.representatives.size() < numberOfRepresentatives) { double farthestWeightedDistance = 0.0; int farthestIndex = Integer.MAX_VALUE; for (int j = 0; j < savedRepresentatives.size(); j++) { if (savedRepresentatives.get(j).weight > weight / (2 * numberOfRepresentatives)) { double newWeightedDist = distance.apply(this.representatives.get(0).index, savedRepresentatives.get(j).index) * savedRepresentatives.get(j).weight; checkArgument(newWeightedDist >= 0, " weights or distances cannot be negative"); for (int i = 1; i < this.representatives.size(); i++) { newWeightedDist = min(newWeightedDist, distance.apply(this.representatives.get(i).index, savedRepresentatives.get(j).index)) * savedRepresentatives.get(j).weight; checkArgument(newWeightedDist >= 0, " weights or distances cannot be negative"); } if (newWeightedDist > farthestWeightedDistance) { farthestWeightedDistance = newWeightedDist; farthestIndex = j; } } } if (farthestWeightedDistance == 0.0) { break; } this.representatives.add(savedRepresentatives.get(farthestIndex)); savedRepresentatives.remove(farthestIndex); } // absorb the remainder into existing representatives for (Weighted<R> representative : savedRepresentatives) { double dist = distance.apply(representative.index, this.representatives.get(0).index); checkArgument(dist >= 0, "distance cannot be negative"); double minDist = dist; int minIndex = 0; for (int i = 1; i < this.representatives.size(); i++) { double newDist = distance.apply(this.representatives.get(i).index, representative.index); checkArgument(newDist >= 0, "distance cannot be negative"); if (newDist < minDist) { minDist = newDist; minIndex = i; } } this.representatives.get(minIndex).weight += representative.weight; sumOfRadius += representative.weight * ((1 - shrinkage) * minDist + dist * shrinkage); } } @Override public double distance(R point, BiFunction<R, R, Double> distanceFunction) { double dist = distanceFunction.apply(this.representatives.get(0).index, point); checkArgument(dist >= 0, "distance cannot be negative"); double newDist = dist; for (int i = 1; i < this.representatives.size(); i++) { newDist = min(newDist, distanceFunction.apply(this.representatives.get(i).index, point)); checkArgument(newDist >= 0, "distance cannot be negative"); } return (1 - shrinkage) * newDist + shrinkage * dist; } @Override public double distance(ICluster<R> other, BiFunction<R, R, Double> distanceFunction) { List<Weighted<R>> representatives = other.getRepresentatives(); double dist = distanceFunction.apply(this.representatives.get(0).index, representatives.get(0).index); checkArgument(dist >= 0, "distance cannot be negative"); double newDist = dist; for (int i = 1; i < this.representatives.size(); i++) { for (int j = 1; j < representatives.size(); j++) { newDist = min(newDist, distanceFunction.apply(this.representatives.get(i).index, representatives.get(j).index)); checkArgument(newDist >= 0, "distance cannot be negative"); } } return (1 - shrinkage) * newDist + shrinkage * dist; } @Override public List<Weighted<R>> getRepresentatives() { return representatives; } }
484
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/summarization/MultiCenter.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.summarization; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import java.util.ArrayList; import java.util.List; import java.util.function.BiFunction; import java.util.function.Function; import com.amazon.randomcutforest.util.Weighted; public class MultiCenter extends GenericMultiCenter<float[]> { ArrayList<Weighted<Integer>> assignedPoints; MultiCenter(float[] coordinate, float weight, double shrinkage, int numberOfRepresentatives) { super(coordinate, weight, shrinkage, numberOfRepresentatives); this.assignedPoints = new ArrayList<>(); } public static MultiCenter initialize(float[] coordinate, float weight, double shrinkage, int numberOfRepresentatives) { checkArgument(shrinkage >= 0 && shrinkage <= 1.0, " parameter has to be in [0,1]"); checkArgument(numberOfRepresentatives > 0 && numberOfRepresentatives <= 100, " the number of representatives has to be in (0,100]"); return new MultiCenter(coordinate, weight, shrinkage, numberOfRepresentatives); } public void addPoint(int index, float weight, double dist, float[] point, BiFunction<float[], float[], Double> distance) { super.addPoint(index, weight, dist, point, distance); assignedPoints.add(new Weighted<>(index, weight)); } // the following sets up reassignment of the coordinate based on the points // assigned to the center public void reset() { super.reset(); assignedPoints = new ArrayList<>(); } // a standard reassignment using the median values and NOT the mean; the mean is // unlikely to // provide robust convergence public double recompute(Function<Integer, float[]> getPoint, boolean force, BiFunction<float[], float[], Double> distanceFunction) { if (assignedPoints.size() == 0 || weight == 0.0 || !force) { return 0; } previousSumOFRadius = sumOfRadius; sumOfRadius = 0; for (int j = 0; j < assignedPoints.size(); j++) { // distance will check for -negative internally double addTerm = distance(getPoint.apply(assignedPoints.get(j).index), distanceFunction) * assignedPoints.get(j).weight; sumOfRadius += addTerm; } return (previousSumOFRadius - sumOfRadius); } @Override public List<Weighted<Integer>> getAssignedPoints() { return assignedPoints; } }
485
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/summarization/ICluster.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.summarization; import java.util.Collections; import java.util.List; import java.util.function.BiFunction; import java.util.function.Function; import com.amazon.randomcutforest.util.Weighted; /** * a set of cunstions that a conceptual "cluster" should satisfy for any generic * distance based clustering where a distance function of type from (R,R) into * double is provided externally. It is not feasible (short of various * assumptions) to check for the validity of a distance function and the * clustering would not perform any validity checks. The user is referred to * https://en.wikipedia.org/wiki/Metric_(mathematics) * * It does not escape our attention that the clustering can use multiple * different distance functions over its execution. But such should be performed * with caution. */ public interface ICluster<R> { // restting statistics for a potential reassignment void reset(); double averageRadius(); // a measure of the noise/blur around a cluster; for single centroid clustering // this is the average distance of a point from a cluster representative double extentMeasure(); // weight computation double getWeight(); // merge another cluster of same type void absorb(ICluster<R> other, BiFunction<R, R, Double> distance); // distance of apoint from a cluster, has to be non-negative double distance(R point, BiFunction<R, R, Double> distance); // distance of another cluster from this cluster, has to be non negative double distance(ICluster<R> other, BiFunction<R, R, Double> distance); // all potential representativess of a cluster these are typically chosen to be // well scattered // by default the first entry is the primary representative List<Weighted<R>> getRepresentatives(); // a primary representative of the cluster; by default it is the first in the // list of representatives // this additional function allows an option for optimization of runtime as well // as alternate // representations. For example the distance metric can be altered to be a fixed // linear combination // of the primary and secondary representatives, as in CURE // https://en.wikipedia.org/wiki/CURE_algorithm default R primaryRepresentative(BiFunction<R, R, Double> distance) { return getRepresentatives().get(0).index; } // Some of the algorithms, in particular the geometric ones may store the // assigned points for // iterative refinement. However that can be extremely inefficient if the // distance measure does not // have sufficient range, for example, string edit distances (for bounded // strings) are bounded in a // short interval. A soft assignment would create multiple copies of points (as // is appropriate) and // that can be significantly slower. default List<Weighted<Integer>> getAssignedPoints() { return Collections.emptyList(); } // optimize the cluster representation based on assigned points; this is classic // iterative optimization // useful in EM type algorithms /** * optimize the cluster representation based on assigned points; this is classic * iterative optimization useful in EM type algorithms * * @param getPoint a function that provides a point given an integer index * @param force it set as true perform a slow and accurate recomputation; * otherwise approximation would suffice * @param distance the distance function * @return a measure of improvement (if any); this can be useful in the future * as a part of the stopping condition */ double recompute(Function<Integer, R> getPoint, boolean force, BiFunction<R, R, Double> distance); // adding a point to a cluster, and possibly updates the extent measure and the // assigned points void addPoint(int index, float weight, double dist, R point, BiFunction<R, R, Double> distance); }
486
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/summarization/Summarizer.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.summarization; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.util.Weighted.createSample; import static com.amazon.randomcutforest.util.Weighted.prefixPick; import static java.lang.Math.max; import java.util.ArrayList; import java.util.Arrays; import java.util.Comparator; import java.util.List; import java.util.Random; import java.util.function.BiFunction; import java.util.function.Function; import com.amazon.randomcutforest.returntypes.SampleSummary; import com.amazon.randomcutforest.util.Weighted; public class Summarizer { /** * a factor that controls weight assignment for soft clustering; this is the * multiple of the minimum distance and should be greater or equal 1. */ public static double WEIGHT_ALLOCATION_THRESHOLD = 1.25; /** * the following determines the ratio between the sum of the (average) radius * and the separation between centers for a merge; ratio greater than 1 means * significant overlap a ratio of 0 means merge closest pairs without * consideration of separartion * **/ public static double DEFAULT_SEPARATION_RATIO_FOR_MERGE = 0.8; public static int PHASE2_THRESHOLD = 2; public static int LENGTH_BOUND = 1000; public static Double L1distance(float[] a, float[] b) { double dist = 0; for (int i = 0; i < a.length; i++) { dist += Math.abs(a[i] - b[i]); } return dist; } public static Double L2distance(float[] a, float[] b) { double dist = 0; for (int i = 0; i < a.length; i++) { double t = Math.abs(a[i] - b[i]); dist += t * t; } return Math.sqrt(dist); } public static Double LInfinitydistance(float[] a, float[] b) { double dist = 0; for (int i = 0; i < a.length; i++) { dist = max(Math.abs(a[i] - b[i]), dist); } return dist; } /** * a function that reassigns points to clusters * * @param sampledPoints a list of sampled points with weights * @param clusters a list of current clusters, because random access to * the elements is necessary * @param distance a distance function * @param parallelEnabled a flag enabling limited parallelism; only during * cluster by cluster recomputation. Using parallel mode * during the assignment of points does not seem to help */ public static <R> void assignAndRecompute(List<Weighted<Integer>> sampledPoints, Function<Integer, R> getPoint, List<ICluster<R>> clusters, BiFunction<R, R, Double> distance, boolean parallelEnabled) { checkArgument(clusters.size() > 0, " cannot be empty list of clusters"); checkArgument(sampledPoints.size() > 0, " cannot be empty list of points"); for (ICluster<R> cluster : clusters) { cluster.reset(); } for (Weighted<Integer> point : sampledPoints) { if (point.weight > 0) { double[] dist = new double[clusters.size()]; Arrays.fill(dist, Double.MAX_VALUE); double minDist = Double.MAX_VALUE; int minDistNbr = -1; for (int i = 0; i < clusters.size(); i++) { // will check for negative distances dist[i] = clusters.get(i).distance(getPoint.apply(point.index), distance); if (minDist > dist[i]) { minDist = dist[i]; minDistNbr = i; } if (minDist == 0) { break; } } if (minDist == 0) { clusters.get(minDistNbr).addPoint(point.index, point.weight, 0, getPoint.apply(point.index), distance); } else { double sum = 0; for (int i = 0; i < clusters.size(); i++) { if (dist[i] <= WEIGHT_ALLOCATION_THRESHOLD * minDist) { sum += minDist / dist[i]; // setting up harmonic mean } } for (int i = 0; i < clusters.size(); i++) { if (dist[i] <= WEIGHT_ALLOCATION_THRESHOLD * minDist) { // harmonic mean clusters.get(i).addPoint(point.index, (float) (point.weight * minDist / (dist[i] * sum)), dist[i], getPoint.apply(point.index), distance); } } } } } if (parallelEnabled) { clusters.parallelStream().forEach(e -> e.recompute(getPoint, true, distance)); } else { clusters.stream().forEach(e -> e.recompute(getPoint, true, distance)); } } /** * The core subroutine for iterative clustering used herein. The clustering * algorithm borrows from CURE https://en.wikipedia.org/wiki/CURE_algorithm, * which used sampling as a tradeoff of representationa accuracy versus * algorithmic efficiency. Note however random sampling can also perform * denoising and reduce space as a filtering mechanism. Note that hierarchical * iterative merging strategies can be proven to not degrade clustering * https://en.wikipedia.org/wiki/Data_stream_clustering with the benefit of * small space. The algorithm herein proceeds in three phases, where the first * phase corresponds from the initial seeding to about twice the maximum number * of clusters one wishes to consider. The second phase corresponds to reducing * that number to the maximum allowable number. The third phase corresponds to * continuing the clustering as long as the conditions are similar to the end of * phase two, thereby enabling us to use a rough estimate for the maximum * allowed. By default, recomputation of the cluster makes sense in phases 2 and * 3 -- however can be enabled for phase 1 as well, thereby enabling the regular * K-Means algorithm to be expressed in the below algorithm as well. The * algorithm can also express Minimum Spanning Tree based clustering with * repeated merging of closest pair (which is a capability derived from CURE) * * The primary reason for the number of parameters is the ability to invoke this * function without creating a copy of the points (or equivalent large objects), * and hence the functions as parameters * * @param maxAllowed number of maximum clusters one is interested in * @param initial an initial number of sampled centers to start * from * @param stopAt a hard lower bound on the number of clusters * @param refs a (possibly sampled) list of references with * weight * @param getPoint a function which retrives the point/object given * an index in the refs * @param distance a distance function * @param clusterInitializer a function that creates a cluster given an object * aand a weight * @param seed a random seed * @param parallelEnabled enabling parallel computation in the first phase * when points are assigned to different sampled * centers; and the centers are possibly adjusted * @param phase2GlobalReassign a flag that determines if the points would be * reassigned when the clusters fall below 1.2 * * maxAllowed -- this serves as a denoising. * @param overlapParameter a parameter that controls the ordering of the * merges as well as the stopping condition of the * merges * @param previousClustering a possibly null list of clusters seen previously, * used as zero weight seeds to smoothen the * continuous clustering * @param <R> type of object being clustered * @return a list of clusters */ public static <R> List<ICluster<R>> iterativeClustering(int maxAllowed, int initial, int stopAt, List<Weighted<Integer>> refs, Function<Integer, R> getPoint, BiFunction<R, R, Double> distance, BiFunction<R, Float, ICluster<R>> clusterInitializer, long seed, boolean parallelEnabled, boolean phase2GlobalReassign, double overlapParameter, List<ICluster<R>> previousClustering) { checkArgument(refs.size() > 0, "empty list, nothing to do"); checkArgument(stopAt > 0, "has to stop at 1 cluster"); checkArgument(stopAt <= maxAllowed, "cannot stop before achieving the limit"); Random rng = new Random(seed); double sampledSum = refs.stream().map(e -> { checkArgument(Double.isFinite(e.weight), " weights have to be finite"); checkArgument(e.weight >= 0.0, () -> "negative weights are not meaningful" + e.weight); return (double) e.weight; }).reduce(0.0, Double::sum); checkArgument(sampledSum > 0, " total weight has to be positive"); ArrayList<ICluster<R>> centers = new ArrayList<>(); if (refs.size() < 10 * (initial + 5)) { for (Weighted<Integer> point : refs) { centers.add(clusterInitializer.apply(getPoint.apply(point.index), 0f)); } } else { for (int k = 0; k < 2 * (initial + 5); k++) { double wt = rng.nextDouble() * sampledSum; Weighted<Integer> picked = prefixPick(refs, wt); centers.add(clusterInitializer.apply(getPoint.apply(picked.index), 0f)); } } if (previousClustering != null) { for (ICluster<R> previousCluster : previousClustering) { List<Weighted<R>> representatives = previousCluster.getRepresentatives(); for (Weighted<R> representative : representatives) { centers.add(clusterInitializer.apply(representative.index, 0f)); } } } assignAndRecompute(refs, getPoint, centers, distance, parallelEnabled); // assignment would change weights, sorting in non-decreasing order centers.sort(Comparator.comparingDouble(ICluster::getWeight)); while (centers.get(0).getWeight() == 0) { centers.remove(0); } double phase3Distance = 0; double runningPhase3Distance = 0; boolean keepReducingCenters = (centers.size() > maxAllowed); while (keepReducingCenters) { double measure = 0; double measureDist = Double.MAX_VALUE; int lower = 0; int firstOfMerge = lower; int secondOfMerge = lower + 1;// will be reset before exiting the loop boolean foundMerge = false; double minDist = Double.MAX_VALUE; while (lower < centers.size() - 1 && !foundMerge) { // we will keep searching int minNbr = -1; for (int j = lower + 1; j < centers.size(); j++) { double dist = centers.get(lower).distance(centers.get(j), distance); if (dist == 0) { foundMerge = true; firstOfMerge = lower; secondOfMerge = minNbr = j; minDist = measureDist = 0.0; break; } else { if (minDist > dist) { minNbr = j; minDist = dist; } double temp = (centers.get(lower).extentMeasure() + centers.get(j).extentMeasure() + phase3Distance) / dist; if (temp > overlapParameter && measure < temp) { firstOfMerge = lower; secondOfMerge = j; measure = temp; measureDist = dist; } } } if (lower == 0 && !foundMerge) { measureDist = minDist; // this is set assuming we may be interested in merging the minimum weight // cluster which corresponds to lower == 0 secondOfMerge = minNbr; } ++lower; } int inital = centers.size(); if (inital > maxAllowed || foundMerge || (inital > stopAt && measure > overlapParameter)) { centers.get(secondOfMerge).absorb(centers.get(firstOfMerge), distance); if (phase2GlobalReassign && centers.size() <= PHASE2_THRESHOLD * maxAllowed + 1) { centers.remove(firstOfMerge); assignAndRecompute(refs, getPoint, centers, distance, parallelEnabled); } else { centers.get(secondOfMerge).recompute(getPoint, false, distance); centers.remove(firstOfMerge); } centers.sort(Comparator.comparingDouble(ICluster::getWeight)); while (centers.get(0).getWeight() == 0.0) { // this line is reachable via zeroTest() in // SampleSummaryTest centers.remove(0); } if (inital < 1.2 * maxAllowed + 1) { // phase 3 kicks in; but this will execute at most once // note that measureDist can be 0 as well runningPhase3Distance = max(runningPhase3Distance, measureDist); if (inital > maxAllowed && centers.size() <= maxAllowed) { phase3Distance = runningPhase3Distance; } } } else { keepReducingCenters = false; } } // sort in decreasing weight centers.sort((o1, o2) -> Double.compare(o2.getWeight(), o1.getWeight())); return centers; } /** * the following function returns a summary of the input points * * @param points points with associated weights * @param maxAllowed the maximum number of clusters/summary points * @param initial the initial number of clusters/summary points, * chosen at random * @param stopAt a hard lower bound on the number of clusters * @param phase2GlobalReassign a flag that performs global reassignments when * the number of clusters is in the range * [maxAllowed, ceil(1.2*maxAllowed)] * @param overlapParameter a control for merging clusters * @param distance a distance function for the points, that * determines the order of the reverse delete * however the EM like step uses L1 measure (to be * robust to noise) * @param clusterInitializer a function that creates the cluster type given a * single object and a weight * @param seed a random seed for controlling the randomness * @param parallelEnabled flag enabling (limited) parallelism * @param previousClustering any previous clustering that can be used as zero * weight seeds to ensure smoothness * @return a clustering of the input points (Note: the median returned is an * approximate median; exact computation is unlikely to be critical for * true applications of summarization) */ public static <R> List<ICluster<R>> summarize(List<Weighted<R>> points, int maxAllowed, int initial, int stopAt, boolean phase2GlobalReassign, double overlapParameter, BiFunction<R, R, Double> distance, BiFunction<R, Float, ICluster<R>> clusterInitializer, long seed, boolean parallelEnabled, List<ICluster<R>> previousClustering) { checkArgument(maxAllowed < 100, "are you sure you want more elements in the summary?"); checkArgument(maxAllowed <= initial, "initial parameter should be at least maximum allowed in final result"); double totalWeight = points.stream().map(e -> { checkArgument(Double.isFinite(e.weight), " weights have to be finite"); checkArgument(e.weight >= 0.0, () -> "negative weights are not meaningful" + e.weight); return (double) e.weight; }).reduce(0.0, Double::sum); checkArgument(totalWeight > 0, " total weight has to be positive"); Random rng = new Random(seed); // the following list is explicity copied and sorted for potential efficiency List<Weighted<R>> sampledPoints = createSample(points, rng.nextLong(), 5 * LENGTH_BOUND, 0.005, 1.0); ArrayList<Weighted<Integer>> refs = new ArrayList<>(); for (int i = 0; i < sampledPoints.size(); i++) { refs.add(new Weighted<>(i, sampledPoints.get(i).weight)); } Function<Integer, R> getPoint = (i) -> sampledPoints.get(i).index; return iterativeClustering(maxAllowed, initial, stopAt, refs, getPoint, distance, clusterInitializer, rng.nextLong(), parallelEnabled, phase2GlobalReassign, overlapParameter, previousClustering); } // same as above, specific for single centroid clustering of float[] // with an explicit stopping condition as well as a global reassignment option public static List<ICluster<float[]>> singleCentroidSummarize(List<Weighted<float[]>> points, int maxAllowed, int initial, int stopAt, boolean phase2GlobalReassign, BiFunction<float[], float[], Double> distance, long seed, boolean parallelEnabled, List<ICluster<float[]>> previousClustering) { return summarize(points, maxAllowed, initial, stopAt, phase2GlobalReassign, DEFAULT_SEPARATION_RATIO_FOR_MERGE, distance, Center::initialize, seed, parallelEnabled, previousClustering); } /** * the following function returns a summary of the input points * * @param points points with associated weights * @param maxAllowed the maximum number of clusters/summary points * @param initial the initial number of clusters/summary points, chosen * at random * @param phase1reassign (this parameter is ignored in the current version, but * the signature is unchanged for convenience) * @param distance a distance function for the points, that determines * the order of the reverse delete however the EM like * step uses L1 measure (to be robust to noise) * @param seed a random seed for controlling the randomness * @param parallelEnabled flag enabling (limited) parallelism * @return a summary of the input points (Note: the median returned is an * approximate median; exact computation is unlikely to be critical for * true applications of summarization) */ public static SampleSummary summarize(List<Weighted<float[]>> points, int maxAllowed, int initial, boolean phase1reassign, BiFunction<float[], float[], Double> distance, long seed, boolean parallelEnabled) { checkArgument(maxAllowed < 100, "are you sure you want more elements in the summary?"); checkArgument(maxAllowed <= initial, "initial parameter should be at least maximum allowed in final result"); double totalWeight = points.stream().map(e -> { checkArgument(Double.isFinite(e.weight), " weights have to be finite"); checkArgument(e.weight >= 0.0, () -> "negative weights are not meaningful" + e.weight); return (double) e.weight; }).reduce(0.0, Double::sum); checkArgument(totalWeight > 0, " total weight has to be positive"); Random rng = new Random(seed); // the following list is explicity copied and sorted for potential efficiency List<Weighted<float[]>> sampledPoints = createSample(points, rng.nextLong(), 5 * LENGTH_BOUND, 0.005, 1.0); List<ICluster<float[]>> centers = summarize(sampledPoints, maxAllowed, initial, 1, true, DEFAULT_SEPARATION_RATIO_FOR_MERGE, distance, Center::initialize, seed, parallelEnabled, null); float[][] pointList = new float[centers.size()][]; float[] likelihood = new float[centers.size()]; int dimensions = centers.get(0).primaryRepresentative(distance).length; for (int i = 0; i < centers.size(); i++) { pointList[i] = Arrays.copyOf(centers.get(i).primaryRepresentative(distance), dimensions); likelihood[i] = (float) (centers.get(i).getWeight() / totalWeight); } return new SampleSummary(sampledPoints, pointList, likelihood); } /** * Same as previous over a flat collection of unweighted float[] * * @param points points represented by float[][] * @param maxAllowed maximum number of clusters in output * @param initial initial number of points to seed; a control parameter * that serves both as a denoiser, as well as as a * facilitator of coninuity (large numbers would * correspond to MST like clustering) * @param reassignPerStep unusued in current version * @param distance distance metric over float [] * @param seed random seed * @param parallelEnabled flag enabling (limited) parallelism * @return a list of centers with weights */ public static SampleSummary summarize(float[][] points, int maxAllowed, int initial, boolean reassignPerStep, BiFunction<float[], float[], Double> distance, long seed, Boolean parallelEnabled) { ArrayList<Weighted<float[]>> weighted = new ArrayList<>(); for (float[] point : points) { weighted.add(new Weighted<>(point, 1.0f)); } return summarize(weighted, maxAllowed, initial, reassignPerStep, distance, seed, parallelEnabled); } /** * same as before with common cases filled in, used in analysis of * ConditionalSamples * * @param points points in ProjectedPoint{} * @param maxAllowed maximum number of groups/clusters * @param initial a parameter controlling the initialization * @param reassignPerStep if reassignment is to be performed each step * @param seed random seed * @return a summarization */ public static SampleSummary l2summarize(List<Weighted<float[]>> points, int maxAllowed, int initial, boolean reassignPerStep, long seed) { return summarize(points, maxAllowed, initial, reassignPerStep, Summarizer::L2distance, seed, false); } /** * Same as above, with the most common use cases filled in * * @param points points in float[][], each of weight 1.0 * @param maxAllowed maximum number of clusters one is interested in * @param seed random seed * @return a summarization */ public static SampleSummary l2summarize(float[][] points, int maxAllowed, long seed) { return summarize(points, maxAllowed, 4 * maxAllowed, false, Summarizer::L2distance, seed, false); } /** * * @param points points represented by R[] * @param maxAllowed maximum number of clusters in output * @param initial initial number of points to seed; a control * parameter that serves both as a denoiser, as * well as as a facilitator of coninuity (large * numbers would correspond to MST like * clustering) * @param phase2GlobalReassign a boolean determining global reassignment * @param overlapParameter a parameter controlling order of mergers * @param distance distance metric over float [] * @param seed random seed * @param parallelEnabled flag enabling (limited) parallelism * @param shrinkage a parameter that morphs from centroidal * behavior (=1) to robust Minimum Spanning Tree * (=0) * @param numberOfRepresentatives the number of representatives ina multicentrod * representation used to cluster potentially * non-spherical shapes * @return a list of centers with weights */ public static <R> List<ICluster<R>> multiSummarize(List<R> points, int maxAllowed, int initial, int stopAt, boolean phase2GlobalReassign, double overlapParameter, BiFunction<R, R, Double> distance, long seed, Boolean parallelEnabled, double shrinkage, int numberOfRepresentatives) { ArrayList<Weighted<R>> weighted = new ArrayList<>(); for (R point : points) { weighted.add(new Weighted<>(point, 1.0f)); } BiFunction<R, Float, ICluster<R>> clusterInitializer = (a, b) -> GenericMultiCenter.initialize(a, b, shrinkage, numberOfRepresentatives); return summarize(weighted, maxAllowed, initial, stopAt, phase2GlobalReassign, overlapParameter, distance, clusterInitializer, seed, parallelEnabled, null); } // same as above, different input public static <R> List<ICluster<R>> multiSummarize(R[] points, int maxAllowed, int initial, int stopAt, boolean phase2GlobalReassign, double overlapParameter, BiFunction<R, R, Double> distance, long seed, Boolean parallelEnabled, double shrinkage, int numberOfRepresentatives) { ArrayList<Weighted<R>> weighted = new ArrayList<>(); for (R point : points) { weighted.add(new Weighted<>(point, 1.0f)); } BiFunction<R, Float, ICluster<R>> clusterInitializer = (a, b) -> GenericMultiCenter.initialize(a, b, shrinkage, numberOfRepresentatives); return summarize(weighted, maxAllowed, initial, stopAt, phase2GlobalReassign, overlapParameter, distance, clusterInitializer, seed, parallelEnabled, null); } // same as above, with multicenter instead of generic public static List<ICluster<float[]>> multiSummarize(float[][] points, int maxAllowed, double shrinkage, int numberOfRepresentatives, long seed) { ArrayList<Weighted<float[]>> weighted = new ArrayList<>(); for (float[] point : points) { weighted.add(new Weighted<>(point, 1.0f)); } BiFunction<float[], Float, ICluster<float[]>> clusterInitializer = (a, b) -> MultiCenter.initialize(a, b, shrinkage, numberOfRepresentatives); return summarize(weighted, maxAllowed, 4 * maxAllowed, 1, true, DEFAULT_SEPARATION_RATIO_FOR_MERGE, Summarizer::L2distance, clusterInitializer, seed, true, null); } }
487
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/ArgumentParser.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.checkNotNull; import java.util.HashMap; import java.util.Map; import java.util.function.Consumer; import java.util.function.Function; /** * A utility class for parsing command-line arguments. */ public class ArgumentParser { public static final String ARCHIVE_NAME = "randomcutforest-core-1.0.jar"; private final String runnerClass; private final String runnerDescription; private final Map<String, Argument<?>> shortFlags; private final Map<String, Argument<?>> longFlags; private final IntegerArgument numberOfTrees; private final IntegerArgument sampleSize; private final IntegerArgument windowSize; private final IntegerArgument shingleSize; private final BooleanArgument shingleCyclic; private final StringArgument delimiter; private final BooleanArgument headerRow; private final IntegerArgument randomSeed; /** * Create a new ArgumentParser.The runner class and runner description will be * used in help text. * * @param runnerClass The name of the runner class where this argument * parser is being invoked. * @param runnerDescription A description of the runner class where this * argument parser is being invoked. */ public ArgumentParser(String runnerClass, String runnerDescription) { this.runnerClass = runnerClass; this.runnerDescription = runnerDescription; shortFlags = new HashMap<>(); longFlags = new HashMap<>(); numberOfTrees = new IntegerArgument("-n", "--number-of-trees", "Number of trees to use in the forest.", 100, n -> checkArgument(n > 0, "number of trees should be greater than 0")); addArgument(numberOfTrees); sampleSize = new IntegerArgument("-s", "--sample-size", "Number of points to keep in sample for each tree.", 256, n -> checkArgument(n > 0, "sample size should be greater than 0")); addArgument(sampleSize); windowSize = new IntegerArgument("-w", "--window-size", "Window size of the sample or 0 for no window.", 0, n -> checkArgument(n > 0, "window size should be greater than 0")); addArgument(windowSize); shingleSize = new IntegerArgument("-g", "--shingle-size", "Shingle size to use.", 1, n -> checkArgument(n > 0, "shingle size should be greater than 0")); addArgument(shingleSize); shingleCyclic = new BooleanArgument("-c", "--shingle-cyclic", "Set to 'true' to use cyclic shingles instead of linear shingles.", false); addArgument(shingleCyclic); delimiter = new StringArgument("-d", "--delimiter", "The character or string used as a field delimiter.", ","); addArgument(delimiter); headerRow = new BooleanArgument(null, "--header-row", "Set to 'true' if the data contains a header row.", false); addArgument(headerRow); randomSeed = new IntegerArgument(null, "--random-seed", "Random seed to use in the Random Cut Forest", 42); addArgument(randomSeed); } /** * Add a new argument to this argument parser. * * @param argument An Argument instance for a command-line argument that should * be parsed. */ protected void addArgument(Argument<?> argument) { checkNotNull(argument, "argument should not be null"); checkArgument(argument.getShortFlag() == null || !shortFlags.containsKey(argument.getShortFlag()), String.format("An argument mapping already exists for %s", argument.getShortFlag())); checkArgument(!longFlags.containsKey(argument.getLongFlag()), String.format("An argument mapping already exists for %s", argument.getLongFlag())); if (argument.getShortFlag() != null) { shortFlags.put(argument.getShortFlag(), argument); } longFlags.put(argument.getLongFlag(), argument); } /** * Remove the argument with the given long flag from help messages. This allows * subclasses to suppress arguments as needed. The argument will still exist in * this object with its default value. * * @param longFlag The long flag corresponding to the argument being removed */ protected void removeArgument(String longFlag) { Argument<?> argument = longFlags.get(longFlag); if (argument != null) { longFlags.remove(longFlag); shortFlags.remove(argument.getShortFlag()); } } /** * Parse the given array of command-line arguments. * * @param arguments An array of command-line arguments. */ public void parse(String... arguments) { int i = 0; while (i < arguments.length) { String flag = arguments[i]; try { if (shortFlags.containsKey(flag)) { shortFlags.get(flag).parse(arguments[++i]); } else if (longFlags.containsKey(flag)) { longFlags.get(flag).parse(arguments[++i]); } else if ("-h".equals(flag) || "--help".equals(flag)) { printUsage(); Runtime.getRuntime().exit(0); } else { throw new IllegalArgumentException("Unknown argument: " + flag); } } catch (Exception e) { printUsageAndExit("%s: %s", e.getClass().getName(), e.getMessage()); } i++; } } /** * Print a usage message to STDOUT. */ public void printUsage() { System.out.println( String.format("Usage: java -cp %s %s [options] < input_file > output_file", ARCHIVE_NAME, runnerClass)); System.out.println(); System.out.println(runnerDescription); System.out.println(); System.out.println("Options:"); longFlags.values().stream().map(Argument::getHelpMessage).sorted() .forEach(msg -> System.out.println("\t" + msg)); System.out.println(); System.out.println("\t--help, -h: Print this help message and exit."); } /** * Print an error message, the usage message, and exit the application. * * @param errorMessage An error message to show the user. * @param formatObjects An array of format objects that will be interpolated * into the error message using {@link String#format}. */ public void printUsageAndExit(String errorMessage, Object... formatObjects) { System.err.println("Error: " + String.format(errorMessage, formatObjects)); printUsage(); System.exit(1); } /** * @return the user-specified value of the number-of-trees parameter. */ public int getNumberOfTrees() { return numberOfTrees.getValue(); } /** * @return the user-specified value of the sample-size parameter. */ public int getSampleSize() { return sampleSize.getValue(); } /** * @return the user-specified value of the window-size parameter */ public int getWindowSize() { return windowSize.getValue(); } /** * @return the user-specified value of the time-decay parameter */ public double getTimeDecay() { if (getWindowSize() > 0) { return 1.0 / getWindowSize(); } else { return 0.0; } } /** * @return the user-specified value of the shingle-size parameter */ public int getShingleSize() { return shingleSize.getValue(); } /** * @return the user-specified value of the shingle-cyclic parameter */ public boolean getShingleCyclic() { return shingleCyclic.getValue(); } /** * @return the user-specified value of the delimiter parameter */ public String getDelimiter() { return delimiter.getValue(); } /** * @return the user-specified value of the header-row parameter */ public boolean getHeaderRow() { return headerRow.getValue(); } /** * @return the user-specified value of the random-seed parameter */ public int getRandomSeed() { return randomSeed.getValue(); } public static class Argument<T> { private final String shortFlag; private final String longFlag; private final String description; private final T defaultValue; private final Function<String, T> parseFunction; private final Consumer<T> validateFunction; private T value; public Argument(String shortFlag, String longFlag, String description, T defaultValue, Function<String, T> parseFunction, Consumer<T> validateFunction) { this.shortFlag = shortFlag; this.longFlag = longFlag; this.description = description; this.defaultValue = defaultValue; this.parseFunction = parseFunction; this.validateFunction = validateFunction; value = defaultValue; } public Argument(String shortFlag, String longFlag, String description, T defaultValue, Function<String, T> parseFunction) { this(shortFlag, longFlag, description, defaultValue, parseFunction, t -> { }); } public String getShortFlag() { return shortFlag; } public String getLongFlag() { return longFlag; } public String getDescription() { return description; } public T getDefaultValue() { return defaultValue; } public String getHelpMessage() { if (shortFlag != null) { return String.format("%s, %s: %s (default: %s)", longFlag, shortFlag, description, defaultValue); } else { return String.format("%s: %s (default: %s)", longFlag, description, defaultValue); } } public void parse(String string) { value = parseFunction.apply(string); validateFunction.accept(value); } public T getValue() { return value; } } public static class StringArgument extends Argument<String> { public StringArgument(String shortFlag, String longFlag, String description, String defaultValue, Consumer<String> validateFunction) { super(shortFlag, longFlag, description, defaultValue, x -> x, validateFunction); } public StringArgument(String shortFlag, String longFlag, String description, String defaultValue) { super(shortFlag, longFlag, description, defaultValue, x -> x); } } public static class BooleanArgument extends Argument<Boolean> { public BooleanArgument(String shortFlag, String longFlag, String description, boolean defaultValue) { super(shortFlag, longFlag, description, defaultValue, Boolean::parseBoolean); } } public static class IntegerArgument extends Argument<Integer> { public IntegerArgument(String shortFlag, String longFlag, String description, int defaultValue, Consumer<Integer> validateFunction) { super(shortFlag, longFlag, description, defaultValue, Integer::parseInt, validateFunction); } public IntegerArgument(String shortFlag, String longFlag, String description, int defaultValue) { super(shortFlag, longFlag, description, defaultValue, Integer::parseInt); } } public static class DoubleArgument extends Argument<Double> { public DoubleArgument(String shortFlag, String longFlag, String description, double defaultValue, Consumer<Double> validateFunction) { super(shortFlag, longFlag, description, defaultValue, Double::parseDouble, validateFunction); } public DoubleArgument(String shortFlag, String longFlag, String description, double defaultValue) { super(shortFlag, longFlag, description, defaultValue, Double::parseDouble); } } }
488
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/LineTransformer.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import java.util.List; import com.amazon.randomcutforest.RandomCutForest; /** * This interface is used by SimpleRunner to transform input lines into output * lines. */ public interface LineTransformer { /** * For the given parsed input point, return a list of string values that should * be written as output. The list of strings will be joined together using the * user-specified delimiter. * * @param point A point value that was parsed from the input stream. * @return a list of string values that should be written as output. */ List<String> getResultValues(double[] point); /** * @return a list of string values that should be written to the output when * processing a line if there is no input point available. This method * is invoked when shingling is enabled before the first shingle is * full. */ List<String> getEmptyResultValue(); /** * @return a list of column names to write to the output if headers are enabled. */ List<String> getResultColumnNames(); /** * @return the RandomCutForest instance which is being used internally to * process lines. */ RandomCutForest getForest(); }
489
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/AnomalyScoreRunner.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import java.io.BufferedReader; import java.io.IOException; import java.io.InputStreamReader; import java.io.OutputStreamWriter; import java.io.PrintWriter; import java.nio.charset.StandardCharsets; import java.util.Collections; import java.util.List; import com.amazon.randomcutforest.RandomCutForest; /** * A command-line application that computes anomaly scores. Points are read from * STDIN and output is written to STDOUT. Output consists of the original input * point with the anomaly score appended. */ public class AnomalyScoreRunner extends SimpleRunner { public AnomalyScoreRunner() { super(AnomalyScoreRunner.class.getName(), "Compute scalar anomaly scores from the input rows and append them to the output rows.", AnomalyScoreTransformer::new); } public static void main(String... args) throws IOException { AnomalyScoreRunner runner = new AnomalyScoreRunner(); runner.parse(args); System.out.println("Reading from stdin... (Ctrl-c to exit)"); runner.run(new BufferedReader(new InputStreamReader(System.in, StandardCharsets.UTF_8)), new PrintWriter(new OutputStreamWriter(System.out, StandardCharsets.UTF_8))); System.out.println("Done."); } public static class AnomalyScoreTransformer implements LineTransformer { private final RandomCutForest forest; public AnomalyScoreTransformer(RandomCutForest forest) { this.forest = forest; } @Override public List<String> getResultValues(double... point) { double score = forest.getAnomalyScore(point); forest.update(point); return Collections.singletonList(Double.toString(score)); } @Override public List<String> getEmptyResultValue() { return Collections.singletonList("NA"); } @Override public List<String> getResultColumnNames() { return Collections.singletonList("anomaly_score"); } @Override public RandomCutForest getForest() { return forest; } } }
490
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/ImputeRunner.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import java.io.BufferedReader; import java.io.IOException; import java.io.InputStreamReader; import java.io.OutputStreamWriter; import java.io.PrintWriter; import java.nio.charset.StandardCharsets; import java.util.Arrays; import java.util.StringJoiner; /** * A command-line application that imputes missing values. Points are read from * STDIN and output is written to STDOUT. One output point is written for each * input point. If the input point does not contain any missing value * indicators, then it is copied as-is to the output. If an input point contains * one or more missing value indicators, then the missing values are imputed and * the imputed point is written to the output. */ public class ImputeRunner extends SimpleRunner { private String missingValueMarker; private int numberOfMissingValues; private int[] missingIndexes; public ImputeRunner() { super(new ImputeArgumentParser(), UpdateOnlyTransformer::new); } public static void main(String... args) throws IOException { ImputeRunner runner = new ImputeRunner(); runner.parse(args); System.out.println("Reading from stdin... (Ctrl-c to exit)"); runner.run(new BufferedReader(new InputStreamReader(System.in, StandardCharsets.UTF_8)), new PrintWriter(new OutputStreamWriter(System.out, StandardCharsets.UTF_8))); System.out.println("Done."); } @Override protected void prepareAlgorithm(int dimensions) { super.prepareAlgorithm(dimensions); missingIndexes = new int[dimensions]; missingValueMarker = ((ImputeArgumentParser) argumentParser).getMissingValueMarker(); } @Override protected void processLine(String[] values, PrintWriter out) { numberOfMissingValues = 0; for (int i = 0; i < getPointSize(); i++) { if (missingValueMarker.equals(values[i])) { missingIndexes[numberOfMissingValues++] = i; values[i] = "0"; } } if (numberOfMissingValues > 0) { parsePoint(values); double[] imputedPoint = algorithm.getForest().imputeMissingValues(pointBuffer, numberOfMissingValues, missingIndexes); StringJoiner joiner = new StringJoiner(argumentParser.getDelimiter()); Arrays.stream(imputedPoint).mapToObj(Double::toString).forEach(joiner::add); out.println(joiner.toString()); } else { super.processLine(values, out); } } public static class ImputeArgumentParser extends ArgumentParser { private final StringArgument missingValueMarker; public ImputeArgumentParser() { super(ImputeRunner.class.getName(), "Read rows with missing values and write rows with missing values imputed."); missingValueMarker = new StringArgument(null, "--missing-value-marker", "String used to represent a missing value in the data.", "NA"); addArgument(missingValueMarker); removeArgument("--shingle-size"); removeArgument("--shingle-cyclic"); } public String getMissingValueMarker() { return missingValueMarker.getValue(); } } }
491
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/SimpleRunner.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import java.io.BufferedReader; import java.io.IOException; import java.io.PrintWriter; import java.util.Arrays; import java.util.List; import java.util.StringJoiner; import java.util.function.Function; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.util.ShingleBuilder; /** * A simple command-line application that parses command-line arguments, creates * a RandomCutForest instance based on those arguments, reads values from STDIN * and writes results to STDOUT. */ public class SimpleRunner { protected final ArgumentParser argumentParser; protected final Function<RandomCutForest, LineTransformer> algorithmInitializer; protected LineTransformer algorithm; protected ShingleBuilder shingleBuilder; protected double[] pointBuffer; protected double[] shingleBuffer; protected int lineNumber; /** * Create a new SimpleRunner. * * @param runnerClass The name of the runner class. This will be * displayed in the help text. * @param runnerDescription A description of the runner class. This will be * displayed in the help text. * @param algorithmInitializer A factory method to create a new LineTransformer * instance from a RandomCutForest. */ public SimpleRunner(String runnerClass, String runnerDescription, Function<RandomCutForest, LineTransformer> algorithmInitializer) { this(new ArgumentParser(runnerClass, runnerDescription), algorithmInitializer); } /** * Create a new SimpleRunner. * * @param argumentParser A argument parser that will be used by this * runner to parse command-line arguments. * @param algorithmInitializer A factory method to create a new LineTransformer * instance from a RandomCutForest. */ public SimpleRunner(ArgumentParser argumentParser, Function<RandomCutForest, LineTransformer> algorithmInitializer) { this.argumentParser = argumentParser; this.algorithmInitializer = algorithmInitializer; } /** * Parse the given command-line arguments. * * @param arguments An array of command-line arguments. */ public void parse(String... arguments) { argumentParser.parse(arguments); } /** * Read data from an input stream, apply the desired transformation, and write * the result to an output stream. * * @param in An input stream where input values will be read. * @param out An output stream where the result values will be written. * @throws IOException if IO errors are encountered during reading or writing. */ public void run(BufferedReader in, PrintWriter out) throws IOException { String line; while ((line = in.readLine()) != null) { lineNumber++; String[] values = line.split(argumentParser.getDelimiter()); if (pointBuffer == null) { prepareAlgorithm(values.length); } if (lineNumber == 1 && argumentParser.getHeaderRow()) { writeHeader(values, out); continue; } processLine(values, out); } finish(out); out.flush(); } /** * Set up the internal RandomCutForest instance and line transformer. * * @param dimensions The number of dimensions in the input data. */ protected void prepareAlgorithm(int dimensions) { pointBuffer = new double[dimensions]; shingleBuilder = new ShingleBuilder(dimensions, argumentParser.getShingleSize(), argumentParser.getShingleCyclic()); shingleBuffer = new double[shingleBuilder.getShingledPointSize()]; RandomCutForest forest = RandomCutForest.builder().numberOfTrees(argumentParser.getNumberOfTrees()) .sampleSize(argumentParser.getSampleSize()).dimensions(shingleBuilder.getShingledPointSize()) .timeDecay(argumentParser.getTimeDecay()).randomSeed(argumentParser.getRandomSeed()).build(); algorithm = algorithmInitializer.apply(forest); } /** * Write a header row to the output stream. * * @param values The array of values that are used to create the header. These * values will be joined together using the user-specified * delimiter. * @param out The output stream where the header will be written. */ protected void writeHeader(String[] values, PrintWriter out) { StringJoiner joiner = new StringJoiner(argumentParser.getDelimiter()); Arrays.stream(values).forEach(joiner::add); algorithm.getResultColumnNames().forEach(joiner::add); out.println(joiner.toString()); } /** * Process a single line of input data and write the result to the output * stream. * * @param values An array of string values taken from the input stream. These * values will be parsed into an array of doubles before being * transformed and written to the output stream. * @param out The output stream where the transformed line will be written. */ protected void processLine(String[] values, PrintWriter out) { if (values.length != pointBuffer.length) { throw new IllegalArgumentException( String.format("Wrong number of values on line %d. Exected %d but found %d.", lineNumber, pointBuffer.length, values.length)); } parsePoint(values); shingleBuilder.addPoint(pointBuffer); List<String> result; if (shingleBuilder.isFull()) { shingleBuilder.getShingle(shingleBuffer); result = algorithm.getResultValues(shingleBuffer); } else { result = algorithm.getEmptyResultValue(); } StringJoiner joiner = new StringJoiner(argumentParser.getDelimiter()); Arrays.stream(values).forEach(joiner::add); result.forEach(joiner::add); out.println(joiner.toString()); } /** * Parse the array of string values into doubles and write them to an internal * buffer. * * @param stringValues An array of string-encoded double values. */ protected void parsePoint(String[] stringValues) { for (int i = 0; i < pointBuffer.length; i++) { pointBuffer[i] = Double.parseDouble(stringValues[i]); } } /** * This method is used to write any final output to the output stream after the * input stream has beeen fully processed. * * @param out The output stream where additional output text may be written. */ protected void finish(PrintWriter out) { } /** * @return the size of the internal point buffer. */ protected int getPointSize() { return pointBuffer != null ? pointBuffer.length : 0; } /** * @return the size of the internal shingled point buffer. */ protected int getShingleSize() { return shingleBuffer != null ? shingleBuffer.length : 0; } }
492
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/UpdateOnlyTransformer.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import java.util.Collections; import java.util.List; import com.amazon.randomcutforest.RandomCutForest; /** * A line transformer that updates its internal RandomCutForest instance but * does not produce any output. */ public class UpdateOnlyTransformer implements LineTransformer { private final RandomCutForest forest; public UpdateOnlyTransformer(RandomCutForest forest) { this.forest = forest; } @Override public List<String> getResultValues(double... point) { forest.update(point); return Collections.emptyList(); } @Override public List<String> getEmptyResultValue() { return Collections.emptyList(); } @Override public List<String> getResultColumnNames() { return Collections.emptyList(); } @Override public RandomCutForest getForest() { return forest; } }
493
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/SimpleDensityRunner.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import java.io.BufferedReader; import java.io.IOException; import java.io.InputStreamReader; import java.io.OutputStreamWriter; import java.io.PrintWriter; import java.nio.charset.StandardCharsets; import java.util.ArrayList; import java.util.List; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.returntypes.DiVector; /** * A command-line application that computes directional density. Points are read * from STDIN and output is written to STDOUT. Output consists of the original * input point with the directional density vector appended. */ public class SimpleDensityRunner extends SimpleRunner { public SimpleDensityRunner() { super(SimpleDensityRunner.class.getName(), "Compute directional density vectors from the input rows and append them to the output rows.", SimpleDensityRunner.SimpleDensityTransformer::new); } public static void main(String... args) throws IOException { SimpleDensityRunner runner = new SimpleDensityRunner(); runner.parse(args); System.out.println("Reading from stdin... (Ctrl-c to exit)"); runner.run(new BufferedReader(new InputStreamReader(System.in, StandardCharsets.UTF_8)), new PrintWriter(new OutputStreamWriter(System.out, StandardCharsets.UTF_8))); System.out.println("Done."); } public static class SimpleDensityTransformer implements LineTransformer { private final RandomCutForest forest; public SimpleDensityTransformer(RandomCutForest forest) { this.forest = forest; } @Override public List<String> getResultValues(double... point) { DiVector densityFactors = forest.getSimpleDensity(point).getDirectionalDensity(); forest.update(point); List<String> result = new ArrayList<>(2 * forest.getDimensions()); for (int i = 0; i < forest.getDimensions(); i++) { result.add(String.format("%f", densityFactors.high[i])); result.add(String.format("%f", densityFactors.low[i])); } return result; } @Override public List<String> getEmptyResultValue() { List<String> result = new ArrayList<>(2 * forest.getDimensions()); for (int i = 0; i < 2 * forest.getDimensions(); i++) { result.add("NA"); } return result; } @Override public List<String> getResultColumnNames() { List<String> result = new ArrayList<>(2 * forest.getDimensions()); for (int i = 0; i < forest.getDimensions(); i++) { result.add(String.format("prob_mass_%d_up", i)); result.add(String.format("prob_mass_%d_down", i)); } return result; } @Override public RandomCutForest getForest() { return forest; } } }
494
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/runner/AnomalyAttributionRunner.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.runner; import java.io.BufferedReader; import java.io.IOException; import java.io.InputStreamReader; import java.io.OutputStreamWriter; import java.io.PrintWriter; import java.nio.charset.StandardCharsets; import java.util.ArrayList; import java.util.List; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.returntypes.DiVector; /** * A command-line application that computes anomaly attribution. Points are read * from STDIN and output is written to STDOUT. Output consists of the original * input point with the anomaly attribution vector appended. */ public class AnomalyAttributionRunner extends SimpleRunner { public AnomalyAttributionRunner() { super(AnomalyAttributionRunner.class.getName(), "Compute directional anomaly scores from the input rows and append them to the output rows.", AnomalyAttributionTransformer::new); } public static void main(String... args) throws IOException { AnomalyAttributionRunner runner = new AnomalyAttributionRunner(); runner.parse(args); System.out.println("Reading from stdin... (Ctrl-c to exit)"); runner.run(new BufferedReader(new InputStreamReader(System.in, StandardCharsets.UTF_8)), new PrintWriter(new OutputStreamWriter(System.out, StandardCharsets.UTF_8))); System.out.println("Done."); } public static class AnomalyAttributionTransformer implements LineTransformer { private final RandomCutForest forest; public AnomalyAttributionTransformer(RandomCutForest forest) { this.forest = forest; } @Override public List<String> getResultValues(double... point) { DiVector attribution = forest.getAnomalyAttribution(point); forest.update(point); List<String> result = new ArrayList<>(2 * forest.getDimensions()); for (int i = 0; i < attribution.getDimensions(); i++) { result.add(Double.toString(attribution.low[i])); result.add(Double.toString(attribution.high[i])); } return result; } @Override public List<String> getEmptyResultValue() { List<String> result = new ArrayList<>(2 * forest.getDimensions()); for (int i = 0; i < 2 * forest.getDimensions(); i++) { result.add("NA"); } return result; } @Override public List<String> getResultColumnNames() { List<String> result = new ArrayList<>(2 * forest.getDimensions()); for (int i = 0; i < forest.getDimensions(); i++) { result.add("anomaly_low_" + i); result.add("anomaly_high_" + i); } return result; } @Override public RandomCutForest getForest() { return forest; } } }
495
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/state/IContextualStateMapper.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.state; public interface IContextualStateMapper<Model, State, ContextState> { State toState(Model model); Model toModel(State state, ContextState contextState, long seed); default Model toModel(State state, ContextState contextState) { return toModel(state, contextState, 0L); } }
496
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/state/ExecutionContext.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.state; import java.io.Serializable; import lombok.Data; @Data public class ExecutionContext implements Serializable { private static final long serialVersionUID = 1L; private boolean parallelExecutionEnabled; private int threadPoolSize; /** * A string to define an "execution mode" that can be used to set multiple * configuration options. This field is not currently in use. */ private String mode; }
497
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/state/RandomCutForestMapper.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.state; import static com.amazon.randomcutforest.CommonUtils.checkArgument; import static com.amazon.randomcutforest.CommonUtils.checkNotNull; import java.util.ArrayList; import java.util.List; import java.util.Random; import java.util.stream.Collectors; import lombok.Getter; import lombok.Setter; import com.amazon.randomcutforest.ComponentList; import com.amazon.randomcutforest.IComponentModel; import com.amazon.randomcutforest.RandomCutForest; import com.amazon.randomcutforest.config.Config; import com.amazon.randomcutforest.config.Precision; import com.amazon.randomcutforest.executor.PointStoreCoordinator; import com.amazon.randomcutforest.executor.SamplerPlusTree; import com.amazon.randomcutforest.sampler.CompactSampler; import com.amazon.randomcutforest.sampler.IStreamSampler; import com.amazon.randomcutforest.sampler.Weighted; import com.amazon.randomcutforest.state.sampler.CompactSamplerMapper; import com.amazon.randomcutforest.state.sampler.CompactSamplerState; import com.amazon.randomcutforest.state.store.PointStoreMapper; import com.amazon.randomcutforest.state.store.PointStoreState; import com.amazon.randomcutforest.state.tree.CompactRandomCutTreeContext; import com.amazon.randomcutforest.state.tree.CompactRandomCutTreeState; import com.amazon.randomcutforest.state.tree.RandomCutTreeMapper; import com.amazon.randomcutforest.store.IPointStore; import com.amazon.randomcutforest.store.PointStore; import com.amazon.randomcutforest.tree.ITree; import com.amazon.randomcutforest.tree.RandomCutTree; /** * A utility class for creating a {@link RandomCutForestState} instance from a * {@link RandomCutForest} instance and vice versa. */ @Getter @Setter public class RandomCutForestMapper implements IContextualStateMapper<RandomCutForest, RandomCutForestState, ExecutionContext> { /** * A flag indicating whether the structure of the trees in the forest should be * included in the state object. If true, then data describing the bounding * boxes and cuts defining each tree will be written to the * {@link RandomCutForestState} object produced by the mapper. Tree state is not * saved by default. */ private boolean saveTreeStateEnabled = false; /** * A flag indicating whether the point store should be included in the * {@link RandomCutForestState} object produced by the mapper. This is saved by * default for compact trees */ private boolean saveCoordinatorStateEnabled = true; /** * A flag indicating whether the samplers should be included in the * {@link RandomCutForestState} object produced by the mapper. This is saved by * default for all trees. */ private boolean saveSamplerStateEnabled = true; /** * A flag indicating whether the executor context should be included in the * {@link RandomCutForestState} object produced by the mapper. Executor context * is not saved by defalt. */ private boolean saveExecutorContextEnabled = false; /** * If true, then the arrays are compressed via simple data dependent scheme */ private boolean compressionEnabled = true; /** * if true would require that the samplers populate the trees before the trees * can be used gain. That woukld correspond to extra time, at the benefit of a * smaller serialization. */ private boolean partialTreeStateEnabled = false; /** * Create a {@link RandomCutForestState} object representing the state of the * given forest. If the forest is compact and the {@code saveTreeState} flag is * set to true, then structure of the trees in the forest will be included in * the state object. If the flag is set to false, then the state object will * only contain the sampler data for each tree. If the * {@code saveExecutorContext} is true, then the executor context will be * included in the state object. * * @param forest A Random Cut Forest whose state we want to capture. * @return a {@link RandomCutForestState} object representing the state of the * given forest. * @throws IllegalArgumentException if the {@code saveTreeState} flag is true * and the forest is not compact. */ @Override public RandomCutForestState toState(RandomCutForest forest) { if (saveTreeStateEnabled) { checkArgument(forest.isCompact(), "tree state cannot be saved for noncompact forests"); } RandomCutForestState state = new RandomCutForestState(); state.setNumberOfTrees(forest.getNumberOfTrees()); state.setDimensions(forest.getDimensions()); state.setTimeDecay(forest.getTimeDecay()); state.setSampleSize(forest.getSampleSize()); state.setShingleSize(forest.getShingleSize()); state.setCenterOfMassEnabled(forest.isCenterOfMassEnabled()); state.setOutputAfter(forest.getOutputAfter()); state.setStoreSequenceIndexesEnabled(forest.isStoreSequenceIndexesEnabled()); state.setTotalUpdates(forest.getTotalUpdates()); state.setCompact(forest.isCompact()); state.setInternalShinglingEnabled(forest.isInternalShinglingEnabled()); state.setBoundingBoxCacheFraction(forest.getBoundingBoxCacheFraction()); state.setSaveSamplerStateEnabled(saveSamplerStateEnabled); state.setSaveTreeStateEnabled(saveTreeStateEnabled); state.setSaveCoordinatorStateEnabled(saveCoordinatorStateEnabled); state.setPrecision(forest.getPrecision().name()); state.setCompressed(compressionEnabled); state.setPartialTreeState(partialTreeStateEnabled); if (saveExecutorContextEnabled) { ExecutionContext executionContext = new ExecutionContext(); executionContext.setParallelExecutionEnabled(forest.isParallelExecutionEnabled()); executionContext.setThreadPoolSize(forest.getThreadPoolSize()); state.setExecutionContext(executionContext); } if (saveCoordinatorStateEnabled) { PointStoreCoordinator<?> pointStoreCoordinator = (PointStoreCoordinator<?>) forest.getUpdateCoordinator(); PointStoreMapper mapper = new PointStoreMapper(); mapper.setCompressionEnabled(compressionEnabled); mapper.setNumberOfTrees(forest.getNumberOfTrees()); PointStoreState pointStoreState = mapper.toState((PointStore) pointStoreCoordinator.getStore()); state.setPointStoreState(pointStoreState); } List<CompactSamplerState> samplerStates = null; if (saveSamplerStateEnabled) { samplerStates = new ArrayList<>(); } List<ITree<Integer, ?>> trees = null; if (saveTreeStateEnabled) { trees = new ArrayList<>(); } CompactSamplerMapper samplerMapper = new CompactSamplerMapper(); samplerMapper.setCompressionEnabled(compressionEnabled); for (IComponentModel<?, ?> component : forest.getComponents()) { SamplerPlusTree<Integer, ?> samplerPlusTree = (SamplerPlusTree<Integer, ?>) component; CompactSampler sampler = (CompactSampler) samplerPlusTree.getSampler(); if (samplerStates != null) { samplerStates.add(samplerMapper.toState(sampler)); } if (trees != null) { trees.add(samplerPlusTree.getTree()); } } state.setCompactSamplerStates(samplerStates); if (trees != null) { RandomCutTreeMapper treeMapper = new RandomCutTreeMapper(); List<CompactRandomCutTreeState> treeStates = trees.stream().map(t -> treeMapper.toState((RandomCutTree) t)) .collect(Collectors.toList()); state.setCompactRandomCutTreeStates(treeStates); } return state; } /** * Create a {@link RandomCutForest} instance from a * {@link RandomCutForestState}. If the state contains tree states, then trees * will be constructed from the tree state objects. Otherwise, empty trees are * created and populated from the sampler data. The resulting forest should be * equal in distribution to the forest that the state object was created from. * * @param state A Random Cut Forest state object. * @param executionContext An executor context that will be used to initialize * new executors in the Random Cut Forest. If this * argument is null, then the mapper will look for an * executor context in the state object. * @param seed A random seed. * @return A Random Cut Forest corresponding to the state object. * @throws NullPointerException if both the {@code executorContext} method * argument and the executor context field in the * state object are null. */ public RandomCutForest toModel(RandomCutForestState state, ExecutionContext executionContext, long seed) { ExecutionContext ec; if (executionContext != null) { ec = executionContext; } else { checkNotNull(state.getExecutionContext(), "The executor context in the state object is null, an executor context must be passed explicitly to toModel()"); ec = state.getExecutionContext(); } RandomCutForest.Builder<?> builder = RandomCutForest.builder().numberOfTrees(state.getNumberOfTrees()) .dimensions(state.getDimensions()).timeDecay(state.getTimeDecay()).sampleSize(state.getSampleSize()) .centerOfMassEnabled(state.isCenterOfMassEnabled()).outputAfter(state.getOutputAfter()) .parallelExecutionEnabled(ec.isParallelExecutionEnabled()).threadPoolSize(ec.getThreadPoolSize()) .storeSequenceIndexesEnabled(state.isStoreSequenceIndexesEnabled()).shingleSize(state.getShingleSize()) .boundingBoxCacheFraction(state.getBoundingBoxCacheFraction()).compact(state.isCompact()) .internalShinglingEnabled(state.isInternalShinglingEnabled()).randomSeed(seed); if (Precision.valueOf(state.getPrecision()) == Precision.FLOAT_32) { return singlePrecisionForest(builder, state, null, null, null); } Random random = builder.getRandom(); PointStore pointStore = new PointStoreMapper().convertFromDouble(state.getPointStoreState()); ComponentList<Integer, float[]> components = new ComponentList<>(); PointStoreCoordinator<float[]> coordinator = new PointStoreCoordinator<>(pointStore); coordinator.setTotalUpdates(state.getTotalUpdates()); CompactRandomCutTreeContext context = new CompactRandomCutTreeContext(); context.setPointStore(pointStore); context.setMaxSize(state.getSampleSize()); checkArgument(state.isSaveSamplerStateEnabled(), " conversion cannot proceed without samplers"); List<CompactSamplerState> samplerStates = state.getCompactSamplerStates(); CompactSamplerMapper samplerMapper = new CompactSamplerMapper(); for (int i = 0; i < state.getNumberOfTrees(); i++) { CompactSampler compactData = samplerMapper.toModel(samplerStates.get(i)); RandomCutTree tree = RandomCutTree.builder().capacity(state.getSampleSize()).pointStoreView(pointStore) .storeSequenceIndexesEnabled(state.isStoreSequenceIndexesEnabled()) .outputAfter(state.getOutputAfter()).centerOfMassEnabled(state.isCenterOfMassEnabled()) .randomSeed(random.nextLong()).build(); CompactSampler sampler = CompactSampler.builder().capacity(state.getSampleSize()) .timeDecay(state.getTimeDecay()).randomSeed(random.nextLong()).build(); sampler.setMaxSequenceIndex(compactData.getMaxSequenceIndex()); sampler.setMostRecentTimeDecayUpdate(compactData.getMostRecentTimeDecayUpdate()); for (Weighted<Integer> sample : compactData.getWeightedSample()) { Integer reference = sample.getValue(); Integer newReference = tree.addPoint(reference, sample.getSequenceIndex()); if (newReference.intValue() != reference.intValue()) { pointStore.incrementRefCount(newReference); pointStore.decrementRefCount(reference); } sampler.addPoint(newReference, sample.getWeight(), sample.getSequenceIndex()); } components.add(new SamplerPlusTree<>(sampler, tree)); } return new RandomCutForest(builder, coordinator, components, random); } /** * Create a {@link RandomCutForest} instance from a {@link RandomCutForestState} * using the executor context in the state object. See * {@link #toModel(RandomCutForestState, ExecutionContext, long)}. * * @param state A Random Cut Forest state object. * @param seed A random seed. * @return A Random Cut Forest corresponding to the state object. * @throws NullPointerException if the executor context field in the state * object are null. */ public RandomCutForest toModel(RandomCutForestState state, long seed) { return toModel(state, null, seed); } /** * Create a {@link RandomCutForest} instance from a {@link RandomCutForestState} * using the executor context in the state object. See * {@link #toModel(RandomCutForestState, ExecutionContext, long)}. * * @param state A Random Cut Forest state object. * @return A Random Cut Forest corresponding to the state object. * @throws NullPointerException if the executor context field in the state * object are null. */ public RandomCutForest toModel(RandomCutForestState state) { return toModel(state, null); } public RandomCutForest singlePrecisionForest(RandomCutForest.Builder<?> builder, RandomCutForestState state, IPointStore<Integer, float[]> extPointStore, List<ITree<Integer, float[]>> extTrees, List<IStreamSampler<Integer>> extSamplers) { checkArgument(builder != null, "builder cannot be null"); checkArgument(extTrees == null || extTrees.size() == state.getNumberOfTrees(), "incorrect number of trees"); checkArgument(extSamplers == null || extSamplers.size() == state.getNumberOfTrees(), "incorrect number of samplers"); checkArgument(extSamplers != null | state.isSaveSamplerStateEnabled(), " need samplers "); checkArgument(extPointStore != null || state.isSaveCoordinatorStateEnabled(), " need coordinator state "); Random random = builder.getRandom(); ComponentList<Integer, float[]> components = new ComponentList<>(); CompactRandomCutTreeContext context = new CompactRandomCutTreeContext(); IPointStore<Integer, float[]> pointStore = (extPointStore == null) ? new PointStoreMapper().toModel(state.getPointStoreState()) : extPointStore; PointStoreCoordinator<float[]> coordinator = new PointStoreCoordinator<>(pointStore); coordinator.setTotalUpdates(state.getTotalUpdates()); context.setPointStore(pointStore); context.setMaxSize(state.getSampleSize()); RandomCutTreeMapper treeMapper = new RandomCutTreeMapper(); List<CompactRandomCutTreeState> treeStates = state.isSaveTreeStateEnabled() ? state.getCompactRandomCutTreeStates() : null; CompactSamplerMapper samplerMapper = new CompactSamplerMapper(); List<CompactSamplerState> samplerStates = state.isSaveSamplerStateEnabled() ? state.getCompactSamplerStates() : null; for (int i = 0; i < state.getNumberOfTrees(); i++) { IStreamSampler<Integer> sampler = (extSamplers != null) ? extSamplers.get(i) : samplerMapper.toModel(samplerStates.get(i), random.nextLong()); ITree<Integer, float[]> tree; if (extTrees != null) { tree = extTrees.get(i); } else if (treeStates != null) { tree = treeMapper.toModel(treeStates.get(i), context, random.nextLong()); sampler.getSample().forEach(s -> tree.addPointToPartialTree(s.getValue(), s.getSequenceIndex())); tree.setConfig(Config.BOUNDING_BOX_CACHE_FRACTION, treeStates.get(i).getBoundingBoxCacheFraction()); tree.validateAndReconstruct(); } else { // using boundingBoxCahce for the new tree tree = new RandomCutTree.Builder().capacity(state.getSampleSize()).randomSeed(random.nextLong()) .pointStoreView(pointStore).boundingBoxCacheFraction(state.getBoundingBoxCacheFraction()) .centerOfMassEnabled(state.isCenterOfMassEnabled()) .storeSequenceIndexesEnabled(state.isStoreSequenceIndexesEnabled()).build(); sampler.getSample().forEach(s -> tree.addPoint(s.getValue(), s.getSequenceIndex())); } components.add(new SamplerPlusTree<>(sampler, tree)); } builder.precision(Precision.FLOAT_32); return new RandomCutForest(builder, coordinator, components, random); } }
498
0
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest
Create_ds/random-cut-forest-by-aws/Java/core/src/main/java/com/amazon/randomcutforest/state/RandomCutForestState.java
/* * Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"). * You may not use this file except in compliance with the License. * A copy of the License is located at * * http://www.apache.org/licenses/LICENSE-2.0 * * or in the "license" file accompanying this file. This file is distributed * on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. See the License for the specific language governing * permissions and limitations under the License. */ package com.amazon.randomcutforest.state; import static com.amazon.randomcutforest.state.Version.V3_0; import java.io.Serializable; import java.util.List; import lombok.Data; import com.amazon.randomcutforest.state.sampler.CompactSamplerState; import com.amazon.randomcutforest.state.store.PointStoreState; import com.amazon.randomcutforest.state.tree.CompactRandomCutTreeState; /** * A class that encapsulates most of the data used in a RandomCutForest such * that the forest can be serialized and deserialized. */ @Data public class RandomCutForestState implements Serializable { private static final long serialVersionUID = 1L; private String version = V3_0; private long totalUpdates; private double timeDecay; private int numberOfTrees; private int sampleSize; private int shingleSize; private int dimensions; private int outputAfter; private boolean compressed; private boolean partialTreeState; private double boundingBoxCacheFraction; private boolean storeSequenceIndexesEnabled; private boolean compact; private boolean internalShinglingEnabled; private boolean centerOfMassEnabled; private String precision; private PointStoreState pointStoreState; private List<CompactSamplerState> compactSamplerStates; private List<CompactRandomCutTreeState> compactRandomCutTreeStates; private ExecutionContext executionContext; // Mapper options private boolean saveTreeStateEnabled; private boolean saveSamplerStateEnabled; private boolean saveCoordinatorStateEnabled; }
499