Spaces:
Running
Running
File size: 21,087 Bytes
6d11371 4dbfec8 6d11371 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 |
#!/usr/bin/env python3
"""
Performance Evaluation Script for AskVeracity.
This script evaluates the performance of the AskVeracity fact-checking system
using a predefined set of test claims with known ground truth labels.
It collects metrics on accuracy, safety rate, processing time, and confidence scores
without modifying the core codebase.
Usage:
python evaluate_performance.py [--limit N] [--output FILE]
Options:
--limit N Limit evaluation to first N claims (default: all)
--output FILE Save results to FILE (default: performance_results.json)
"""
import os
import sys
import json
import time
import argparse
from datetime import datetime
import matplotlib.pyplot as plt
from tabulate import tabulate
import numpy as np
# Add the parent directory to sys.path if this script is run directly
if __name__ == "__main__":
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
# Import the agent and performance tracker
import agent
from utils.performance import PerformanceTracker
from utils.models import initialize_models
# IMPORTANT NOTE FOR DEVELOPERS:
# The test claims below include many recent events that will become outdated.
# When using this script for testing or evaluation, please update these claims
# with relevant and up-to-date examples to ensure meaningful results.
# Performance metrics are heavily influenced by the recency and verifiability
# of these claims, so using outdated claims will likely lead to poor results.
# Define the test claims with ground truth labels
TEST_CLAIMS = [
# True claims
{"claim": "Dozens killed as gunmen massacre tourists in Kashmir beauty spot.", "expected": "True"},
{"claim": "Pope Francis dies at 88.", "expected": "True"},
{"claim": "OpenAI released new reasoning models called o3 and o4-mini.", "expected": "True"},
{"claim": "Trump And Zelensky Clash Again As US Says Crimea Now Russian Territory.", "expected": "True"},
{"claim": "Twelve states sue Donald Trump administration in trade court over chaotic and illegal tariff policy.", "expected": "True"},
{"claim": "Zomato has been renamed to Eternal Limited.", "expected": "True"},
{"claim": "The Taj Mahal is located in Agra.", "expected": "True"},
{"claim": "ISRO achieves second docking with SpaDeX satellites.", "expected": "True"},
{"claim": "The TV series Adolescence is streaming on Netflix.", "expected": "True"},
{"claim": "Vladimir Putin offers to halt Ukraine invasion.", "expected": "True"},
{"claim": "Meta released its Llama 4 language model.", "expected": "True"},
{"claim": "Google launched Gemini 2.5 Pro Experimental, the first model in the Gemini 2.5 family.", "expected": "True"},
{"claim": "Microsoft is rolling out improved Recall feature for Windows Insiders.", "expected": "True"},
{"claim": "Microsoft announced a 1-bit language model that can run on CPU.", "expected": "True"},
{"claim": "Royal Challengers Bengaluru beat Rajasthan Royals by 11 runs in yesterday's IPL match.", "expected": "True"},
{"claim": "Anthropic introduced Claude Research.", "expected": "True"},
{"claim": "The IMF has lowered India's growth projection for the fiscal year 2025-26 to 6.2 per cent.", "expected": "True"},
{"claim": "In Bundesliga, Bayern Munich beat Heidenheim 4-0 last week.", "expected": "True"},
{"claim": "Manchester United in Europa League semi-finals.", "expected": "True"},
# False claims
{"claim": "The Eiffel Tower is in Rome.", "expected": "False"},
{"claim": "The earth is flat.", "expected": "False"},
{"claim": "Rishi Sunak is the current Prime Minister of the UK.", "expected": "False"},
{"claim": "New Zealand won the ICC Champions Trophy in 2025.", "expected": "False"},
{"claim": "US President Donald trump to visit India next week.", "expected": "False"},
{"claim": "Quantum computers have definitively solved the protein folding problem.", "expected": "False"},
{"claim": "CRISPR gene editing has successfully cured type 1 diabetes in human clinical trials.", "expected": "False"},
{"claim": "Google's new quantum computer, Willow, has demonstrated remarkable capabilities by solving mathematical problems far beyond the reach of the fastest supercomputers.", "expected": "False"},
{"claim": "NASA confirmed that the James Webb Space Telescope has found definitive evidence of alien life on an exoplanet.", "expected": "False"},
{"claim": "Google launched Gemini 3.", "expected": "False"},
{"claim": "A solar eclipse was be seen in India on October 17, 2024.", "expected": "False"},
{"claim": "Tom Cruise and Shah Rukh Khan have starred in a Bollywood movie in the past.", "expected": "False"},
{"claim": "Germany has the highest GDP in the world.", "expected": "False"},
# Uncertain claims
{"claim": "Aliens have visited the Earth.", "expected": "Uncertain"},
{"claim": "Information that falls into a black hole is permanently lost or destroyed.", "expected": "Uncertain"},
{"claim": "Time travel into the past is possible.", "expected": "Uncertain"},
{"claim": "Bigfoot (or Yeti) exists in remote wilderness areas.", "expected": "Uncertain"},
{"claim": "Intelligent life exists elsewhere in the universe.", "expected": "Uncertain"},
{"claim": "Yogi Adityanath will be the next Prime Minister of India.", "expected": "Uncertain"},
{"claim": "Consciousness continues to exist after biological death.", "expected": "Uncertain"},
{"claim": "There are multiple parallel universes.", "expected": "Uncertain"}
]
def setup_argument_parser():
"""
Set up command line argument parsing.
Returns:
argparse.Namespace: Parsed command line arguments
"""
parser = argparse.ArgumentParser(description="Evaluate AskVeracity performance")
parser.add_argument("--limit", type=int, help="Limit evaluation to first N claims")
parser.add_argument("--output", type=str, default="performance_results.json",
help="Output file for results (default: performance_results.json)")
return parser.parse_args()
def initialize_system():
"""
Initialize the system for evaluation.
Returns:
object: Initialized LangGraph agent
"""
print("Initializing models and agent...")
initialize_models()
eval_agent = agent.setup_agent()
return eval_agent
def normalize_classification(classification):
"""
Normalize classification labels for consistent comparison.
Args:
classification (str): Classification label from the system
Returns:
str: Normalized classification label ("True", "False", or "Uncertain")
"""
if not classification:
return "Uncertain"
if "true" in classification.lower():
return "True"
elif "false" in classification.lower():
return "False"
else:
return "Uncertain"
def is_correct(actual, expected):
"""
Determine if the actual classification matches the expected classification.
Args:
actual (str): Actual classification from the system
expected (str): Expected (ground truth) classification
Returns:
bool: True if classifications match, False otherwise
"""
# Normalize both for comparison
normalized_actual = normalize_classification(actual)
normalized_expected = expected
return normalized_actual == normalized_expected
def is_safe(actual, expected):
"""
Determine if the classification is "safe" - either correct or abstained (Uncertain)
instead of making an incorrect assertion.
Args:
actual (str): Actual classification from the system
expected (str): Expected (ground truth) classification
Returns:
bool: True if the classification is safe, False otherwise
"""
# Normalize both for comparison
normalized_actual = normalize_classification(actual)
normalized_expected = expected
# If the classification is correct, it's definitely safe
if normalized_actual == normalized_expected:
return True
# If the system classified as "Uncertain", that's safe (abstaining rather than wrong assertion)
if normalized_actual == "Uncertain":
return True
# Otherwise, the system made an incorrect assertion (False as True or True as False)
return False
def evaluate_claims(test_claims, eval_agent, limit=None):
"""
Evaluate a list of claims using the fact-checking system.
Args:
test_claims (list): List of test claims with expected classifications
eval_agent (object): Initialized LangGraph agent
limit (int, optional): Maximum number of claims to evaluate
Returns:
tuple: (results, metrics)
- results (list): Detailed results for each claim
- metrics (dict): Aggregated performance metrics
"""
# Initialize performance tracker
performance_tracker = PerformanceTracker()
# Limit the number of claims if requested
if limit and limit > 0:
claims_to_evaluate = test_claims[:limit]
else:
claims_to_evaluate = test_claims
results = []
total_count = len(claims_to_evaluate)
correct_count = 0
safe_count = 0
# Classification counts
classification_counts = {"True": 0, "False": 0, "Uncertain": 0}
# Track processing times by expected classification
processing_times = {"True": [], "False": [], "Uncertain": []}
# Confidence scores by expected classification
confidence_scores = {"True": [], "False": [], "Uncertain": []}
# Track correct classifications by expected classification
correct_by_class = {"True": 0, "False": 0, "Uncertain": 0}
safe_by_class = {"True": 0, "False": 0, "Uncertain": 0}
total_by_class = {"True": 0, "False": 0, "Uncertain": 0}
print(f"Evaluating {len(claims_to_evaluate)} claims...")
# Process each claim
for idx, test_case in enumerate(claims_to_evaluate):
claim = test_case["claim"]
expected = test_case["expected"]
print(f"\nProcessing claim {idx+1}/{len(claims_to_evaluate)}: {claim}")
try:
# Process the claim and measure time
start_time = time.time()
result = agent.process_claim(claim, eval_agent)
total_time = time.time() - start_time
# Extract classification and confidence
classification = result.get("classification", "Uncertain")
confidence = result.get("confidence", 0.0)
# Normalize classification for comparison
normalized_classification = normalize_classification(classification)
# Check if classification is correct
correct = is_correct(normalized_classification, expected)
if correct:
correct_count += 1
correct_by_class[expected] += 1
# Check if classification is safe
safe = is_safe(normalized_classification, expected)
if safe:
safe_count += 1
safe_by_class[expected] += 1
# Update classification count
classification_counts[normalized_classification] = classification_counts.get(normalized_classification, 0) + 1
# Update counts by expected class
total_by_class[expected] += 1
# Update processing times
processing_times[expected].append(total_time)
# Update confidence scores
confidence_scores[expected].append(confidence)
# Save detailed result
detail_result = {
"claim": claim,
"expected": expected,
"actual": normalized_classification,
"correct": correct,
"safe": safe,
"confidence": confidence,
"processing_time": total_time
}
results.append(detail_result)
# Print progress indicator
outcome = "β" if correct else "β"
safety = "(safe)" if safe and not correct else ""
print(f" Result: {normalized_classification} (Expected: {expected}) {outcome} {safety}")
print(f" Time: {total_time:.2f}s, Confidence: {confidence:.2f}")
except Exception as e:
print(f"Error processing claim: {str(e)}")
results.append({
"claim": claim,
"expected": expected,
"error": str(e)
})
# Calculate performance metrics
accuracy = correct_count / total_count if total_count > 0 else 0
safety_rate = safe_count / total_count if total_count > 0 else 0
# Calculate per-class metrics
class_metrics = {}
for cls in ["True", "False", "Uncertain"]:
class_accuracy = correct_by_class[cls] / total_by_class[cls] if total_by_class[cls] > 0 else 0
class_safety_rate = safe_by_class[cls] / total_by_class[cls] if total_by_class[cls] > 0 else 0
avg_time = sum(processing_times[cls]) / len(processing_times[cls]) if processing_times[cls] else 0
avg_confidence = sum(confidence_scores[cls]) / len(confidence_scores[cls]) if confidence_scores[cls] else 0
class_metrics[cls] = {
"accuracy": class_accuracy,
"safety_rate": class_safety_rate,
"count": total_by_class[cls],
"correct": correct_by_class[cls],
"safe": safe_by_class[cls],
"avg_processing_time": avg_time,
"avg_confidence": avg_confidence
}
# Calculate overall metrics
all_times = [r.get("processing_time", 0) for r in results if "processing_time" in r]
all_confidence = [r.get("confidence", 0) for r in results if "confidence" in r]
metrics = {
"total_claims": total_count,
"correct_claims": correct_count,
"safe_claims": safe_count,
"accuracy": accuracy,
"safety_rate": safety_rate,
"avg_processing_time": sum(all_times) / len(all_times) if all_times else 0,
"avg_confidence": sum(all_confidence) / len(all_confidence) if all_confidence else 0,
"classification_counts": classification_counts,
"per_class_metrics": class_metrics
}
return results, metrics
def save_results(results, metrics, output_file):
"""
Save evaluation results to a JSON file.
Args:
results (list): Detailed results for each claim
metrics (dict): Aggregated performance metrics
output_file (str): Path to output file
"""
output_data = {
"timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
"metrics": metrics,
"detailed_results": results
}
with open(output_file, 'w') as f:
json.dump(output_data, f, indent=2)
print(f"\nResults saved to {output_file}")
def print_summary(metrics):
"""
Print a summary of performance metrics.
Args:
metrics (dict): Aggregated performance metrics
"""
print("\n" + "="*70)
print(f"PERFORMANCE SUMMARY")
print("="*70)
# Overall metrics
print(f"\nOverall Metrics:")
print(f"Total Claims: {metrics['total_claims']}")
print(f"Correctly Classified: {metrics['correct_claims']}")
print(f"Safely Classified: {metrics['safe_claims']}")
print(f"Accuracy: {metrics['accuracy']:.2%}")
print(f"Safety Rate: {metrics['safety_rate']:.2%}")
print(f"Average Processing Time: {metrics['avg_processing_time']:.2f} seconds")
print(f"Average Confidence Score: {metrics['avg_confidence']:.2f}")
# Per-class metrics as table
print("\nPer-Class Performance:")
table_data = []
headers = ["Class", "Count", "Correct", "Safe", "Accuracy", "Safety Rate", "Avg Time", "Avg Confidence"]
for cls, cls_metrics in metrics['per_class_metrics'].items():
table_data.append([
cls,
cls_metrics['count'],
cls_metrics['correct'],
cls_metrics['safe'],
f"{cls_metrics['accuracy']:.2%}",
f"{cls_metrics['safety_rate']:.2%}",
f"{cls_metrics['avg_processing_time']:.2f}s",
f"{cls_metrics['avg_confidence']:.2f}"
])
print(tabulate(table_data, headers=headers, tablefmt="grid"))
def create_charts(metrics, output_dir="."):
"""
Create visualizations of performance metrics.
Args:
metrics (dict): Aggregated performance metrics
output_dir (str): Directory to save charts
"""
try:
# Create output directory if it doesn't exist
os.makedirs(output_dir, exist_ok=True)
# Plot 1: Accuracy by class
plt.figure(figsize=(10, 6))
classes = list(metrics['per_class_metrics'].keys())
accuracies = [metrics['per_class_metrics'][cls]['accuracy'] for cls in classes]
plt.bar(classes, accuracies, color=['green', 'red', 'gray'])
plt.title('Accuracy by Classification Type')
plt.xlabel('Classification')
plt.ylabel('Accuracy')
plt.ylim(0, 1)
for i, v in enumerate(accuracies):
plt.text(i, v + 0.02, f"{v:.2%}", ha='center')
plt.tight_layout()
plt.savefig(os.path.join(output_dir, 'accuracy_by_class.png'))
plt.close() # Close the figure to free memory
# Plot 2: Safety rate by class
plt.figure(figsize=(10, 6))
safety_rates = [metrics['per_class_metrics'][cls]['safety_rate'] for cls in classes]
plt.bar(classes, safety_rates, color=['green', 'red', 'gray'])
plt.title('Safety Rate by Classification Type')
plt.xlabel('Classification')
plt.ylabel('Safety Rate')
plt.ylim(0, 1)
for i, v in enumerate(safety_rates):
plt.text(i, v + 0.02, f"{v:.2%}", ha='center')
plt.tight_layout()
plt.savefig(os.path.join(output_dir, 'safety_rate_by_class.png'))
plt.close() # Close the figure to free memory
# Plot 3: Processing time by class
plt.figure(figsize=(10, 6))
times = [metrics['per_class_metrics'][cls]['avg_processing_time'] for cls in classes]
plt.bar(classes, times, color=['green', 'red', 'gray'])
plt.title('Average Processing Time by Classification Type')
plt.xlabel('Classification')
plt.ylabel('Time (seconds)')
for i, v in enumerate(times):
plt.text(i, v + 0.5, f"{v:.2f}s", ha='center')
plt.tight_layout()
plt.savefig(os.path.join(output_dir, 'processing_time_by_class.png'))
plt.close() # Close the figure to free memory
# Plot 4: Confidence scores by class
plt.figure(figsize=(10, 6))
confidence = [metrics['per_class_metrics'][cls]['avg_confidence'] for cls in classes]
plt.bar(classes, confidence, color=['green', 'red', 'gray'])
plt.title('Average Confidence Score by Classification Type')
plt.xlabel('Classification')
plt.ylabel('Confidence Score')
plt.ylim(0, 1)
for i, v in enumerate(confidence):
plt.text(i, v + 0.02, f"{v:.2f}", ha='center')
plt.tight_layout()
plt.savefig(os.path.join(output_dir, 'confidence_by_class.png'))
plt.close() # Close the figure to free memory
print(f"\nCharts created in {output_dir}")
except Exception as e:
print(f"Error creating charts: {str(e)}")
print("Continuing without charts.")
def main():
"""Main evaluation function that runs the entire evaluation process."""
# Parse arguments
args = setup_argument_parser()
# Initialize the agent
eval_agent = initialize_system()
# Create results directory if it doesn't exist
results_dir = "results"
os.makedirs(results_dir, exist_ok=True)
# Set output file path
output_file = args.output
if not os.path.isabs(output_file):
output_file = os.path.join(results_dir, output_file)
# Evaluate claims
results, metrics = evaluate_claims(TEST_CLAIMS, eval_agent, args.limit)
# results, metrics = evaluate_claims(TEST_CLAIMS, eval_agent, 1)
# Print summary
print_summary(metrics)
# Save results
save_results(results, metrics, output_file)
# Create charts
try:
from tabulate import tabulate
import matplotlib.pyplot as plt
create_charts(metrics, results_dir)
except ImportError:
print("\nCould not create charts. Please install matplotlib and tabulate packages:")
print("pip install matplotlib tabulate")
if __name__ == "__main__":
main() |