Spaces:
Sleeping
Sleeping
File size: 19,609 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 |
"""
Data validation for backend migrations.
This module provides comprehensive validation tools for ensuring
data integrity during backend migrations, including document
validation, embedding verification, and consistency checks.
"""
import logging
from typing import List, Dict, Any, Set, Optional
import numpy as np
from collections import Counter
from src.core.interfaces import Document
logger = logging.getLogger(__name__)
class ValidationError(Exception):
"""Raised when validation fails."""
pass
class DataValidator:
"""
Comprehensive data validator for backend migrations.
This validator performs multiple levels of validation to ensure
data integrity during migration between different vector database
backends. It checks document structure, embedding quality,
metadata consistency, and content integrity.
Validation Categories:
- Document Structure: Content, metadata, embedding presence
- Embedding Quality: Dimension consistency, value ranges, NaN/inf checks
- Metadata Consistency: Required fields, data types, value ranges
- Content Integrity: Text quality, encoding issues, length validation
- Statistical Analysis: Distribution analysis, outlier detection
"""
def __init__(self, strict_mode: bool = False):
"""
Initialize the data validator.
Args:
strict_mode: Whether to use strict validation criteria
"""
self.strict_mode = strict_mode
# Validation thresholds
self.thresholds = {
"min_content_length": 10 if not strict_mode else 50,
"max_content_length": 100000,
"min_embedding_dim": 50,
"max_embedding_dim": 4096,
"max_embedding_value": 10.0,
"min_embedding_norm": 0.01,
"max_nan_ratio": 0.0,
"max_duplicate_ratio": 0.1 if not strict_mode else 0.05
}
logger.info(f"Data validator initialized (strict_mode={strict_mode})")
def validate_documents(self, documents: List[Document]) -> Dict[str, Any]:
"""
Perform comprehensive validation of document list.
Args:
documents: List of documents to validate
Returns:
Dictionary with validation results
"""
logger.info(f"Validating {len(documents)} documents...")
validation_result = {
"is_valid": True,
"total_documents": len(documents),
"issues": [],
"warnings": [],
"statistics": {},
"validation_details": {}
}
try:
# Basic document validation
self._validate_document_structure(documents, validation_result)
# Embedding validation
self._validate_embeddings(documents, validation_result)
# Metadata validation
self._validate_metadata(documents, validation_result)
# Content validation
self._validate_content(documents, validation_result)
# Statistical analysis
self._perform_statistical_analysis(documents, validation_result)
# Duplicate detection
self._detect_duplicates(documents, validation_result)
# Final validation status
validation_result["is_valid"] = len(validation_result["issues"]) == 0
if validation_result["is_valid"]:
logger.info("Document validation passed")
else:
logger.warning(f"Document validation failed with {len(validation_result['issues'])} issues")
return validation_result
except Exception as e:
logger.error(f"Validation failed with exception: {str(e)}")
validation_result["is_valid"] = False
validation_result["issues"].append(f"Validation exception: {str(e)}")
return validation_result
def _validate_document_structure(self, documents: List[Document], result: Dict[str, Any]) -> None:
"""Validate basic document structure."""
issues = []
for i, doc in enumerate(documents):
try:
# Check content presence
if not doc.content:
issues.append(f"Document {i}: Empty content")
# Check content type
if not isinstance(doc.content, str):
issues.append(f"Document {i}: Content is not string")
# Check metadata presence
if not hasattr(doc, 'metadata') or doc.metadata is None:
issues.append(f"Document {i}: Missing metadata")
elif not isinstance(doc.metadata, dict):
issues.append(f"Document {i}: Metadata is not dictionary")
# Check embedding presence
if not hasattr(doc, 'embedding') or doc.embedding is None:
issues.append(f"Document {i}: Missing embedding")
except Exception as e:
issues.append(f"Document {i}: Structure validation error - {str(e)}")
result["issues"].extend(issues)
result["validation_details"]["structure"] = {
"checked_documents": len(documents),
"structure_issues": len(issues)
}
def _validate_embeddings(self, documents: List[Document], result: Dict[str, Any]) -> None:
"""Validate embedding quality and consistency."""
issues = []
warnings = []
embedding_dims = []
embedding_norms = []
for i, doc in enumerate(documents):
if doc.embedding is None:
continue
try:
embedding = np.array(doc.embedding)
# Check dimension
dim = len(embedding)
embedding_dims.append(dim)
if dim < self.thresholds["min_embedding_dim"]:
issues.append(f"Document {i}: Embedding dimension too small ({dim})")
elif dim > self.thresholds["max_embedding_dim"]:
issues.append(f"Document {i}: Embedding dimension too large ({dim})")
# Check for NaN or infinite values
if np.any(np.isnan(embedding)):
issues.append(f"Document {i}: NaN values in embedding")
if np.any(np.isinf(embedding)):
issues.append(f"Document {i}: Infinite values in embedding")
# Check value ranges
max_val = np.max(np.abs(embedding))
if max_val > self.thresholds["max_embedding_value"]:
warnings.append(f"Document {i}: Large embedding values (max: {max_val:.2f})")
# Check embedding norm
norm = np.linalg.norm(embedding)
embedding_norms.append(norm)
if norm < self.thresholds["min_embedding_norm"]:
warnings.append(f"Document {i}: Very small embedding norm ({norm:.6f})")
# Check for zero embeddings
if np.all(embedding == 0):
issues.append(f"Document {i}: Zero embedding vector")
except Exception as e:
issues.append(f"Document {i}: Embedding validation error - {str(e)}")
# Check dimension consistency
if embedding_dims:
dim_counts = Counter(embedding_dims)
if len(dim_counts) > 1:
issues.append(f"Inconsistent embedding dimensions: {dict(dim_counts)}")
result["issues"].extend(issues)
result["warnings"].extend(warnings)
result["validation_details"]["embeddings"] = {
"checked_embeddings": len(embedding_dims),
"embedding_issues": len(issues),
"dimension_consistency": len(set(embedding_dims)) <= 1,
"common_dimension": max(embedding_dims, default=0) if embedding_dims else 0,
"norm_statistics": {
"mean": np.mean(embedding_norms) if embedding_norms else 0,
"std": np.std(embedding_norms) if embedding_norms else 0,
"min": np.min(embedding_norms) if embedding_norms else 0,
"max": np.max(embedding_norms) if embedding_norms else 0
}
}
def _validate_metadata(self, documents: List[Document], result: Dict[str, Any]) -> None:
"""Validate metadata consistency and completeness."""
issues = []
warnings = []
metadata_keys = set()
source_files = set()
for i, doc in enumerate(documents):
if doc.metadata is None:
continue
try:
# Collect metadata keys
metadata_keys.update(doc.metadata.keys())
# Check for source information
if "source" in doc.metadata:
source_files.add(doc.metadata["source"])
elif self.strict_mode:
warnings.append(f"Document {i}: Missing source metadata")
# Check for chunk information
if "chunk_index" in doc.metadata:
chunk_idx = doc.metadata["chunk_index"]
if not isinstance(chunk_idx, int) or chunk_idx < 0:
issues.append(f"Document {i}: Invalid chunk_index: {chunk_idx}")
# Check for page information
if "page" in doc.metadata:
page_num = doc.metadata["page"]
if not isinstance(page_num, (int, float)) or page_num < 0:
warnings.append(f"Document {i}: Invalid page number: {page_num}")
except Exception as e:
issues.append(f"Document {i}: Metadata validation error - {str(e)}")
result["issues"].extend(issues)
result["warnings"].extend(warnings)
result["validation_details"]["metadata"] = {
"unique_metadata_keys": list(metadata_keys),
"unique_sources": len(source_files),
"metadata_issues": len(issues)
}
def _validate_content(self, documents: List[Document], result: Dict[str, Any]) -> None:
"""Validate document content quality."""
issues = []
warnings = []
content_lengths = []
for i, doc in enumerate(documents):
if not doc.content:
continue
try:
content = doc.content
content_length = len(content)
content_lengths.append(content_length)
# Check content length
if content_length < self.thresholds["min_content_length"]:
warnings.append(f"Document {i}: Very short content ({content_length} chars)")
elif content_length > self.thresholds["max_content_length"]:
warnings.append(f"Document {i}: Very long content ({content_length} chars)")
# Check for encoding issues
try:
content.encode('utf-8')
except UnicodeEncodeError:
issues.append(f"Document {i}: Encoding issues detected")
# Check for suspicious characters
if '\x00' in content:
issues.append(f"Document {i}: Null bytes in content")
# Check content quality
if content.strip() != content:
warnings.append(f"Document {i}: Leading/trailing whitespace")
if content.count('\n\n\n') > content_length / 100:
warnings.append(f"Document {i}: Excessive empty lines")
except Exception as e:
issues.append(f"Document {i}: Content validation error - {str(e)}")
result["issues"].extend(issues)
result["warnings"].extend(warnings)
result["validation_details"]["content"] = {
"checked_documents": len(content_lengths),
"content_issues": len(issues),
"length_statistics": {
"mean": np.mean(content_lengths) if content_lengths else 0,
"std": np.std(content_lengths) if content_lengths else 0,
"min": np.min(content_lengths) if content_lengths else 0,
"max": np.max(content_lengths) if content_lengths else 0
}
}
def _perform_statistical_analysis(self, documents: List[Document], result: Dict[str, Any]) -> None:
"""Perform statistical analysis of the document collection."""
stats = {}
try:
# Basic statistics
stats["total_documents"] = len(documents)
stats["documents_with_embeddings"] = sum(1 for doc in documents if doc.embedding)
stats["documents_with_metadata"] = sum(1 for doc in documents if doc.metadata)
# Content statistics
content_lengths = [len(doc.content) for doc in documents if doc.content]
if content_lengths:
stats["content_length"] = {
"mean": np.mean(content_lengths),
"median": np.median(content_lengths),
"std": np.std(content_lengths),
"min": np.min(content_lengths),
"max": np.max(content_lengths)
}
# Embedding statistics
embeddings = [doc.embedding for doc in documents if doc.embedding]
if embeddings:
embedding_matrix = np.array(embeddings)
stats["embeddings"] = {
"dimension": embedding_matrix.shape[1],
"mean_norm": np.mean(np.linalg.norm(embedding_matrix, axis=1)),
"value_range": {
"min": np.min(embedding_matrix),
"max": np.max(embedding_matrix),
"mean": np.mean(embedding_matrix),
"std": np.std(embedding_matrix)
}
}
# Source distribution
sources = [doc.metadata.get("source", "unknown") for doc in documents if doc.metadata]
if sources:
source_counts = Counter(sources)
stats["sources"] = {
"unique_sources": len(source_counts),
"documents_per_source": dict(source_counts.most_common(10))
}
except Exception as e:
result["warnings"].append(f"Statistical analysis failed: {str(e)}")
result["statistics"] = stats
def _detect_duplicates(self, documents: List[Document], result: Dict[str, Any]) -> None:
"""Detect potential duplicate documents."""
issues = []
warnings = []
try:
# Content-based duplicate detection
content_hashes = {}
embedding_similarities = []
for i, doc in enumerate(documents):
if doc.content:
# Simple hash-based duplicate detection
content_hash = hash(doc.content.strip().lower())
if content_hash in content_hashes:
warnings.append(f"Potential duplicate content: documents {content_hashes[content_hash]} and {i}")
else:
content_hashes[content_hash] = i
# Embedding-based similarity (sample check for performance)
embeddings = [(i, doc.embedding) for i, doc in enumerate(documents) if doc.embedding]
if len(embeddings) > 1:
# Sample for performance
sample_size = min(100, len(embeddings))
sample_indices = np.random.choice(len(embeddings), sample_size, replace=False)
for i in range(len(sample_indices)):
for j in range(i + 1, len(sample_indices)):
idx1, emb1 = embeddings[sample_indices[i]]
idx2, emb2 = embeddings[sample_indices[j]]
similarity = np.dot(emb1, emb2) / (np.linalg.norm(emb1) * np.linalg.norm(emb2))
if similarity > 0.99: # Very high similarity threshold
warnings.append(f"Very similar embeddings: documents {idx1} and {idx2} (similarity: {similarity:.3f})")
# Check duplicate ratio
unique_content = len(content_hashes)
total_with_content = sum(1 for doc in documents if doc.content)
if total_with_content > 0:
duplicate_ratio = 1 - (unique_content / total_with_content)
if duplicate_ratio > self.thresholds["max_duplicate_ratio"]:
issues.append(f"High duplicate ratio: {duplicate_ratio:.2%}")
except Exception as e:
warnings.append(f"Duplicate detection failed: {str(e)}")
result["issues"].extend(issues)
result["warnings"].extend(warnings)
result["validation_details"]["duplicates"] = {
"duplicate_issues": len([issue for issue in issues if "duplicate" in issue.lower()]),
"similarity_warnings": len([warning for warning in warnings if "similar" in warning.lower()])
}
def validate_migration_consistency(self,
source_documents: List[Document],
target_count: int) -> Dict[str, Any]:
"""
Validate consistency between source and target after migration.
Args:
source_documents: Original documents
target_count: Number of documents in target system
Returns:
Dictionary with consistency validation results
"""
result = {
"is_consistent": True,
"issues": [],
"warnings": [],
"comparison": {}
}
try:
source_count = len(source_documents)
# Count consistency
if source_count != target_count:
result["is_consistent"] = False
result["issues"].append(f"Document count mismatch: source={source_count}, target={target_count}")
# Additional consistency checks would go here
# (e.g., sampling documents and comparing content)
result["comparison"] = {
"source_count": source_count,
"target_count": target_count,
"count_match": source_count == target_count
}
except Exception as e:
result["is_consistent"] = False
result["issues"].append(f"Consistency validation failed: {str(e)}")
return result |