Spaces:
Sleeping
Sleeping
File size: 18,878 Bytes
8205d6b |
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 |
# DEPENDENCIES
import re
import math
import torch
import numpy as np
from typing import Any
from typing import Dict
from typing import List
from loguru import logger
from config.threshold_config import Domain
from metrics.base_metric import BaseMetric
from metrics.base_metric import MetricResult
from models.model_manager import get_model_manager
from config.threshold_config import get_threshold_for_domain
class PerplexityMetric(BaseMetric):
"""
Text predictability analysis using GPT-2 for perplexity calculation
Measures (Aligned with Documentation):
- Overall text perplexity (lower = more predictable = more AI-like)
- Perplexity distribution across text chunks
- Sentence-level perplexity patterns
- Cross-entropy analysis
"""
def __init__(self):
super().__init__(name = "perplexity",
description = "GPT-2 based perplexity calculation for text predictability analysis",
)
self.model = None
self.tokenizer = None
def initialize(self) -> bool:
"""
Initialize the perplexity metric
"""
try:
logger.info("Initializing perplexity metric...")
# Load GPT-2 model and tokenizer
model_manager = get_model_manager()
model_result = model_manager.load_model(model_name = "perplexity_gpt2")
if isinstance(model_result, tuple):
self.model, self.tokenizer = model_result
else:
logger.error("Failed to load GPT-2 model for perplexity calculation")
return False
self.is_initialized = True
logger.success("Perplexity metric initialized successfully")
return True
except Exception as e:
logger.error(f"Failed to initialize perplexity metric: {repr(e)}")
return False
def compute(self, text: str, **kwargs) -> MetricResult:
"""
Compute perplexity measures with FULL DOMAIN THRESHOLD INTEGRATION
"""
try:
if not text or len(text.strip()) < 50:
return MetricResult(metric_name = self.name,
ai_probability = 0.5,
human_probability = 0.5,
mixed_probability = 0.0,
confidence = 0.1,
error = "Text too short for perplexity analysis",
)
# Get domain-specific thresholds
domain = kwargs.get('domain', Domain.GENERAL)
domain_thresholds = get_threshold_for_domain(domain)
perplexity_thresholds = domain_thresholds.perplexity
# Calculate comprehensive perplexity features
features = self._calculate_perplexity_features(text)
# Calculate raw perplexity score (0-1 scale)
raw_perplexity_score, confidence = self._analyze_perplexity_patterns(features)
# Apply domain-specific thresholds to convert raw score to probabilities
ai_prob, human_prob, mixed_prob = self._apply_domain_thresholds(raw_perplexity_score, perplexity_thresholds, features)
# Apply confidence multiplier from domain thresholds
confidence *= perplexity_thresholds.confidence_multiplier
confidence = max(0.0, min(1.0, confidence))
return MetricResult(metric_name = self.name,
ai_probability = ai_prob,
human_probability = human_prob,
mixed_probability = mixed_prob,
confidence = confidence,
details = {**features,
'domain_used' : domain.value,
'ai_threshold' : perplexity_thresholds.ai_threshold,
'human_threshold' : perplexity_thresholds.human_threshold,
'raw_score' : raw_perplexity_score,
},
)
except Exception as e:
logger.error(f"Error in perplexity computation: {repr(e)}")
return MetricResult(metric_name = self.name,
ai_probability = 0.5,
human_probability = 0.5,
mixed_probability = 0.0,
confidence = 0.0,
error = str(e),
)
def _apply_domain_thresholds(self, raw_score: float, thresholds: Any, features: Dict[str, Any]) -> tuple:
"""
Apply domain-specific thresholds to convert raw score to probabilities
"""
ai_threshold = thresholds.ai_threshold # e.g., 0.60 for GENERAL, 0.55 for ACADEMIC
human_threshold = thresholds.human_threshold # e.g., 0.40 for GENERAL, 0.35 for ACADEMIC
# Calculate probabilities based on threshold distances
if (raw_score >= ai_threshold):
# Above AI threshold - strongly AI
distance_from_threshold = raw_score - ai_threshold
ai_prob = 0.7 + (distance_from_threshold * 0.3) # 0.7 to 1.0
human_prob = 0.3 - (distance_from_threshold * 0.3) # 0.3 to 0.0
elif (raw_score <= human_threshold):
# Below human threshold - strongly human
distance_from_threshold = human_threshold - raw_score
ai_prob = 0.3 - (distance_from_threshold * 0.3) # 0.3 to 0.0
human_prob = 0.7 + (distance_from_threshold * 0.3) # 0.7 to 1.0
else:
# Between thresholds - uncertain zone
range_width = ai_threshold - human_threshold
if (range_width > 0):
position_in_range = (raw_score - human_threshold) / range_width
ai_prob = 0.3 + (position_in_range * 0.4) # 0.3 to 0.7
human_prob = 0.7 - (position_in_range * 0.4) # 0.7 to 0.3
else:
ai_prob = 0.5
human_prob = 0.5
# Ensure probabilities are valid
ai_prob = max(0.0, min(1.0, ai_prob))
human_prob = max(0.0, min(1.0, human_prob))
# Calculate mixed probability based on perplexity variance
mixed_prob = self._calculate_mixed_probability(features)
# Normalize to sum to 1.0
total = ai_prob + human_prob + mixed_prob
if (total > 0):
ai_prob /= total
human_prob /= total
mixed_prob /= total
return ai_prob, human_prob, mixed_prob
def _calculate_perplexity_features(self, text: str) -> Dict[str, Any]:
"""
Calculate comprehensive perplexity measures
"""
if not self.model or not self.tokenizer:
return self._get_default_features()
# Calculate overall perplexity
overall_perplexity = self._calculate_perplexity(text)
# Split into sentences for sentence-level analysis
sentences = self._split_sentences(text)
# Calculate sentence-level perplexities
sentence_perplexities = list()
valid_sentences = 0
for sentence in sentences:
# Minimum sentence length
if (len(sentence.strip()) > 20):
sent_perplexity = self._calculate_perplexity(sentence)
if (sent_perplexity > 0):
sentence_perplexities.append(sent_perplexity)
valid_sentences += 1
# Calculate statistical features
if sentence_perplexities:
avg_sentence_perplexity = np.mean(sentence_perplexities)
std_sentence_perplexity = np.std(sentence_perplexities)
min_sentence_perplexity = np.min(sentence_perplexities)
max_sentence_perplexity = np.max(sentence_perplexities)
else:
avg_sentence_perplexity = overall_perplexity
std_sentence_perplexity = 0.0
min_sentence_perplexity = overall_perplexity
max_sentence_perplexity = overall_perplexity
# Chunk-based analysis for whole-text understanding
chunk_perplexities = self._calculate_chunk_perplexity(text, chunk_size = 200)
perplexity_variance = np.var(chunk_perplexities) if chunk_perplexities else 0.0
avg_chunk_perplexity = np.mean(chunk_perplexities) if chunk_perplexities else overall_perplexity
# Normalize perplexity to 0-1 scale for easier interpretation
normalized_perplexity = self._normalize_perplexity(overall_perplexity)
# Cross-entropy analysis
cross_entropy_score = self._calculate_cross_entropy(text)
return {"overall_perplexity" : round(overall_perplexity, 2),
"normalized_perplexity" : round(normalized_perplexity, 4),
"avg_sentence_perplexity" : round(avg_sentence_perplexity, 2),
"std_sentence_perplexity" : round(std_sentence_perplexity, 2),
"min_sentence_perplexity" : round(min_sentence_perplexity, 2),
"max_sentence_perplexity" : round(max_sentence_perplexity, 2),
"perplexity_variance" : round(perplexity_variance, 4),
"avg_chunk_perplexity" : round(avg_chunk_perplexity, 2),
"cross_entropy_score" : round(cross_entropy_score, 4),
"num_sentences_analyzed" : valid_sentences,
"num_chunks_analyzed" : len(chunk_perplexities),
}
def _calculate_perplexity(self, text: str) -> float:
"""
Calculate perplexity for given text using GPT-2 : Lower perplexity = more predictable = more AI-like
"""
try:
# Check text length before tokenization
if (len(text.strip()) < 10):
return 0.0
# Tokenize the text
encodings = self.tokenizer(text,
return_tensors = 'pt',
truncation = True,
max_length = 1024,
)
input_ids = encodings.input_ids
# Minimum tokens
if ((input_ids.numel() == 0) or (input_ids.size(1) < 5)):
return 0.0
# Calculate loss (cross-entropy)
with torch.no_grad():
outputs = self.model(input_ids, labels = input_ids)
loss = outputs.loss
# Convert loss to perplexity
perplexity = math.exp(loss.item())
return perplexity
except Exception as e:
logger.warning(f"Perplexity calculation failed: {repr(e)}")
return 0.0
def _split_sentences(self, text: str) -> List[str]:
"""
Split text into sentences
"""
sentences = re.split(r'[.!?]+', text)
return [s.strip() for s in sentences if s.strip() and len(s.strip()) > 10]
def _calculate_chunk_perplexity(self, text: str, chunk_size: int = 200) -> List[float]:
"""
Calculate perplexity across text chunks for whole-text analysis
"""
chunks = list()
words = text.split()
# Ensure we have enough words for meaningful chunks
if (len(words) < chunk_size // 2):
return [self._calculate_perplexity(text)] if text.strip() else []
# Create overlapping chunks for better analysis
for i in range(0, len(words), chunk_size // 2):
chunk = ' '.join(words[i:i + chunk_size])
# Minimum chunk size
if (len(chunk) > 50):
perplexity = self._calculate_perplexity(chunk)
# Reasonable range check
if ((perplexity > 0) and (perplexity < 1000)):
chunks.append(perplexity)
return chunks if chunks else [0.0]
def _normalize_perplexity(self, perplexity: float) -> float:
"""
Normalize perplexity using sigmoid transformation
Lower perplexity = higher normalized score = more AI-like
"""
# Use exponential normalization : Typical ranges: AI = 10-40, Human = 20-100
normalized = 1.0 / (1.0 + np.exp((perplexity - 30) / 10))
return normalized
def _calculate_cross_entropy(self, text: str) -> float:
"""
Calculate cross-entropy as an alternative measure
"""
try:
encodings = self.tokenizer(text, return_tensors='pt', truncation=True, max_length=1024)
input_ids = encodings.input_ids
if (input_ids.numel() == 0):
return 0.0
with torch.no_grad():
outputs = self.model(input_ids, labels = input_ids)
loss = outputs.loss
# Normalize cross-entropy to 0-1 scale : Assuming max ~5 nats
cross_entropy = loss.item()
normalized_ce = min(1.0, cross_entropy / 5.0)
return normalized_ce
except Exception as e:
logger.warning(f"Cross-entropy calculation failed: {repr(e)}")
return 0.0
def _analyze_perplexity_patterns(self, features: Dict[str, Any]) -> tuple:
"""
Analyze perplexity patterns to determine RAW perplexity score (0-1 scale) : Higher score = more AI-like
"""
# Check feature validity first
required_features = ['normalized_perplexity', 'perplexity_variance', 'std_sentence_perplexity', 'cross_entropy_score']
valid_features = [features.get(feat, 0) for feat in required_features if features.get(feat, 0) > 0]
if (len(valid_features) < 3):
# Low confidence if insufficient features
return 0.5, 0.3
# Initialize ai_indicator list
ai_indicators = list()
# Low overall perplexity suggests AI
if (features['normalized_perplexity'] > 0.7):
# Very AI-like
ai_indicators.append(0.8)
elif (features['normalized_perplexity'] > 0.5):
# AI-like
ai_indicators.append(0.6)
else:
# Human-like
ai_indicators.append(0.2)
# Low perplexity variance suggests AI (consistent predictability)
if (features['perplexity_variance'] < 50):
ai_indicators.append(0.7)
elif (features['perplexity_variance'] < 200):
ai_indicators.append(0.4)
else:
ai_indicators.append(0.2)
# Low sentence perplexity std suggests AI (consistent across sentences)
if (features['std_sentence_perplexity'] < 20):
ai_indicators.append(0.8)
elif (features['std_sentence_perplexity'] < 50):
ai_indicators.append(0.5)
else:
ai_indicators.append(0.2)
# Low cross-entropy suggests AI (more predictable)
if (features['cross_entropy_score'] < 0.3):
ai_indicators.append(0.7)
elif (features['cross_entropy_score'] < 0.6):
ai_indicators.append(0.4)
else:
ai_indicators.append(0.2)
# Consistent chunk perplexity suggests AI
chunk_variance = features['perplexity_variance']
if (chunk_variance < 25):
ai_indicators.append(0.9)
elif (chunk_variance < 100):
ai_indicators.append(0.6)
else:
ai_indicators.append(0.3)
# Calculate raw score and confidence
raw_score = np.mean(ai_indicators) if ai_indicators else 0.5
confidence = 1.0 - (np.std(ai_indicators) / 0.5) if ai_indicators else 0.5
confidence = max(0.1, min(0.9, confidence))
return raw_score, confidence
def _calculate_mixed_probability(self, features: Dict[str, Any]) -> float:
"""
Calculate probability of mixed AI/Human content
"""
mixed_indicators = list()
# Moderate perplexity values might indicate mixing
if (0.4 <= features['normalized_perplexity'] <= 0.6):
mixed_indicators.append(0.3)
else:
mixed_indicators.append(0.0)
# High perplexity variance suggests mixed content
if (features['perplexity_variance'] > 200):
mixed_indicators.append(0.4)
elif (features['perplexity_variance'] > 100):
mixed_indicators.append(0.2)
else:
mixed_indicators.append(0.0)
# Inconsistent sentence perplexities
if (20 <= features['std_sentence_perplexity'] <= 60):
mixed_indicators.append(0.3)
else:
mixed_indicators.append(0.0)
return min(0.3, np.mean(mixed_indicators)) if mixed_indicators else 0.0
def _get_default_features(self) -> Dict[str, Any]:
"""
Return default features when analysis is not possible
"""
return {"overall_perplexity" : 50.0,
"normalized_perplexity" : 0.5,
"avg_sentence_perplexity" : 50.0,
"std_sentence_perplexity" : 25.0,
"min_sentence_perplexity" : 30.0,
"max_sentence_perplexity" : 70.0,
"perplexity_variance" : 100.0,
"avg_chunk_perplexity" : 50.0,
"cross_entropy_score" : 0.5,
"num_sentences_analyzed" : 0,
"num_chunks_analyzed" : 0,
}
def cleanup(self):
"""
Clean up resources
"""
self.model = None
self.tokenizer = None
super().cleanup()
# Export
__all__ = ["PerplexityMetric"]
|