File size: 8,740 Bytes
198ccb0 | 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 | """Data drift detection utilities."""
import logging
from typing import Dict, List, Optional, Tuple
import numpy as np
import pandas as pd
from datetime import datetime, timedelta
from scipy import stats
from collections import defaultdict
import json
from pathlib import Path
logger = logging.getLogger(__name__)
class DataDriftDetector:
"""
Detect data drift in input features.
Compares current data distribution to reference (training) distribution.
"""
def __init__(
self,
reference_data: Optional[pd.DataFrame] = None,
reference_stats: Optional[Dict] = None,
drift_threshold: float = 0.05,
):
"""
Initialize data drift detector.
Args:
reference_data: Reference DataFrame (training data)
reference_stats: Pre-computed reference statistics
drift_threshold: Threshold for drift detection (5%)
"""
self.drift_threshold = drift_threshold
self.reference_stats = reference_stats or {}
if reference_data is not None:
self.reference_stats = self._compute_statistics(reference_data)
# Store current data for comparison
self.current_data: List[Dict] = []
logger.info("DataDriftDetector initialized")
def _compute_statistics(self, data: pd.DataFrame) -> Dict:
"""
Compute statistics for reference data.
Args:
data: DataFrame with features
Returns:
Dictionary of statistics
"""
stats_dict = {}
# Text length statistics
if "title" in data.columns:
title_lengths = data["title"].str.len()
stats_dict["title_length"] = {
"mean": float(title_lengths.mean()),
"std": float(title_lengths.std()),
"min": float(title_lengths.min()),
"max": float(title_lengths.max()),
}
if "snippet" in data.columns:
snippet_lengths = data["snippet"].str.len()
stats_dict["snippet_length"] = {
"mean": float(snippet_lengths.mean()),
"std": float(snippet_lengths.std()),
"min": float(snippet_lengths.min()),
"max": float(snippet_lengths.max()),
}
# Word count statistics
if "title" in data.columns:
word_counts = data["title"].str.split().str.len()
stats_dict["title_word_count"] = {
"mean": float(word_counts.mean()),
"std": float(word_counts.std()),
}
# Character distribution (for Russian text)
if "title" in data.columns:
cyrillic_ratio = data["title"].apply(
lambda x: len([c for c in str(x) if '\u0400' <= c <= '\u04FF']) / max(len(str(x)), 1)
)
stats_dict["cyrillic_ratio"] = {
"mean": float(cyrillic_ratio.mean()),
"std": float(cyrillic_ratio.std()),
}
return stats_dict
def record_sample(
self,
title: str,
snippet: Optional[str] = None,
metadata: Optional[Dict] = None,
) -> None:
"""
Record a sample for drift detection.
Args:
title: Article title
snippet: Optional snippet
metadata: Optional metadata
"""
sample = {
"timestamp": datetime.now().isoformat(),
"title": title,
"snippet": snippet,
"metadata": metadata or {},
}
self.current_data.append(sample)
# Keep only recent samples (last 1000)
if len(self.current_data) > 1000:
self.current_data = self.current_data[-1000:]
def detect_drift(
self,
window_size: int = 100,
) -> Tuple[bool, Dict]:
"""
Detect data drift in recent samples.
Args:
window_size: Number of recent samples to analyze
Returns:
Tuple of (has_drift, drift_info)
"""
if len(self.current_data) < window_size:
return False, {"error": "Insufficient data"}
if not self.reference_stats:
return False, {"error": "No reference statistics"}
# Get recent samples
recent = self.current_data[-window_size:]
recent_df = pd.DataFrame(recent)
# Compute current statistics
current_stats = self._compute_statistics(recent_df)
# Compare with reference
drift_info = {}
has_drift = False
for feature, ref_stats in self.reference_stats.items():
if feature not in current_stats:
continue
curr_stats = current_stats[feature]
# Compare means (relative change)
if "mean" in ref_stats and "mean" in curr_stats:
ref_mean = ref_stats["mean"]
curr_mean = curr_stats["mean"]
if ref_mean > 0:
relative_change = abs(curr_mean - ref_mean) / ref_mean
drift_info[feature] = {
"reference_mean": ref_mean,
"current_mean": curr_mean,
"relative_change": relative_change,
"drifted": relative_change > self.drift_threshold,
}
if relative_change > self.drift_threshold:
has_drift = True
logger.warning(
f"Data drift detected in {feature}: "
f"relative change {relative_change:.2%} "
f"(ref: {ref_mean:.2f}, curr: {curr_mean:.2f})"
)
return has_drift, drift_info
def statistical_test(
self,
feature: str,
window_size: int = 100,
) -> Dict:
"""
Perform statistical test for drift (KS test).
Args:
feature: Feature to test
window_size: Number of recent samples
Returns:
Dictionary with test results
"""
if len(self.current_data) < window_size:
return {"error": "Insufficient data"}
recent = self.current_data[-window_size:]
recent_df = pd.DataFrame(recent)
# Extract feature values
if feature == "title_length":
current_values = recent_df["title"].str.len().values
elif feature == "snippet_length":
current_values = recent_df["snippet"].str.len().values
elif feature == "title_word_count":
current_values = recent_df["title"].str.split().str.len().values
else:
return {"error": f"Unknown feature: {feature}"}
# Get reference statistics
if feature not in self.reference_stats:
return {"error": f"No reference stats for {feature}"}
ref_stats = self.reference_stats[feature]
# Generate reference distribution (normal approximation)
if "mean" in ref_stats and "std" in ref_stats:
ref_mean = ref_stats["mean"]
ref_std = ref_stats["std"]
# Kolmogorov-Smirnov test
# Create reference distribution sample
ref_sample = np.random.normal(ref_mean, ref_std, size=1000)
# Perform KS test
statistic, p_value = stats.ks_2samp(current_values, ref_sample)
return {
"feature": feature,
"ks_statistic": float(statistic),
"p_value": float(p_value),
"drifted": p_value < 0.05, # Significant drift
"reference_mean": ref_mean,
"current_mean": float(np.mean(current_values)),
}
return {"error": "Insufficient reference statistics"}
def save_reference_stats(self, path: str) -> None:
"""Save reference statistics to file."""
with open(path, 'w') as f:
json.dump(self.reference_stats, f, indent=2)
logger.info(f"Reference statistics saved to {path}")
def load_reference_stats(self, path: str) -> None:
"""Load reference statistics from file."""
with open(path) as f:
self.reference_stats = json.load(f)
logger.info(f"Reference statistics loaded from {path}")
|