Spaces:
Runtime error
Runtime error
add running pii eval script
Browse files
.gitignore
CHANGED
|
@@ -168,4 +168,5 @@ temp.txt
|
|
| 168 |
**.csv
|
| 169 |
binary-classifier/
|
| 170 |
wandb/
|
| 171 |
-
artifacts/
|
|
|
|
|
|
| 168 |
**.csv
|
| 169 |
binary-classifier/
|
| 170 |
wandb/
|
| 171 |
+
artifacts/
|
| 172 |
+
evaluation_results/
|
guardrails_genie/guardrails/entity_recognition/pii_examples/pii_benchmark.py
CHANGED
|
@@ -0,0 +1,215 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from datasets import load_dataset
|
| 2 |
+
from typing import Dict, List, Tuple
|
| 3 |
+
import random
|
| 4 |
+
from tqdm import tqdm
|
| 5 |
+
import json
|
| 6 |
+
from pathlib import Path
|
| 7 |
+
import weave
|
| 8 |
+
|
| 9 |
+
def load_ai4privacy_dataset(num_samples: int = 100, split: str = "validation") -> List[Dict]:
|
| 10 |
+
"""
|
| 11 |
+
Load and prepare samples from the ai4privacy dataset.
|
| 12 |
+
|
| 13 |
+
Args:
|
| 14 |
+
num_samples: Number of samples to evaluate
|
| 15 |
+
split: Dataset split to use ("train" or "validation")
|
| 16 |
+
|
| 17 |
+
Returns:
|
| 18 |
+
List of prepared test cases
|
| 19 |
+
"""
|
| 20 |
+
# Load the dataset
|
| 21 |
+
dataset = load_dataset("ai4privacy/pii-masking-400k")
|
| 22 |
+
|
| 23 |
+
# Get the specified split
|
| 24 |
+
data_split = dataset[split]
|
| 25 |
+
|
| 26 |
+
# Randomly sample entries if num_samples is less than total
|
| 27 |
+
if num_samples < len(data_split):
|
| 28 |
+
indices = random.sample(range(len(data_split)), num_samples)
|
| 29 |
+
samples = [data_split[i] for i in indices]
|
| 30 |
+
else:
|
| 31 |
+
samples = data_split
|
| 32 |
+
|
| 33 |
+
# Convert to test case format
|
| 34 |
+
test_cases = []
|
| 35 |
+
for sample in samples:
|
| 36 |
+
# Extract entities from privacy_mask
|
| 37 |
+
entities: Dict[str, List[str]] = {}
|
| 38 |
+
for entity in sample['privacy_mask']:
|
| 39 |
+
label = entity['label']
|
| 40 |
+
value = entity['value']
|
| 41 |
+
if label not in entities:
|
| 42 |
+
entities[label] = []
|
| 43 |
+
entities[label].append(value)
|
| 44 |
+
|
| 45 |
+
test_case = {
|
| 46 |
+
"description": f"AI4Privacy Sample (ID: {sample['uid']})",
|
| 47 |
+
"input_text": sample['source_text'],
|
| 48 |
+
"expected_entities": entities,
|
| 49 |
+
"masked_text": sample['masked_text'],
|
| 50 |
+
"language": sample['language'],
|
| 51 |
+
"locale": sample['locale']
|
| 52 |
+
}
|
| 53 |
+
test_cases.append(test_case)
|
| 54 |
+
|
| 55 |
+
return test_cases
|
| 56 |
+
|
| 57 |
+
@weave.op()
|
| 58 |
+
def evaluate_model(guardrail, test_cases: List[Dict]) -> Tuple[Dict, List[Dict]]:
|
| 59 |
+
"""
|
| 60 |
+
Evaluate a model on the test cases.
|
| 61 |
+
|
| 62 |
+
Args:
|
| 63 |
+
guardrail: Entity recognition guardrail to evaluate
|
| 64 |
+
test_cases: List of test cases
|
| 65 |
+
|
| 66 |
+
Returns:
|
| 67 |
+
Tuple of (metrics dict, detailed results list)
|
| 68 |
+
"""
|
| 69 |
+
metrics = {
|
| 70 |
+
"total": len(test_cases),
|
| 71 |
+
"passed": 0,
|
| 72 |
+
"failed": 0,
|
| 73 |
+
"entity_metrics": {} # Will store precision/recall per entity type
|
| 74 |
+
}
|
| 75 |
+
|
| 76 |
+
detailed_results = []
|
| 77 |
+
|
| 78 |
+
for test_case in tqdm(test_cases, desc="Evaluating samples"):
|
| 79 |
+
# Run detection
|
| 80 |
+
result = guardrail.guard(test_case['input_text'])
|
| 81 |
+
detected = result.detected_entities
|
| 82 |
+
expected = test_case['expected_entities']
|
| 83 |
+
|
| 84 |
+
# Track entity-level metrics
|
| 85 |
+
all_entity_types = set(list(detected.keys()) + list(expected.keys()))
|
| 86 |
+
entity_results = {}
|
| 87 |
+
|
| 88 |
+
for entity_type in all_entity_types:
|
| 89 |
+
detected_set = set(detected.get(entity_type, []))
|
| 90 |
+
expected_set = set(expected.get(entity_type, []))
|
| 91 |
+
|
| 92 |
+
# Calculate metrics
|
| 93 |
+
true_positives = len(detected_set & expected_set)
|
| 94 |
+
false_positives = len(detected_set - expected_set)
|
| 95 |
+
false_negatives = len(expected_set - detected_set)
|
| 96 |
+
|
| 97 |
+
precision = true_positives / (true_positives + false_positives) if (true_positives + false_positives) > 0 else 0
|
| 98 |
+
recall = true_positives / (true_positives + false_negatives) if (true_positives + false_negatives) > 0 else 0
|
| 99 |
+
f1 = 2 * (precision * recall) / (precision + recall) if (precision + recall) > 0 else 0
|
| 100 |
+
|
| 101 |
+
entity_results[entity_type] = {
|
| 102 |
+
"precision": precision,
|
| 103 |
+
"recall": recall,
|
| 104 |
+
"f1": f1,
|
| 105 |
+
"true_positives": true_positives,
|
| 106 |
+
"false_positives": false_positives,
|
| 107 |
+
"false_negatives": false_negatives
|
| 108 |
+
}
|
| 109 |
+
|
| 110 |
+
# Aggregate metrics
|
| 111 |
+
if entity_type not in metrics["entity_metrics"]:
|
| 112 |
+
metrics["entity_metrics"][entity_type] = {
|
| 113 |
+
"total_true_positives": 0,
|
| 114 |
+
"total_false_positives": 0,
|
| 115 |
+
"total_false_negatives": 0
|
| 116 |
+
}
|
| 117 |
+
metrics["entity_metrics"][entity_type]["total_true_positives"] += true_positives
|
| 118 |
+
metrics["entity_metrics"][entity_type]["total_false_positives"] += false_positives
|
| 119 |
+
metrics["entity_metrics"][entity_type]["total_false_negatives"] += false_negatives
|
| 120 |
+
|
| 121 |
+
# Store detailed result
|
| 122 |
+
detailed_result = {
|
| 123 |
+
"id": test_case.get("description", ""),
|
| 124 |
+
"language": test_case.get("language", ""),
|
| 125 |
+
"locale": test_case.get("locale", ""),
|
| 126 |
+
"input_text": test_case["input_text"],
|
| 127 |
+
"expected_entities": expected,
|
| 128 |
+
"detected_entities": detected,
|
| 129 |
+
"entity_metrics": entity_results,
|
| 130 |
+
"anonymized_text": result.anonymized_text if result.anonymized_text else None
|
| 131 |
+
}
|
| 132 |
+
detailed_results.append(detailed_result)
|
| 133 |
+
|
| 134 |
+
# Update pass/fail counts
|
| 135 |
+
if all(entity_results[et]["f1"] == 1.0 for et in entity_results):
|
| 136 |
+
metrics["passed"] += 1
|
| 137 |
+
else:
|
| 138 |
+
metrics["failed"] += 1
|
| 139 |
+
|
| 140 |
+
# Calculate final entity metrics
|
| 141 |
+
for entity_type, counts in metrics["entity_metrics"].items():
|
| 142 |
+
tp = counts["total_true_positives"]
|
| 143 |
+
fp = counts["total_false_positives"]
|
| 144 |
+
fn = counts["total_false_negatives"]
|
| 145 |
+
|
| 146 |
+
precision = tp / (tp + fp) if (tp + fp) > 0 else 0
|
| 147 |
+
recall = tp / (tp + fn) if (tp + fn) > 0 else 0
|
| 148 |
+
f1 = 2 * (precision * recall) / (precision + recall) if (precision + recall) > 0 else 0
|
| 149 |
+
|
| 150 |
+
metrics["entity_metrics"][entity_type].update({
|
| 151 |
+
"precision": precision,
|
| 152 |
+
"recall": recall,
|
| 153 |
+
"f1": f1
|
| 154 |
+
})
|
| 155 |
+
|
| 156 |
+
return metrics, detailed_results
|
| 157 |
+
|
| 158 |
+
def save_results(metrics: Dict, detailed_results: List[Dict], model_name: str, output_dir: str = "evaluation_results"):
|
| 159 |
+
"""Save evaluation results to files"""
|
| 160 |
+
output_dir = Path(output_dir)
|
| 161 |
+
output_dir.mkdir(exist_ok=True)
|
| 162 |
+
|
| 163 |
+
# Save metrics summary
|
| 164 |
+
with open(output_dir / f"{model_name}_metrics.json", "w") as f:
|
| 165 |
+
json.dump(metrics, f, indent=2)
|
| 166 |
+
|
| 167 |
+
# Save detailed results
|
| 168 |
+
with open(output_dir / f"{model_name}_detailed_results.json", "w") as f:
|
| 169 |
+
json.dump(detailed_results, f, indent=2)
|
| 170 |
+
|
| 171 |
+
def print_metrics_summary(metrics: Dict):
|
| 172 |
+
"""Print a summary of the evaluation metrics"""
|
| 173 |
+
print("\nEvaluation Summary")
|
| 174 |
+
print("=" * 80)
|
| 175 |
+
print(f"Total Samples: {metrics['total']}")
|
| 176 |
+
print(f"Passed: {metrics['passed']}")
|
| 177 |
+
print(f"Failed: {metrics['failed']}")
|
| 178 |
+
print(f"Success Rate: {(metrics['passed']/metrics['total'])*100:.1f}%")
|
| 179 |
+
|
| 180 |
+
print("\nEntity-level Metrics:")
|
| 181 |
+
print("-" * 80)
|
| 182 |
+
print(f"{'Entity Type':<20} {'Precision':>10} {'Recall':>10} {'F1':>10}")
|
| 183 |
+
print("-" * 80)
|
| 184 |
+
for entity_type, entity_metrics in metrics["entity_metrics"].items():
|
| 185 |
+
print(f"{entity_type:<20} {entity_metrics['precision']:>10.2f} {entity_metrics['recall']:>10.2f} {entity_metrics['f1']:>10.2f}")
|
| 186 |
+
|
| 187 |
+
def main():
|
| 188 |
+
"""Main evaluation function"""
|
| 189 |
+
weave.init("guardrails-genie-pii-evaluation")
|
| 190 |
+
|
| 191 |
+
# Load test cases
|
| 192 |
+
test_cases = load_ai4privacy_dataset(num_samples=100)
|
| 193 |
+
|
| 194 |
+
# Initialize models to evaluate
|
| 195 |
+
models = {
|
| 196 |
+
"regex": RegexEntityRecognitionGuardrail(should_anonymize=True),
|
| 197 |
+
"presidio": PresidioEntityRecognitionGuardrail(should_anonymize=True),
|
| 198 |
+
"transformers": TransformersEntityRecognitionGuardrail(should_anonymize=True)
|
| 199 |
+
}
|
| 200 |
+
|
| 201 |
+
# Evaluate each model
|
| 202 |
+
for model_name, guardrail in models.items():
|
| 203 |
+
print(f"\nEvaluating {model_name} model...")
|
| 204 |
+
metrics, detailed_results = evaluate_model(guardrail, test_cases)
|
| 205 |
+
|
| 206 |
+
# Print and save results
|
| 207 |
+
print_metrics_summary(metrics)
|
| 208 |
+
save_results(metrics, detailed_results, model_name)
|
| 209 |
+
|
| 210 |
+
if __name__ == "__main__":
|
| 211 |
+
from guardrails_genie.guardrails.entity_recognition.regex_entity_recognition_guardrail import RegexEntityRecognitionGuardrail
|
| 212 |
+
from guardrails_genie.guardrails.entity_recognition.presidio_entity_recognition_guardrail import PresidioEntityRecognitionGuardrail
|
| 213 |
+
from guardrails_genie.guardrails.entity_recognition.transformers_entity_recognition_guardrail import TransformersEntityRecognitionGuardrail
|
| 214 |
+
|
| 215 |
+
main()
|