Spaces:
Running
Running
File size: 21,820 Bytes
5e1a30c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 |
"""
Enhanced Neural Reranker for Advanced Retrieval.
This module provides a sophisticated neural reranking sub-component that extends
the existing reranker capabilities with advanced features including multiple
model support, adaptive strategies, score fusion, and performance optimization.
This is the architecture-compliant implementation in the proper rerankers/
sub-component location, enhanced with capabilities from the migrated utilities.
"""
import logging
import time
from typing import List, Dict, Any, Tuple, Optional, Union
import numpy as np
from src.core.interfaces import Document
from .base import Reranker
from .utils import (
ScoreFusion, AdaptiveStrategies, CrossEncoderModels, PerformanceOptimizer,
ModelConfig, WeightsConfig, NormalizationConfig
)
logger = logging.getLogger(__name__)
class NeuralRerankingError(Exception):
"""Raised when neural reranking operations fail."""
pass
class NeuralReranker(Reranker):
"""
Enhanced neural reranker with sophisticated capabilities.
This reranker extends the base Reranker interface with advanced features:
- Multiple cross-encoder model support with lazy loading
- Query-type adaptive reranking strategies
- Advanced neural + retrieval score fusion
- Performance optimization with caching and batching
- Graceful degradation and comprehensive error handling
- Real-time performance monitoring and adaptation
The implementation follows proper architecture patterns by enhancing
the existing rerankers sub-component with utilities from the migrated
reranking/ module capabilities.
Features:
- β
Multiple model support (general + technical domains)
- β
Adaptive model selection based on query type
- β
Advanced score fusion with normalization strategies
- β
Performance optimization (<200ms target)
- β
Intelligent caching with LRU eviction
- β
Batch processing with dynamic sizing
- β
Comprehensive error handling and fallbacks
- β
Real-time metrics collection and adaptation
Example:
config = {
"enabled": True,
"models": {
"default_model": {
"name": "cross-encoder/ms-marco-MiniLM-L6-v2",
"max_length": 512,
"batch_size": 16
},
"technical_model": {
"name": "cross-encoder/ms-marco-electra-base",
"max_length": 512,
"batch_size": 8
}
},
"adaptive": {
"enabled": True,
"confidence_threshold": 0.7
},
"score_fusion": {
"method": "weighted",
"weights": {
"neural_score": 0.7,
"retrieval_score": 0.3
}
},
"performance": {
"max_latency_ms": 200,
"enable_caching": True,
"max_cache_size": 10000
}
}
reranker = NeuralReranker(config)
results = reranker.rerank(query, documents, initial_scores)
"""
def __init__(self, config: Dict[str, Any]):
"""
Initialize enhanced neural reranker.
Args:
config: Neural reranking configuration dictionary
"""
self.config = config
self.enabled = config.get("enabled", True)
# Parse configuration sections
self._parse_configuration(config)
# Initialize advanced components
self.models_manager = None
self.adaptive_strategies = None
self.score_fusion = None
self.performance_optimizer = None
# Performance tracking
self.stats = {
"total_queries": 0,
"successful_queries": 0,
"failed_queries": 0,
"total_latency_ms": 0.0,
"model_switches": 0,
"cache_hits": 0,
"cache_misses": 0,
"fallback_activations": 0,
"adaptive_adjustments": 0
}
# State management
self._initialized = False
self._error_count = 0
self._last_performance_check = time.time()
# Initialize immediately if enabled (remove lazy initialization)
initialize_immediately = config.get("initialize_immediately", True)
if self.enabled and initialize_immediately:
try:
self._initialize_if_needed()
except Exception as e:
logger.warning(f"Failed to initialize neural reranker: {e}")
logger.warning("Disabling neural reranker and falling back to identity mode")
self.enabled = False
self._initialized = True # Mark as initialized even when disabled
else:
self._initialized = True # Mark as initialized when disabled
logger.info(f"Enhanced NeuralReranker initialized with {len(self.models_config)} models, "
f"enabled={self.enabled}, initialized={self._initialized}")
def _parse_configuration(self, config: Dict[str, Any]):
"""Parse and validate configuration sections."""
# Models configuration
self.models_config = config.get("models", {
"default_model": {
"name": "cross-encoder/ms-marco-MiniLM-L6-v2",
"max_length": 512,
"batch_size": 16
}
})
# Convert to ModelConfig objects
self.model_configs = {}
for name, model_config in self.models_config.items():
self.model_configs[name] = ModelConfig(**model_config)
# Adaptive configuration
adaptive_config = config.get("adaptive", {})
self.adaptive_enabled = adaptive_config.get("enabled", True)
self.confidence_threshold = adaptive_config.get("confidence_threshold", 0.7)
# Score fusion configuration
fusion_config = config.get("score_fusion", {})
self.fusion_method = fusion_config.get("method", "weighted")
weights_config = fusion_config.get("weights", {})
self.weights = WeightsConfig(
retrieval_score=weights_config.get("retrieval_score", 0.3),
neural_score=weights_config.get("neural_score", 0.7),
graph_score=weights_config.get("graph_score", 0.0),
temporal_score=weights_config.get("temporal_score", 0.0)
)
normalization_config = fusion_config.get("normalization", {})
self.normalization = NormalizationConfig(
method=normalization_config.get("method", "min_max"),
clip_outliers=normalization_config.get("clip_outliers", True),
outlier_threshold=normalization_config.get("outlier_threshold", 3.0)
)
# Performance configuration
perf_config = config.get("performance", {})
self.max_latency_ms = perf_config.get("max_latency_ms", 200)
self.target_latency_ms = perf_config.get("target_latency_ms", 150)
self.enable_caching = perf_config.get("enable_caching", True)
self.max_cache_size = perf_config.get("max_cache_size", 10000)
self.cache_ttl_seconds = perf_config.get("cache_ttl_seconds", 3600)
# Legacy compatibility
self.max_candidates = config.get("max_candidates", 50)
self.default_model = config.get("default_model", list(self.models_config.keys())[0])
def _initialize_if_needed(self) -> None:
"""Initialize advanced components lazily for better startup performance."""
if self._initialized or not self.enabled:
return
try:
# Initialize models manager
self.models_manager = CrossEncoderModels(self.model_configs)
# Initialize adaptive strategies
if self.adaptive_enabled:
self.adaptive_strategies = AdaptiveStrategies(
enabled=True,
confidence_threshold=self.confidence_threshold
)
# Initialize score fusion
self.score_fusion = ScoreFusion(
method=self.fusion_method,
weights=self.weights,
normalization=self.normalization
)
# Initialize performance optimizer
self.performance_optimizer = PerformanceOptimizer(
max_latency_ms=self.max_latency_ms,
target_latency_ms=self.target_latency_ms,
enable_caching=self.enable_caching,
cache_ttl_seconds=self.cache_ttl_seconds,
max_cache_size=self.max_cache_size
)
self._initialized = True
logger.info("Enhanced NeuralReranker initialization completed")
except Exception as e:
logger.error(f"Failed to initialize enhanced neural reranker: {e}")
self.enabled = False
raise NeuralRerankingError(f"Initialization failed: {e}")
def rerank(
self,
query: str,
documents: List[Document],
initial_scores: List[float]
) -> List[Tuple[int, float]]:
"""
Rerank documents using enhanced neural models with advanced strategies.
Args:
query: The search query
documents: List of candidate documents
initial_scores: Initial relevance scores from retrieval
Returns:
List of (document_index, reranked_score) tuples sorted by score
"""
start_time = time.time()
self.stats["total_queries"] += 1
try:
# Check if reranking is enabled
if not self.enabled:
return [(i, score) for i, score in enumerate(initial_scores)]
# Validate inputs
if not documents or not query.strip():
return []
if len(initial_scores) != len(documents):
logger.warning("Mismatch between documents and scores, using defaults")
initial_scores = [1.0] * len(documents)
# Initialize if needed
self._initialize_if_needed()
if not self._initialized:
logger.warning("Enhanced neural reranker not initialized, using initial scores")
return [(i, score) for i, score in enumerate(initial_scores)]
# Check cache first
cached_scores = self.performance_optimizer.get_cached_scores(
query, documents, self.default_model
)
if cached_scores is not None:
self.stats["cache_hits"] += 1
return [(i, score) for i, score in enumerate(cached_scores)]
self.stats["cache_misses"] += 1
# Limit candidates for performance
max_candidates = min(len(documents), self.max_candidates)
if len(documents) > max_candidates:
# Sort by initial scores and take top candidates
sorted_indices = sorted(range(len(initial_scores)),
key=lambda i: initial_scores[i], reverse=True)
top_indices = sorted_indices[:max_candidates]
documents = [documents[i] for i in top_indices]
initial_scores = [initial_scores[i] for i in top_indices]
else:
top_indices = list(range(len(documents)))
# Select optimal model for this query
selected_model = self._select_model_for_query(query)
# Get neural scores
neural_scores = self._get_neural_scores(query, documents, selected_model)
# Fuse scores using advanced fusion strategies
fused_scores = self.score_fusion.fuse_scores(
initial_scores, neural_scores, query, documents
)
# Cache the results
self.performance_optimizer.cache_scores(
query, documents, selected_model, fused_scores
)
# Create final results with original indices
results = []
for i, score in enumerate(fused_scores):
original_idx = top_indices[i] if len(documents) <= max_candidates else top_indices[i]
results.append((original_idx, score))
results.sort(key=lambda x: x[1], reverse=True)
# Update performance statistics
latency_ms = (time.time() - start_time) * 1000
self._update_stats(latency_ms, success=True)
self.performance_optimizer.record_latency(latency_ms)
# Record performance for adaptive learning
if self.adaptive_strategies:
query_analysis = self.adaptive_strategies.detector.classify_query(query) if self.adaptive_strategies.detector else None
query_type = query_analysis.query_type if query_analysis else "general"
quality_score = self._estimate_quality_score(fused_scores, initial_scores)
self.adaptive_strategies.record_performance(
selected_model, query_type, latency_ms, quality_score
)
# Log performance
logger.debug(f"Enhanced neural reranking completed in {latency_ms:.1f}ms for {len(documents)} documents")
return results
except Exception as e:
# Handle errors gracefully
latency_ms = (time.time() - start_time) * 1000
self._update_stats(latency_ms, success=False)
self.stats["fallback_activations"] += 1
logger.error(f"Enhanced neural reranking failed: {e}")
# Return fallback results
return [(i, score) for i, score in enumerate(initial_scores)]
def _select_model_for_query(self, query: str) -> str:
"""
Select the optimal model for the given query using adaptive strategies.
Args:
query: The search query
Returns:
Name of the selected model
"""
if not self.adaptive_strategies or not self.adaptive_strategies.enabled:
return self.default_model
try:
available_models = self.models_manager.get_available_models()
selected_model = self.adaptive_strategies.select_model(
query, available_models, self.default_model
)
if selected_model != self.default_model:
self.stats["model_switches"] += 1
self.stats["adaptive_adjustments"] += 1
return selected_model
except Exception as e:
logger.warning(f"Adaptive model selection failed: {e}, using default")
return self.default_model
def _get_neural_scores(
self,
query: str,
documents: List[Document],
model_name: str
) -> List[float]:
"""
Get neural relevance scores for query-document pairs.
Args:
query: The search query
documents: List of documents
model_name: Name of the model to use
Returns:
List of neural relevance scores
"""
try:
# Prepare query-document pairs
query_doc_pairs = []
for doc in documents:
doc_text = doc.content
# Smart truncation for model max_length
model_config = self.model_configs.get(model_name)
if model_config:
max_length = model_config.max_length
if len(doc_text) > max_length:
# Try to keep complete sentences
truncated = doc_text[:max_length - 50]
last_period = truncated.rfind('.')
if last_period > max_length // 2:
doc_text = truncated[:last_period + 1]
else:
doc_text = truncated + "..."
query_doc_pairs.append([query, doc_text])
# Get scores using models manager
neural_scores = self.models_manager.predict(query_doc_pairs, model_name)
return neural_scores
except Exception as e:
logger.error(f"Neural scoring failed: {e}")
return [0.0] * len(documents)
def _estimate_quality_score(
self,
fused_scores: List[float],
initial_scores: List[float]
) -> float:
"""
Estimate quality improvement from neural reranking.
Args:
fused_scores: Final fused scores
initial_scores: Initial retrieval scores
Returns:
Quality score (0-1)
"""
try:
# Simple heuristic: how much did the score distribution improve?
if not fused_scores or not initial_scores:
return 0.5
# Calculate variance of scores (higher variance = better discrimination)
fused_variance = np.var(fused_scores)
initial_variance = np.var(initial_scores)
# Calculate improvement ratio
if initial_variance > 0:
improvement_ratio = fused_variance / initial_variance
# Normalize to 0-1 range
quality_score = min(1.0, improvement_ratio / 2.0)
else:
quality_score = 0.5
return quality_score
except Exception:
return 0.5
def _update_stats(self, latency_ms: float, success: bool) -> None:
"""Update performance statistics."""
if success:
self.stats["successful_queries"] += 1
else:
self.stats["failed_queries"] += 1
self._error_count += 1
self.stats["total_latency_ms"] += latency_ms
def is_enabled(self) -> bool:
"""
Check if neural reranking is enabled.
Returns:
True if reranking should be performed
"""
# Return True if configured to be enabled, regardless of initialization status
# Initialization happens lazily when rerank() is called
return self.enabled
def get_reranker_info(self) -> Dict[str, Any]:
"""
Get information about the enhanced neural reranker.
Returns:
Dictionary with reranker configuration and statistics
"""
base_info = {
"type": "enhanced_neural_reranker",
"enabled": self.enabled,
"initialized": self._initialized,
"default_model": self.default_model,
"total_models": len(self.models_config),
"adaptive_enabled": self.adaptive_enabled,
"score_fusion_method": self.fusion_method,
"max_latency_ms": self.max_latency_ms,
"target_latency_ms": self.target_latency_ms,
"caching_enabled": self.enable_caching
}
# Add model information
if self._initialized and self.models_manager:
base_info["models"] = {
name: config.name
for name, config in self.model_configs.items()
}
base_info["model_stats"] = self.models_manager.get_model_stats()
# Add statistics
base_info["statistics"] = self.stats.copy()
# Add component statistics
if self._initialized:
if self.adaptive_strategies:
base_info["adaptive_stats"] = self.adaptive_strategies.get_stats()
if self.score_fusion:
base_info["fusion_stats"] = self.score_fusion.get_stats()
if self.performance_optimizer:
base_info["performance_stats"] = self.performance_optimizer.get_stats()
# Add performance metrics
if self.stats["total_queries"] > 0:
base_info["avg_latency_ms"] = self.stats["total_latency_ms"] / self.stats["total_queries"]
base_info["success_rate"] = self.stats["successful_queries"] / self.stats["total_queries"]
cache_total = self.stats["cache_hits"] + self.stats["cache_misses"]
if cache_total > 0:
base_info["cache_hit_rate"] = self.stats["cache_hits"] / cache_total
return base_info
def reset_stats(self) -> None:
"""Reset all statistics."""
self.stats = {
"total_queries": 0,
"successful_queries": 0,
"failed_queries": 0,
"total_latency_ms": 0.0,
"model_switches": 0,
"cache_hits": 0,
"cache_misses": 0,
"fallback_activations": 0,
"adaptive_adjustments": 0
}
if self._initialized:
if self.adaptive_strategies:
self.adaptive_strategies.reset_stats()
if self.score_fusion:
self.score_fusion.reset_stats()
if self.performance_optimizer:
self.performance_optimizer.reset_stats() |