File size: 11,728 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 | """
Tests for SparseConvergenceLoop and SparseMultiHop.
Verifies that sparse co-occurrence convergence produces the same
logical behavior as the dense convergence loop:
- Converges on related words
- Does not converge on empty/unknown input
- Query anchor prevents drift
- Per-hop specialization (early=broad, late=narrow)
- Mutual attention boosts coherent clusters
- Softmax-weighted blending
- Multi-hop finds distant concepts
- Trace is inspectable
"""
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
from sparse_convergence import (
SparseConvergenceLoop, SparseMultiHop,
sparse_cosine, sparse_blend, sparse_norm, sparse_normalize,
SparseConvergenceResult, SparseMultiHopResult,
)
def make_graph():
"""Build a small co-occurrence graph for testing.
Words: paris(0), capital(1), france(2), london(3), england(4),
river(5), seine(6), thames(7), bridge(8), tower(9)
Co-occurrence edges encode knowledge:
paris <-> capital, france, seine, bridge
london <-> capital, england, thames, tower, bridge
capital <-> paris, london, france, england
seine <-> paris, river
thames <-> london, river
"""
words = ["paris", "capital", "france", "london", "england",
"river", "seine", "thames", "bridge", "tower"]
word_idx = {w: i for i, w in enumerate(words)}
word_neurons = {w: i + 100 for i, w in enumerate(words)}
cooc = {}
for i in range(len(words)):
cooc[i] = {i: 1.0}
def link(a, b, weight=0.3):
cooc[a][b] = cooc[a].get(b, 0) + weight
cooc[b][a] = cooc[b].get(a, 0) + weight
# paris cluster
link(0, 1, 0.5) # paris-capital
link(0, 2, 0.6) # paris-france
link(0, 6, 0.4) # paris-seine
link(0, 8, 0.2) # paris-bridge
# london cluster
link(3, 1, 0.5) # london-capital
link(3, 4, 0.6) # london-england
link(3, 7, 0.4) # london-thames
link(3, 9, 0.3) # london-tower
link(3, 8, 0.2) # london-bridge
# shared
link(1, 2, 0.4) # capital-france
link(1, 4, 0.4) # capital-england
# river connections
link(5, 6, 0.5) # river-seine
link(5, 7, 0.5) # river-thames
return cooc, word_idx, words, word_neurons
def make_loop(cooc, word_idx, words, word_neurons, **kwargs):
defaults = dict(max_hops=10, k=5, convergence_threshold=0.99,
min_confidence=0.05, min_relevance=0.1, temperature=1.0)
defaults.update(kwargs)
return SparseConvergenceLoop(
cooc=cooc, word_idx=word_idx, words=words,
word_neurons=word_neurons, **defaults
)
class TestSparseConvergenceBasic:
def test_converges_on_related_words(self):
"""Query [paris] should converge and find france/capital."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
result = loop.converge([word_idx["paris"]])
assert result.converged is True
assert len(result.concepts) > 0
assert result.confidence > 0
def test_empty_query_does_not_converge(self):
"""Empty query should not converge."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
result = loop.converge([])
assert result.converged is False
assert result.concepts == []
def test_unknown_index_does_not_crash(self):
"""Index not in cooc should handle gracefully."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
result = loop.converge([999])
assert result.converged is False
def test_convergence_finds_correct_cluster(self):
"""Query [paris, france] should find paris-related words, not london."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
result = loop.converge([word_idx["paris"], word_idx["france"]])
concept_indices = {widx for widx, _ in result.concepts}
# Should find paris-cluster words
assert word_idx["capital"] in concept_indices or word_idx["seine"] in concept_indices
class TestSparseAnchor:
def test_anchor_keeps_query_relevant(self):
"""With query anchor, result should stay near query, not drift."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons, max_hops=10)
result = loop.converge([word_idx["paris"]])
if result.converged and result.hops:
# Last hop's current profile should still have paris-related keys
last_profile = result.hops[-1].current
# Paris (idx 0) should still have weight in the profile
assert last_profile.get(0, 0) > 0 or last_profile.get(2, 0) > 0
def test_movement_decreases(self):
"""Movement should generally decrease (convergence)."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons, max_hops=10)
result = loop.converge([word_idx["paris"]])
if len(result.hops) >= 3:
first = result.hops[0].movement
last = result.hops[-1].movement
assert last <= first + 0.05 # tolerance
class TestSparseMutualAttention:
def test_coherent_cluster_boosted(self):
"""Words that co-occur with each other should get boosted."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
# paris(0) and france(2) mutually co-occur
neighbors = [(0, 0.5), (2, 0.4), (5, 0.3)] # paris, france, river
boosted = loop._mutual_attention(neighbors)
# paris and france should be boosted more than river
paris_sim = next(s for w, s in boosted if w == 0)
river_sim = next(s for w, s in boosted if w == 5)
assert paris_sim > river_sim
class TestSparseSoftmaxBlend:
def test_high_similarity_dominates(self):
"""Higher similarity word should dominate the blend."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons, temperature=0.5)
neighbors = [(0, 0.9), (3, 0.1)] # paris strong, london weak
blended = loop._softmax_blend(neighbors)
# Blended profile should lean toward paris's co-occurrences
# paris has france(2), london has england(4)
paris_weight = blended.get(2, 0) # france
london_weight = blended.get(4, 0) # england
assert paris_weight > london_weight
def test_uniform_at_inf_temperature(self):
"""Infinite temperature should give uniform weighting."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons,
temperature=float('inf'))
neighbors = [(0, 0.9), (3, 0.1)]
blended = loop._softmax_blend(neighbors)
# Both should contribute roughly equally
assert len(blended) > 0
class TestSparseTrace:
def test_trace_has_hops(self):
"""Trace should record each hop."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
result = loop.converge([word_idx["paris"]])
assert len(result.hops) > 0
for hop in result.hops:
assert hop.hop_number >= 0
assert len(hop.neighbors) > 0
def test_trace_string_readable(self):
"""Trace should produce human-readable string."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
result = loop.converge([word_idx["paris"]])
trace_str = result.trace()
assert "SparseConvergence:" in trace_str
assert "Hop 0:" in trace_str
class TestSparseMultiHop:
def test_single_hop_works(self):
"""Single query should work through multi-hop."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
mh = SparseMultiHop(loop, max_rounds=3)
result = mh.reason([word_idx["paris"]])
assert len(result.rounds) >= 1
assert len(result.concepts) > 0
def test_multi_hop_discovers_distant_concept(self):
"""Multi-hop from [seine] should eventually find [london] via river->thames."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons,
min_relevance=0.05, min_confidence=0.01)
mh = SparseMultiHop(loop, max_rounds=3, concept_blend_weight=0.5)
result = mh.reason([word_idx["seine"]])
concept_indices = {widx for widx, _ in result.concepts}
# Should discover river or thames through the graph
found_river_cluster = (word_idx["river"] in concept_indices or
word_idx["thames"] in concept_indices)
assert found_river_cluster, f"Expected to find river/thames, got {concept_indices}"
def test_no_duplicate_concepts(self):
"""Same concept should not appear twice across rounds."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
mh = SparseMultiHop(loop, max_rounds=3)
result = mh.reason([word_idx["paris"]])
indices = [widx for widx, _ in result.concepts]
assert len(indices) == len(set(indices))
def test_max_rounds_respected(self):
"""Should not exceed max_rounds."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
mh = SparseMultiHop(loop, max_rounds=2)
result = mh.reason([word_idx["paris"]])
assert len(result.rounds) <= 2
def test_multi_hop_trace_readable(self):
"""Multi-hop trace should be human-readable."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
mh = SparseMultiHop(loop, max_rounds=3)
result = mh.reason([word_idx["paris"]])
trace_str = result.trace()
assert "SparseMultiHop:" in trace_str
assert "Round 1" in trace_str
def test_empty_query_multi_hop(self):
"""Empty query should not crash."""
cooc, word_idx, words, word_neurons = make_graph()
loop = make_loop(cooc, word_idx, words, word_neurons)
mh = SparseMultiHop(loop, max_rounds=3)
result = mh.reason([])
assert result.converged is False
assert result.concepts == []
class TestSparseUtils:
def test_sparse_cosine_identical(self):
a = {0: 1.0, 1: 2.0}
assert abs(sparse_cosine(a, a) - 1.0) < 1e-6
def test_sparse_cosine_orthogonal(self):
a = {0: 1.0}
b = {1: 1.0}
assert sparse_cosine(a, b) == 0.0
def test_sparse_cosine_empty(self):
assert sparse_cosine({}, {0: 1.0}) == 0.0
assert sparse_cosine({}, {}) == 0.0
def test_sparse_blend_uniform(self):
p1 = {0: 1.0, 1: 2.0}
p2 = {1: 4.0, 2: 6.0}
blended = sparse_blend([p1, p2])
assert abs(blended[0] - 0.5) < 1e-6
assert abs(blended[1] - 3.0) < 1e-6
assert abs(blended[2] - 3.0) < 1e-6
def test_sparse_normalize(self):
d = {0: 3.0, 1: 4.0}
n = sparse_normalize(d)
assert abs(sparse_norm(n) - 1.0) < 1e-6
if __name__ == "__main__":
import pytest
pytest.main([__file__, "-v"])
|