Spaces:
Running
Running
File size: 10,743 Bytes
a9c06ad | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 | # backend/tests/integration/test_raptor.py
# Integration tests for RAPTOR hierarchical summarisation.
#
# Task 8: Validates that the RAPTOR builder produces coherent hierarchies
# with proper clustering, summarisation, and embedding integration.
#
# Tests run with synthetic corpus fixtures to avoid dependency on real
# knowledge base content.
import os
import sys
import pytest
import numpy as np
from unittest.mock import AsyncMock, MagicMock, patch
# Add parent directory to path so ingestion module is accessible
sys.path.insert(0, os.path.join(os.path.dirname(__file__), '../../..'))
from ingestion.raptor import RaptorBuilder, _n_clusters, _gmm_soft_assign
class TestRaptorClustering:
"""Unit tests for RAPTOR clustering logic."""
def test_n_clusters_formula(self):
"""sqrt(N) heuristic with bounds."""
assert _n_clusters(4) == 2
assert _n_clusters(100) == 10
assert _n_clusters(400) == 20
assert _n_clusters(500) == 20
assert _n_clusters(1) == 2
def test_gmm_soft_assign_shape(self):
"""GMM returns correct shapes for responsibilities and labels."""
rng = np.random.default_rng(seed=42)
embeddings = rng.standard_normal((20, 384))
labels, responsibilities = _gmm_soft_assign(embeddings, n_components=3)
assert labels.shape == (20,)
assert responsibilities.shape == (20, 3)
assert np.all((labels >= 0) & (labels < 3))
assert np.allclose(responsibilities.sum(axis=1), 1.0)
def test_gmm_cluster_determinism(self):
"""GMM with fixed random_state is deterministic."""
rng = np.random.default_rng(seed=42)
embeddings = rng.standard_normal((15, 384))
labels1, _ = _gmm_soft_assign(embeddings, n_components=2, random_state=42)
labels2, _ = _gmm_soft_assign(embeddings, n_components=2, random_state=42)
np.testing.assert_array_equal(labels1, labels2)
class TestRaptorSummarisation:
"""Integration tests for RAPTOR cluster summarisation."""
@pytest.fixture
def synthetic_chunks(self):
"""10-item fixture: 5 project chunks + 5 blog chunks."""
return [
{
"id": f"chunk_{i}",
"text": f"Project {i}: Built a Python async service using FastAPI and PostgreSQL. "
f"Key features include real-time validation, caching layers, and REST API.",
"metadata": {
"doc_id": f"project_{i % 3}",
"source_title": f"Project {i % 3}",
"source_type": "project",
"chunk_index": i,
},
}
for i in range(5)
] + [
{
"id": f"blog_{i}",
"text": f"Blog Post {i}: Exploring RAG systems with LangGraph, semantic caching, "
f"and multi-modal retrieval. Discusses production challenges and solutions.",
"metadata": {
"doc_id": f"blog_{i}",
"source_title": f"Blog {i}",
"source_type": "blog",
"chunk_index": i,
},
}
for i in range(5)
]
@pytest.fixture
def synthetic_embeddings(self):
"""10 random 384-dim vectors (BGE-small dimension)."""
rng = np.random.default_rng(seed=42)
return rng.standard_normal((10, 384)).astype(np.float32)
def test_raptor_builder_initialization(self):
"""RaptorBuilder instantiates without errors."""
mock_vector_store = MagicMock()
mock_embedder = MagicMock()
mock_gemini = MagicMock()
builder = RaptorBuilder(
store=mock_vector_store,
embedder=mock_embedder,
gemini_client=mock_gemini,
)
assert builder._store is mock_vector_store
@pytest.mark.asyncio
async def test_raptor_build_creates_hierarchy(
self,
synthetic_chunks,
synthetic_embeddings,
):
"""
RAPTOR build produces hierarchical summary nodes.
Assertions:
• Cluster count is sqrt(N) within bounds
• No degenerate single-item clusters
• Summary nodes are created and upserted
"""
mock_vector_store = MagicMock()
mock_embedder = MagicMock()
mock_gemini = MagicMock()
def mock_summarise(text: str):
return "Summary of cluster content"
mock_gemini.summarise = AsyncMock(side_effect=mock_summarise)
# Mock embedder to return synthetic vectors
def mock_embed(texts, is_query=False):
rng = np.random.default_rng(seed=42)
return rng.standard_normal((len(texts), 384)).astype(np.float32)
mock_embedder.embed = AsyncMock(side_effect=mock_embed)
mock_embedder.embed_texts_async = mock_embedder.embed
# Mock vector store to capture upserts
upserted_count = [0]
def capture_upsert(nodes, dense_embeddings, sparse_embeddings=None):
# Detect raptor_summary nodes by inspecting their metadata.
raptor_nodes = [
n for n in nodes
if n.get("metadata", {}).get("chunk_type") == "raptor_summary"
]
if raptor_nodes:
upserted_count[0] = len(raptor_nodes)
return [f"uuid_{i}" for i in range(len(nodes))]
mock_vector_store.upsert_chunks = MagicMock(side_effect=capture_upsert)
builder = RaptorBuilder(
store=mock_vector_store,
embedder=mock_embedder,
gemini_client=mock_gemini,
)
leaf_uuids = [f"uuid_chunk_{i}" for i in range(len(synthetic_chunks))]
await builder.build(
leaf_chunks=synthetic_chunks,
dense_embeddings=synthetic_embeddings.tolist(),
leaf_uuids=leaf_uuids,
)
# At least one summary node should be created
assert upserted_count[0] > 0 or len(synthetic_chunks) < 2
@pytest.mark.asyncio
async def test_raptor_child_leaf_mapping(self, synthetic_chunks, synthetic_embeddings):
"""Child leaf IDs correctly reference original chunks."""
mock_vector_store = MagicMock()
mock_embedder = MagicMock()
mock_gemini = MagicMock()
def mock_summarise(text: str):
return "Cluster summary"
mock_gemini.summarise = AsyncMock(side_effect=mock_summarise)
def mock_embed(texts, is_query=False):
rng = np.random.default_rng(seed=43)
return rng.standard_normal((len(texts), 384)).astype(np.float32)
mock_embedder.embed = AsyncMock(side_effect=mock_embed)
mock_embedder.embed_texts_async = mock_embedder.embed
# Capture child_leaf_ids for validation
captured_mappings = []
def capture_upsert(nodes, dense_embeddings, sparse_embeddings=None):
for node in nodes:
if node.get("metadata", {}).get("chunk_type") == "raptor_summary":
child_ids = node.get("metadata", {}).get("child_leaf_ids", [])
captured_mappings.append(child_ids)
return [f"uuid_{i}" for i in range(len(nodes))]
mock_vector_store.upsert_chunks = MagicMock(side_effect=capture_upsert)
builder = RaptorBuilder(
store=mock_vector_store,
embedder=mock_embedder,
gemini_client=mock_gemini,
)
leaf_uuids = [f"uuid_chunk_{i}" for i in range(len(synthetic_chunks))]
await builder.build(
leaf_chunks=synthetic_chunks,
dense_embeddings=synthetic_embeddings.tolist(),
leaf_uuids=leaf_uuids,
)
# All child references should use leaf UUIDs
for child_list in captured_mappings:
for child_uuid in child_list:
assert child_uuid in leaf_uuids
def test_raptor_builder_store_reference(self):
"""RaptorBuilder stores reference to vector store."""
mock_vector_store = MagicMock()
mock_embedder = MagicMock()
builder = RaptorBuilder(
store=mock_vector_store,
embedder=mock_embedder,
)
assert builder._store is mock_vector_store
class TestRaptorErrorHandling:
"""Robustness tests for RAPTOR failure modes."""
@pytest.mark.asyncio
async def test_raptor_graceful_gemini_failure(self):
"""If Gemini fails, RAPTOR continues with fallback summary."""
mock_vector_store = MagicMock()
mock_embedder = MagicMock()
mock_gemini = MagicMock()
def mock_summarise_fail(text: str):
raise RuntimeError("Gemini API timeout")
mock_gemini.summarise = AsyncMock(side_effect=mock_summarise_fail)
def mock_embed(texts, is_query=False):
rng = np.random.default_rng(seed=44)
return rng.standard_normal((len(texts), 384)).astype(np.float32)
mock_embedder.embed = AsyncMock(side_effect=mock_embed)
mock_embedder.embed_texts_async = mock_embedder.embed
mock_vector_store.upsert_chunks = MagicMock(return_value=[])
builder = RaptorBuilder(
store=mock_vector_store,
embedder=mock_embedder,
gemini_client=mock_gemini,
)
chunks = [
{
"id": "c1",
"text": "Sample chunk about project architecture",
"metadata": {"doc_id": "d1", "source_type": "blog"},
}
]
rng = np.random.default_rng(seed=42)
embeddings = rng.standard_normal((1, 384)).astype(np.float32)
# Should handle gracefully
try:
await builder.build(
leaf_chunks=chunks,
dense_embeddings=embeddings.tolist(),
leaf_uuids=["uuid_c1"],
)
except Exception:
pytest.fail("RAPTOR should handle Gemini failure gracefully")
@pytest.mark.asyncio
async def test_raptor_empty_corpus(self):
"""Empty chunk list skips RAPTOR."""
mock_vector_store = MagicMock()
mock_embedder = MagicMock()
mock_vector_store.upsert_chunks = MagicMock(return_value={})
builder = RaptorBuilder(
store=mock_vector_store,
embedder=mock_embedder,
)
await builder.build(
leaf_chunks=[],
dense_embeddings=[],
leaf_uuids=[],
)
# Should complete without error
assert mock_vector_store.upsert_chunks.call_count == 0 or len(
mock_vector_store.upsert_chunks.call_args_list[0][0][0]
) == 0
|