Add Ikario Architecture v2 - Phases 1-8 complete
Implements the processual architecture based on Whitehead's Process Philosophy and Peirce's Semiotics. Core paradigm: "L'espace latent pense. Le LLM traduit." (The latent space thinks. The LLM translates.) Phase 1-4: Core semiotic cycle - StateTensor 8x1024 (8 Peircean dimensions) - Dissonance computation with hard negatives - Fixation via 4 Peircean methods (Tenacity, Authority, A Priori, Science) - LatentEngine orchestrating the full cycle Phase 5: StateToLanguage - LLM as pure translator (zero-reasoning, T=0) - Projection on interpretable directions - Reasoning markers detection (Amendment #4) Phase 6: Vigilance - x_ref (David) as guard-rail, NOT attractor - Drift detection per dimension and globally - Alerts: ok, warning, critical Phase 7: Autonomous Daemon - Two modes: CONVERSATION (always verbalize), AUTONOMOUS (~1000 cycles/day) - Amendment #5: 50% probability on unresolved impacts - TriggerGenerator with weighted random selection Phase 8: Integration & Metrics - ProcessMetrics for daily/weekly reports - Health status monitoring - Integration tests validating all modules 297 tests passing, version 0.7.0 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
371
ikario_processual/tests/test_dissonance.py
Normal file
371
ikario_processual/tests/test_dissonance.py
Normal file
@@ -0,0 +1,371 @@
|
||||
#!/usr/bin/env python3
|
||||
"""
|
||||
Tests pour le module de dissonance - Phase 2.
|
||||
|
||||
Exécuter: pytest ikario_processual/tests/test_dissonance.py -v
|
||||
"""
|
||||
|
||||
import numpy as np
|
||||
import pytest
|
||||
from datetime import datetime
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
sys.path.insert(0, str(Path(__file__).parent.parent.parent))
|
||||
|
||||
from ikario_processual.state_tensor import StateTensor, DIMENSION_NAMES, EMBEDDING_DIM
|
||||
from ikario_processual.dissonance import (
|
||||
DissonanceConfig,
|
||||
DissonanceResult,
|
||||
compute_dissonance,
|
||||
compute_dissonance_enhanced,
|
||||
compute_self_dissonance,
|
||||
cosine_similarity,
|
||||
Impact,
|
||||
create_impact_from_dissonance,
|
||||
)
|
||||
|
||||
|
||||
def create_random_tensor() -> StateTensor:
|
||||
"""Crée un tenseur avec des vecteurs aléatoires normalisés."""
|
||||
tensor = StateTensor(
|
||||
state_id=0,
|
||||
timestamp=datetime.now().isoformat(),
|
||||
)
|
||||
for dim_name in DIMENSION_NAMES:
|
||||
v = np.random.randn(EMBEDDING_DIM)
|
||||
v = v / np.linalg.norm(v)
|
||||
setattr(tensor, dim_name, v)
|
||||
return tensor
|
||||
|
||||
|
||||
def create_zero_tensor() -> StateTensor:
|
||||
"""Crée un tenseur avec des vecteurs zéro."""
|
||||
return StateTensor(
|
||||
state_id=0,
|
||||
timestamp=datetime.now().isoformat(),
|
||||
)
|
||||
|
||||
|
||||
class TestCosineSimiliarity:
|
||||
"""Tests pour la fonction cosine_similarity."""
|
||||
|
||||
def test_identical_vectors(self):
|
||||
"""Vecteurs identiques → similarité = 1."""
|
||||
v = np.random.randn(EMBEDDING_DIM)
|
||||
v = v / np.linalg.norm(v)
|
||||
assert np.isclose(cosine_similarity(v, v), 1.0)
|
||||
|
||||
def test_opposite_vectors(self):
|
||||
"""Vecteurs opposés → similarité = -1."""
|
||||
v = np.random.randn(EMBEDDING_DIM)
|
||||
v = v / np.linalg.norm(v)
|
||||
assert np.isclose(cosine_similarity(v, -v), -1.0)
|
||||
|
||||
def test_orthogonal_vectors(self):
|
||||
"""Vecteurs orthogonaux → similarité ≈ 0."""
|
||||
v1 = np.zeros(EMBEDDING_DIM)
|
||||
v1[0] = 1.0
|
||||
v2 = np.zeros(EMBEDDING_DIM)
|
||||
v2[1] = 1.0
|
||||
assert np.isclose(cosine_similarity(v1, v2), 0.0)
|
||||
|
||||
def test_zero_vector(self):
|
||||
"""Vecteur zéro → similarité = 0."""
|
||||
v1 = np.random.randn(EMBEDDING_DIM)
|
||||
v2 = np.zeros(EMBEDDING_DIM)
|
||||
assert cosine_similarity(v1, v2) == 0.0
|
||||
|
||||
|
||||
class TestDissonanceConfig:
|
||||
"""Tests pour DissonanceConfig."""
|
||||
|
||||
def test_default_weights_sum(self):
|
||||
"""Les poids par défaut doivent sommer à ~1.0."""
|
||||
config = DissonanceConfig()
|
||||
weights = config.get_dimension_weights()
|
||||
total = sum(weights.values())
|
||||
assert np.isclose(total, 1.0), f"Total des poids: {total}"
|
||||
|
||||
def test_all_dimensions_have_weight(self):
|
||||
"""Chaque dimension doit avoir un poids."""
|
||||
config = DissonanceConfig()
|
||||
weights = config.get_dimension_weights()
|
||||
for dim in DIMENSION_NAMES:
|
||||
assert dim in weights
|
||||
assert weights[dim] >= 0
|
||||
|
||||
|
||||
class TestComputeDissonance:
|
||||
"""Tests pour compute_dissonance (version basique)."""
|
||||
|
||||
def test_self_dissonance_is_zero(self):
|
||||
"""E(X_t, X_t) ≈ 0."""
|
||||
X_t = create_random_tensor()
|
||||
|
||||
# Utiliser une dimension comme input (simuler entrée identique)
|
||||
e_input = X_t.firstness.copy()
|
||||
|
||||
result = compute_dissonance(e_input, X_t)
|
||||
|
||||
# La dissonance avec firstness devrait être ~0
|
||||
assert result.dissonances_by_dimension['firstness'] < 0.01
|
||||
|
||||
def test_orthogonal_input_high_dissonance(self):
|
||||
"""Entrée orthogonale → haute dissonance."""
|
||||
X_t = create_random_tensor()
|
||||
|
||||
# Créer un vecteur orthogonal (difficile en haute dimension, mais différent)
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
result = compute_dissonance(e_input, X_t)
|
||||
|
||||
# La dissonance totale devrait être significative
|
||||
assert result.total > 0.1
|
||||
|
||||
def test_result_structure(self):
|
||||
"""Vérifier la structure du résultat."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
result = compute_dissonance(e_input, X_t)
|
||||
|
||||
assert isinstance(result, DissonanceResult)
|
||||
assert hasattr(result, 'total')
|
||||
assert hasattr(result, 'is_choc')
|
||||
assert hasattr(result, 'dissonances_by_dimension')
|
||||
assert len(result.dissonances_by_dimension) == 8
|
||||
|
||||
def test_is_choc_flag(self):
|
||||
"""Le flag is_choc dépend du seuil."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
# Seuil bas → plus de chocs
|
||||
config_low = DissonanceConfig(choc_threshold=0.1)
|
||||
result_low = compute_dissonance(e_input, X_t, config_low)
|
||||
|
||||
# Seuil haut → moins de chocs
|
||||
config_high = DissonanceConfig(choc_threshold=0.9)
|
||||
result_high = compute_dissonance(e_input, X_t, config_high)
|
||||
|
||||
# Avec seuil bas, plus probable d'avoir un choc
|
||||
assert result_low.is_choc or result_high.is_choc is False
|
||||
|
||||
|
||||
class TestComputeDissonanceEnhanced:
|
||||
"""Tests pour compute_dissonance_enhanced avec hard negatives."""
|
||||
|
||||
def test_no_rag_results(self):
|
||||
"""Sans résultats RAG → novelty_penalty = 1.0."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
result = compute_dissonance_enhanced(e_input, X_t, rag_results=[])
|
||||
|
||||
assert result.novelty_penalty == 1.0
|
||||
assert result.rag_results_count == 0
|
||||
|
||||
def test_with_similar_rag_results(self):
|
||||
"""Avec résultats RAG similaires → faible novelty."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
# Créer des résultats RAG très similaires (copie avec très peu de bruit)
|
||||
rag_results = [
|
||||
{'vector': e_input.copy(), 'content': 'identical'},
|
||||
{'vector': e_input + np.random.randn(EMBEDDING_DIM) * 0.01, 'content': 'similar'},
|
||||
]
|
||||
# Normaliser les vecteurs RAG
|
||||
for r in rag_results:
|
||||
r['vector'] = r['vector'] / np.linalg.norm(r['vector'])
|
||||
|
||||
result = compute_dissonance_enhanced(e_input, X_t, rag_results)
|
||||
|
||||
# Le premier vecteur est identique donc max_sim ~= 1.0
|
||||
assert result.max_similarity_to_corpus > 0.9
|
||||
assert result.novelty_penalty == 0.0 # Pas de pénalité si > 0.3
|
||||
|
||||
def test_hard_negatives_detection(self):
|
||||
"""Détection des hard negatives (similarité < seuil)."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
# Créer un vecteur opposé (hard negative)
|
||||
opposite = -e_input
|
||||
|
||||
rag_results = [
|
||||
{'vector': opposite, 'content': 'contradiction', 'source': 'test'},
|
||||
{'vector': e_input, 'content': 'similar', 'source': 'test'},
|
||||
]
|
||||
|
||||
result = compute_dissonance_enhanced(e_input, X_t, rag_results)
|
||||
|
||||
# Au moins un hard negative devrait être détecté
|
||||
assert len(result.hard_negatives) >= 1
|
||||
assert result.contradiction_score > 0
|
||||
|
||||
def test_total_dissonance_combines_all(self):
|
||||
"""La dissonance totale combine base + contradiction + novelty."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
config = DissonanceConfig(
|
||||
contradiction_weight=0.2,
|
||||
novelty_weight=0.1
|
||||
)
|
||||
|
||||
result = compute_dissonance_enhanced(e_input, X_t, [], config)
|
||||
|
||||
expected_total = (
|
||||
result.base_dissonance +
|
||||
config.contradiction_weight * result.contradiction_score +
|
||||
config.novelty_weight * result.novelty_penalty
|
||||
)
|
||||
|
||||
assert np.isclose(result.total, expected_total)
|
||||
|
||||
|
||||
class TestSelfDissonance:
|
||||
"""Tests pour compute_self_dissonance."""
|
||||
|
||||
def test_coherent_tensor(self):
|
||||
"""Tenseur cohérent → faible dissonance interne."""
|
||||
# Créer un tenseur où toutes les dimensions sont identiques
|
||||
base_vector = np.random.randn(EMBEDDING_DIM)
|
||||
base_vector = base_vector / np.linalg.norm(base_vector)
|
||||
|
||||
tensor = StateTensor(
|
||||
state_id=0,
|
||||
timestamp=datetime.now().isoformat(),
|
||||
)
|
||||
for dim_name in DIMENSION_NAMES:
|
||||
# Utiliser le même vecteur (parfaitement cohérent)
|
||||
setattr(tensor, dim_name, base_vector.copy())
|
||||
|
||||
dissonance = compute_self_dissonance(tensor)
|
||||
|
||||
# Devrait être zéro car toutes les dimensions sont identiques
|
||||
assert dissonance < 0.01
|
||||
|
||||
def test_incoherent_tensor(self):
|
||||
"""Tenseur incohérent → haute dissonance interne."""
|
||||
tensor = create_random_tensor() # Dimensions aléatoires = incohérent
|
||||
|
||||
dissonance = compute_self_dissonance(tensor)
|
||||
|
||||
# Devrait être plus élevé
|
||||
assert dissonance > 0.3
|
||||
|
||||
|
||||
class TestImpact:
|
||||
"""Tests pour la création d'Impact."""
|
||||
|
||||
def test_create_impact_from_dissonance(self):
|
||||
"""Créer un Impact à partir d'un résultat de dissonance."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
dissonance_result = compute_dissonance(e_input, X_t)
|
||||
|
||||
impact = create_impact_from_dissonance(
|
||||
dissonance=dissonance_result,
|
||||
trigger_type='user',
|
||||
trigger_content='Test message',
|
||||
trigger_vector=e_input,
|
||||
state_id=0,
|
||||
impact_id=1,
|
||||
)
|
||||
|
||||
assert impact.impact_id == 1
|
||||
assert impact.trigger_type == 'user'
|
||||
assert impact.dissonance_total == dissonance_result.total
|
||||
assert impact.resolved is False
|
||||
|
||||
def test_impact_to_dict(self):
|
||||
"""Impact.to_dict() retourne un dictionnaire valide."""
|
||||
impact = Impact(
|
||||
impact_id=1,
|
||||
timestamp=datetime.now().isoformat(),
|
||||
state_id_at_impact=0,
|
||||
trigger_type='user',
|
||||
trigger_content='Test',
|
||||
dissonance_total=0.5,
|
||||
)
|
||||
|
||||
d = impact.to_dict()
|
||||
|
||||
assert 'impact_id' in d
|
||||
assert 'timestamp' in d
|
||||
assert d['timestamp'].endswith('Z')
|
||||
assert d['resolved'] is False
|
||||
|
||||
|
||||
class TestDissonanceMonotonicity:
|
||||
"""Tests de monotonie de la dissonance."""
|
||||
|
||||
def test_more_different_more_dissonance(self):
|
||||
"""Plus différent = plus de dissonance."""
|
||||
X_t = create_random_tensor()
|
||||
|
||||
# Entrée identique à une dimension
|
||||
identical = X_t.firstness.copy()
|
||||
result_identical = compute_dissonance(identical, X_t)
|
||||
|
||||
# Entrée légèrement différente
|
||||
slightly_different = X_t.firstness + np.random.randn(EMBEDDING_DIM) * 0.1
|
||||
slightly_different = slightly_different / np.linalg.norm(slightly_different)
|
||||
result_slight = compute_dissonance(slightly_different, X_t)
|
||||
|
||||
# Entrée très différente
|
||||
very_different = np.random.randn(EMBEDDING_DIM)
|
||||
very_different = very_different / np.linalg.norm(very_different)
|
||||
result_very = compute_dissonance(very_different, X_t)
|
||||
|
||||
# Vérifier la monotonie sur la dimension firstness
|
||||
assert result_identical.dissonances_by_dimension['firstness'] < \
|
||||
result_slight.dissonances_by_dimension['firstness']
|
||||
|
||||
|
||||
class TestDissonanceResultSerialization:
|
||||
"""Tests de sérialisation."""
|
||||
|
||||
def test_to_dict(self):
|
||||
"""DissonanceResult.to_dict() fonctionne."""
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
result = compute_dissonance(e_input, X_t)
|
||||
d = result.to_dict()
|
||||
|
||||
assert 'total' in d
|
||||
assert 'is_choc' in d
|
||||
assert 'dissonances_by_dimension' in d
|
||||
|
||||
def test_to_json(self):
|
||||
"""DissonanceResult.to_json() produit du JSON valide."""
|
||||
import json
|
||||
|
||||
X_t = create_random_tensor()
|
||||
e_input = np.random.randn(EMBEDDING_DIM)
|
||||
e_input = e_input / np.linalg.norm(e_input)
|
||||
|
||||
result = compute_dissonance(e_input, X_t)
|
||||
json_str = result.to_json()
|
||||
|
||||
# Doit être parseable
|
||||
parsed = json.loads(json_str)
|
||||
assert parsed['total'] == result.total
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
pytest.main([__file__, "-v"])
|
||||
Reference in New Issue
Block a user