File size: 12,210 Bytes
a5ae1ac | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 | """
Tests for the Neuron and NeuronDB.
Verifies against HLD spec:
- Neuron: vector(384) + confidence + successors(10) + predecessors(3) + timestamp + temporal + level
- Confidence: ×1.1 on useful, ×0.9 on useless, capped at ±0.8
- Successors: top-K=10 eviction (new competes for slot, replaces lowest)
- Predecessors: top-3 most recent
- Delete = gone (invariant #3)
- Proximity = connection (spatial search finds nearest)
- Temporal neurons decay with age
"""
import sys
import time
from pathlib import Path
import numpy as np
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
from neuron import (
CONFIDENCE_BOOST, CONFIDENCE_CAP, CONFIDENCE_DECAY, CONFIDENCE_FLOOR,
DEFAULT_CONFIDENCE, MAX_PREDECESSORS, MAX_SUCCESSORS, VECTOR_DIM,
Level, Neuron, NeuronDB,
)
def make_vector(seed: int = 0) -> np.ndarray:
"""Generate a deterministic 384-dim vector."""
rng = np.random.RandomState(seed)
v = rng.randn(VECTOR_DIM).astype(np.float32)
return v / np.linalg.norm(v)
# --- Neuron dataclass tests ---
class TestNeuronUnit:
def test_default_fields(self):
v = make_vector(0)
n = Neuron(id=0, vector=v)
assert n.confidence == DEFAULT_CONFIDENCE
assert n.successors == []
assert n.predecessors == []
assert n.temporal is False
assert n.level == Level.WORD
def test_reinforce_increases_confidence(self):
n = Neuron(id=0, vector=make_vector(0), confidence=0.5)
n.reinforce()
assert abs(n.confidence - 0.5 * CONFIDENCE_BOOST) < 1e-6
def test_reinforce_capped_at_max(self):
n = Neuron(id=0, vector=make_vector(0), confidence=0.79)
n.reinforce() # 0.79 * 1.1 = 0.869 > 0.8
assert n.confidence == CONFIDENCE_CAP
def test_weaken_decreases_confidence(self):
n = Neuron(id=0, vector=make_vector(0), confidence=0.5)
n.weaken()
assert abs(n.confidence - 0.5 * CONFIDENCE_DECAY) < 1e-6
def test_weaken_floored(self):
n = Neuron(id=0, vector=make_vector(0), confidence=-0.75)
n.weaken() # -0.75 * 0.9 = -0.675, but floor check
# Actually -0.75 * 0.9 = -0.675 which is > -0.8, so no floor hit
assert n.confidence > CONFIDENCE_FLOOR
def test_weaken_at_floor(self):
n = Neuron(id=0, vector=make_vector(0), confidence=CONFIDENCE_FLOOR)
n.weaken() # -0.8 * 0.9 = -0.72, which is > -0.8
assert n.confidence >= CONFIDENCE_FLOOR
def test_successor_add(self):
n = Neuron(id=0, vector=make_vector(0))
n.add_successor(1, 0.9)
n.add_successor(2, 0.8)
assert len(n.successors) == 2
assert (1, 0.9) in n.successors
def test_successor_update_existing(self):
n = Neuron(id=0, vector=make_vector(0))
n.add_successor(1, 0.5)
n.add_successor(1, 0.9) # update, not duplicate
assert len(n.successors) == 1
assert n.successors[0] == (1, 0.9)
def test_successor_eviction_at_max(self):
n = Neuron(id=0, vector=make_vector(0))
# Fill to MAX_SUCCESSORS
for i in range(MAX_SUCCESSORS):
n.add_successor(i, 0.1 * (i + 1))
assert len(n.successors) == MAX_SUCCESSORS
# Add one with higher confidence than the lowest
n.add_successor(99, 0.95)
assert len(n.successors) == MAX_SUCCESSORS
# Lowest (id=0, conf=0.1) should be evicted
ids = [s[0] for s in n.successors]
assert 99 in ids
assert 0 not in ids # evicted
def test_successor_no_eviction_if_weaker(self):
n = Neuron(id=0, vector=make_vector(0))
for i in range(MAX_SUCCESSORS):
n.add_successor(i, 0.5)
n.add_successor(99, 0.1) # weaker than all existing
ids = [s[0] for s in n.successors]
assert 99 not in ids
def test_predecessor_add(self):
n = Neuron(id=0, vector=make_vector(0))
n.add_predecessor(1)
n.add_predecessor(2)
assert n.predecessors == [1, 2]
def test_predecessor_max_keeps_recent(self):
n = Neuron(id=0, vector=make_vector(0))
n.add_predecessor(1)
n.add_predecessor(2)
n.add_predecessor(3)
assert len(n.predecessors) == MAX_PREDECESSORS
n.add_predecessor(4)
assert len(n.predecessors) == MAX_PREDECESSORS
assert 1 not in n.predecessors # oldest evicted
assert 4 in n.predecessors
def test_predecessor_no_duplicates(self):
n = Neuron(id=0, vector=make_vector(0))
n.add_predecessor(1)
n.add_predecessor(1)
assert n.predecessors == [1]
def test_effective_confidence_non_temporal(self):
n = Neuron(id=0, vector=make_vector(0), confidence=0.7, temporal=False)
assert n.effective_confidence(current_time=999999) == 0.7
def test_effective_confidence_temporal_decays(self):
now = int(time.time())
n = Neuron(id=0, vector=make_vector(0), confidence=0.7,
temporal=True, timestamp=now - 86400) # 24 hours ago
eff = n.effective_confidence(current_time=now)
assert eff < 0.7 # should decay
assert eff > 0.0 # shouldn't be zero
def test_effective_confidence_temporal_fresh(self):
now = int(time.time())
n = Neuron(id=0, vector=make_vector(0), confidence=0.7,
temporal=True, timestamp=now)
eff = n.effective_confidence(current_time=now)
assert abs(eff - 0.7) < 0.01 # fresh = no decay
def test_level_enum(self):
assert Level.CHARACTER == 0
assert Level.WORD == 1
assert Level.CONCEPT == 2
# --- NeuronDB tests ---
class TestNeuronDB:
def setup_method(self):
self.db = NeuronDB() # in-memory
def teardown_method(self):
self.db.close()
def test_insert_and_get(self):
v = make_vector(42)
neuron = self.db.insert(v, confidence=0.7)
assert neuron.id == 0
assert neuron.confidence == 0.7
retrieved = self.db.get(neuron.id)
assert retrieved is not None
assert retrieved.id == neuron.id
assert abs(retrieved.confidence - 0.7) < 1e-6
def test_insert_auto_increments_id(self):
n0 = self.db.insert(make_vector(0))
n1 = self.db.insert(make_vector(1))
n2 = self.db.insert(make_vector(2))
assert n0.id == 0
assert n1.id == 1
assert n2.id == 2
def test_count(self):
assert self.db.count() == 0
self.db.insert(make_vector(0))
self.db.insert(make_vector(1))
assert self.db.count() == 2
def test_get_nonexistent(self):
assert self.db.get(999) is None
def test_search_finds_nearest(self):
# Insert two distant vectors
v1 = np.zeros(VECTOR_DIM, dtype=np.float32)
v1[0] = 1.0 # unit vector along dim 0
v2 = np.zeros(VECTOR_DIM, dtype=np.float32)
v2[1] = 1.0 # unit vector along dim 1
self.db.insert(v1, confidence=0.5) # id=0
self.db.insert(v2, confidence=0.5) # id=1
# Query near v1
query = np.zeros(VECTOR_DIM, dtype=np.float32)
query[0] = 0.9
query[2] = 0.1
results = self.db.search(query, k=1)
assert len(results) == 1
assert results[0].id == 0 # closer to v1
def test_search_returns_k_results(self):
for i in range(20):
self.db.insert(make_vector(i))
results = self.db.search(make_vector(5), k=5)
assert len(results) == 5
def test_search_k_larger_than_db(self):
self.db.insert(make_vector(0))
self.db.insert(make_vector(1))
results = self.db.search(make_vector(0), k=10)
assert len(results) == 2 # only 2 exist
def test_search_empty_db(self):
results = self.db.search(make_vector(0), k=5)
assert results == []
def test_delete_removes_neuron(self):
"""Invariant #3: Delete = gone. No retraining."""
n = self.db.insert(make_vector(42))
assert self.db.get(n.id) is not None
assert self.db.count() == 1
deleted = self.db.delete(n.id)
assert deleted is True
assert self.db.get(n.id) is None
assert self.db.count() == 0
def test_delete_removes_from_search(self):
"""After delete, search should not find the neuron."""
v = np.zeros(VECTOR_DIM, dtype=np.float32)
v[0] = 1.0
n = self.db.insert(v)
# Searchable before delete
results = self.db.search(v, k=1)
assert len(results) == 1
self.db.delete(n.id)
# Gone from search after delete
results = self.db.search(v, k=1)
assert len(results) == 0
def test_delete_nonexistent(self):
assert self.db.delete(999) is False
def test_update_confidence_useful(self):
n = self.db.insert(make_vector(0), confidence=0.5)
self.db.update_confidence(n.id, useful=True)
updated = self.db.get(n.id)
assert abs(updated.confidence - 0.5 * CONFIDENCE_BOOST) < 1e-6
def test_update_confidence_not_useful(self):
n = self.db.insert(make_vector(0), confidence=0.5)
self.db.update_confidence(n.id, useful=False)
updated = self.db.get(n.id)
assert abs(updated.confidence - 0.5 * CONFIDENCE_DECAY) < 1e-6
def test_update_confidence_respects_cap(self):
n = self.db.insert(make_vector(0), confidence=0.79)
self.db.update_confidence(n.id, useful=True)
updated = self.db.get(n.id)
assert updated.confidence == CONFIDENCE_CAP
def test_update_successors(self):
n = self.db.insert(make_vector(0))
other = self.db.insert(make_vector(1))
self.db.update_successors(n.id, other.id, 0.85)
updated = self.db.get(n.id)
assert len(updated.successors) == 1
assert updated.successors[0][0] == other.id
assert abs(updated.successors[0][1] - 0.85) < 1e-5
def test_update_predecessors(self):
n = self.db.insert(make_vector(0))
pred = self.db.insert(make_vector(1))
self.db.update_predecessors(n.id, pred.id)
updated = self.db.get(n.id)
assert pred.id in updated.predecessors
def test_successor_persistence_through_get(self):
"""Successors survive serialize→deserialize round-trip."""
n = self.db.insert(make_vector(0))
self.db.update_successors(n.id, 10, 0.9)
self.db.update_successors(n.id, 20, 0.8)
self.db.update_successors(n.id, 30, 0.7)
retrieved = self.db.get(n.id)
assert len(retrieved.successors) == 3
ids = [s[0] for s in retrieved.successors]
assert 10 in ids
assert 20 in ids
assert 30 in ids
def test_vector_normalized_on_insert(self):
"""Vectors should be L2-normalized for cosine similarity."""
v = np.ones(VECTOR_DIM, dtype=np.float32) * 5.0 # not normalized
n = self.db.insert(v)
retrieved = self.db.get(n.id)
norm = np.linalg.norm(retrieved.vector)
assert abs(norm - 1.0) < 1e-5
def test_multiple_deletes_and_inserts(self):
"""DB stays consistent through insert-delete-insert cycles."""
n0 = self.db.insert(make_vector(0))
n1 = self.db.insert(make_vector(1))
n2 = self.db.insert(make_vector(2))
assert self.db.count() == 3
self.db.delete(n1.id)
assert self.db.count() == 2
n3 = self.db.insert(make_vector(3))
assert self.db.count() == 3
# n1 gone, others present
assert self.db.get(n0.id) is not None
assert self.db.get(n1.id) is None
assert self.db.get(n2.id) is not None
assert self.db.get(n3.id) is not None
def test_level_persists(self):
n = self.db.insert(make_vector(0), level=Level.CONCEPT)
retrieved = self.db.get(n.id)
assert retrieved.level == Level.CONCEPT
def test_temporal_persists(self):
n = self.db.insert(make_vector(0), temporal=True)
retrieved = self.db.get(n.id)
assert retrieved.temporal is True
# --- Run ---
if __name__ == "__main__":
import pytest
pytest.main([__file__, "-v"])
|