Domi31tls
340348b820
- Sélection et copie de 27 documents représentatifs (10 simples, 12 moyens, 5 complexes) - Outil d'annotation CLI complet (tools/annotation_tool.py) - Guide d'annotation détaillé (docs/annotation_guide.md) - Évaluateur de qualité (evaluation/quality_evaluator.py) * Calcul Précision, Rappel, F1-Score * Identification faux positifs/négatifs * Métriques par type de PII * Export JSON et rapports texte - Scanner de fuite (evaluation/leak_scanner.py) * Détection PII résiduels (CRITIQUE) * Détection nouveaux PII (HAUTE) * Scan métadonnées PDF (MOYENNE) - Benchmark de performance (evaluation/benchmark.py) * Mesure temps de traitement * Mesure CPU/RAM * Export JSON/CSV - Tests unitaires complets pour tous les composants - Documentation complète du module d'évaluation Tâches complétées: - 1.1.1 Sélection de 27 documents (au lieu de 30) - 1.1.2 Outil d'annotation CLI - 1.2.1 Évaluateur de qualité - 1.2.2 Scanner de fuite - 1.2.3 Benchmark de performance Prochaines étapes: - 1.1.3 Annotation des 27 documents (manuel) - 1.1.4 Enrichissement stopwords médicaux - 1.3 Mesure de la baseline
146 lines
4.8 KiB
Python
146 lines
4.8 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Tests unitaires pour l'évaluateur de qualité.
|
|
"""
|
|
import pytest
|
|
from pathlib import Path
|
|
from evaluation.quality_evaluator import QualityEvaluator, EvaluationResult
|
|
|
|
|
|
class TestQualityEvaluator:
|
|
"""Tests pour QualityEvaluator."""
|
|
|
|
def test_normalize_text(self):
|
|
"""Test de normalisation de texte."""
|
|
evaluator = QualityEvaluator(Path("tests/ground_truth"))
|
|
|
|
assert evaluator.normalize_text("DUPONT") == "dupont"
|
|
assert evaluator.normalize_text(" DUPONT ") == "dupont"
|
|
assert evaluator.normalize_text("DUPONT\n\nMARTIN") == "dupont martin"
|
|
assert evaluator.normalize_text("Jean-Pierre") == "jean-pierre"
|
|
|
|
def test_types_match(self):
|
|
"""Test de correspondance des types."""
|
|
evaluator = QualityEvaluator(Path("tests/ground_truth"))
|
|
|
|
# Correspondance directe
|
|
assert evaluator.types_match("NOM", "NOM")
|
|
assert evaluator.types_match("NOM", "NOM_GLOBAL")
|
|
assert evaluator.types_match("TEL", "TEL_GLOBAL")
|
|
|
|
# Correspondance croisée
|
|
assert evaluator.types_match("NOM", "PRENOM")
|
|
assert evaluator.types_match("PRENOM", "NOM")
|
|
|
|
# Non correspondance
|
|
assert not evaluator.types_match("NOM", "TEL")
|
|
assert not evaluator.types_match("EMAIL", "ADRESSE")
|
|
|
|
def test_calculate_metrics(self):
|
|
"""Test de calcul des métriques."""
|
|
evaluator = QualityEvaluator(Path("tests/ground_truth"))
|
|
|
|
# Cas parfait
|
|
precision, recall, f1 = evaluator.calculate_metrics(10, 0, 0)
|
|
assert precision == 1.0
|
|
assert recall == 1.0
|
|
assert f1 == 1.0
|
|
|
|
# Cas avec erreurs
|
|
precision, recall, f1 = evaluator.calculate_metrics(8, 2, 2)
|
|
assert precision == 0.8 # 8 / (8 + 2)
|
|
assert recall == 0.8 # 8 / (8 + 2)
|
|
assert f1 == 0.8
|
|
|
|
# Cas zéro
|
|
precision, recall, f1 = evaluator.calculate_metrics(0, 0, 0)
|
|
assert precision == 0.0
|
|
assert recall == 0.0
|
|
assert f1 == 0.0
|
|
|
|
def test_compare_simple(self):
|
|
"""Test de comparaison simple."""
|
|
evaluator = QualityEvaluator(Path("tests/ground_truth"))
|
|
|
|
annotations = [
|
|
{"page": 0, "type": "NOM", "text": "DUPONT", "context": "Dr. DUPONT"},
|
|
{"page": 0, "type": "TEL", "text": "01 23 45 67 89", "context": "Tel: 01 23 45 67 89"}
|
|
]
|
|
|
|
detections = [
|
|
{"page": 0, "kind": "NOM", "original": "DUPONT"},
|
|
{"page": 0, "kind": "TEL", "original": "01 23 45 67 89"}
|
|
]
|
|
|
|
tp, fn, fp = evaluator.compare(annotations, detections)
|
|
|
|
assert len(tp) == 2
|
|
assert len(fn) == 0
|
|
assert len(fp) == 0
|
|
|
|
def test_compare_with_false_negative(self):
|
|
"""Test avec faux négatif."""
|
|
evaluator = QualityEvaluator(Path("tests/ground_truth"))
|
|
|
|
annotations = [
|
|
{"page": 0, "type": "NOM", "text": "DUPONT", "context": "Dr. DUPONT"},
|
|
{"page": 0, "type": "TEL", "text": "01 23 45 67 89", "context": "Tel: 01 23 45 67 89"}
|
|
]
|
|
|
|
detections = [
|
|
{"page": 0, "kind": "NOM", "original": "DUPONT"}
|
|
# TEL manquant
|
|
]
|
|
|
|
tp, fn, fp = evaluator.compare(annotations, detections)
|
|
|
|
assert len(tp) == 1
|
|
assert len(fn) == 1
|
|
assert len(fp) == 0
|
|
assert fn[0]["type"] == "TEL"
|
|
assert fn[0]["reason"] == "not_detected"
|
|
|
|
def test_compare_with_false_positive(self):
|
|
"""Test avec faux positif."""
|
|
evaluator = QualityEvaluator(Path("tests/ground_truth"))
|
|
|
|
annotations = [
|
|
{"page": 0, "type": "NOM", "text": "DUPONT", "context": "Dr. DUPONT"}
|
|
]
|
|
|
|
detections = [
|
|
{"page": 0, "kind": "NOM", "original": "DUPONT"},
|
|
{"page": 0, "kind": "NOM", "original": "MARTIN"} # Faux positif
|
|
]
|
|
|
|
tp, fn, fp = evaluator.compare(annotations, detections)
|
|
|
|
assert len(tp) == 1
|
|
assert len(fn) == 0
|
|
assert len(fp) == 1
|
|
assert fp[0]["text"] == "MARTIN"
|
|
|
|
def test_evaluation_result_to_dict(self):
|
|
"""Test de conversion en dictionnaire."""
|
|
result = EvaluationResult(
|
|
pdf_path="test.pdf",
|
|
true_positives=10,
|
|
false_positives=2,
|
|
false_negatives=1,
|
|
precision=0.8333,
|
|
recall=0.9091,
|
|
f1_score=0.8696
|
|
)
|
|
|
|
data = result.to_dict()
|
|
|
|
assert data["pdf_path"] == "test.pdf"
|
|
assert data["true_positives"] == 10
|
|
assert data["precision"] == 0.8333
|
|
assert data["recall"] == 0.9091
|
|
assert data["f1_score"] == 0.8696
|
|
|
|
|
|
if __name__ == "__main__":
|
|
pytest.main([__file__, "-v"])
|