Spaces:
Sleeping
Sleeping
File size: 4,775 Bytes
a4a265d | 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 | import os
import json
import numpy as np
import time
from typing import List, Any
from sklearn.metrics import accuracy_score, f1_score
from src.monitoring.logger import get_logger
from src.monitoring.alerting import AlertManager
from src.monitoring.drift_detector import DriftDetector
from typing import List, Any
logger = get_logger(__name__)
class PerformanceTracker:
"""Tracks continuous model performance degradation via feedback loops."""
FEEDBACK_FILE = "models/performance_log.json"
STAGING_FILE = "models/prediction_staging.json"
ANALYSIS_LOG = "models/analysis_log.json"
@classmethod
def stage_predictions(cls, request_ids: List[str], predictions: List[Any]) -> None:
"""Stores predictions in a staging area to await ground truth feedback."""
os.makedirs("models", exist_ok=True)
staging = {}
if os.path.exists(cls.STAGING_FILE):
with open(cls.STAGING_FILE, "r") as f:
staging = json.load(f)
for rid, pred in zip(request_ids, predictions):
staging[rid] = pred
# Limit staging size to 10k to prevent OOM
if len(staging) > 10000:
keys = list(staging.keys())
for key in keys[:-10000]:
staging.pop(key)
with open(cls.STAGING_FILE, "w") as f:
json.dump(staging, f)
@classmethod
def log_feedback(cls, request_ids: List[str], truths: List[Any]) -> None:
"""Resolves staged predictions with truth feedback for performance metrics."""
os.makedirs("models", exist_ok=True)
# Load staging
staging = {}
if os.path.exists(cls.STAGING_FILE):
with open(cls.STAGING_FILE, "r") as f:
staging = json.load(f)
history = {"truths": [], "preds": []}
if os.path.exists(cls.FEEDBACK_FILE):
with open(cls.FEEDBACK_FILE, "r") as f:
history = json.load(f)
resolved_count = 0
for rid, truth in zip(request_ids, truths):
if rid in staging:
history["truths"].append(truth)
history["preds"].append(staging.pop(rid))
resolved_count += 1
if resolved_count == 0:
logger.warning("Feedback received but no matching request_ids found in staging!")
return
# Enforce history limit
history["truths"] = history["truths"][-10000:]
history["preds"] = history["preds"][-10000:]
with open(cls.FEEDBACK_FILE, "w") as f:
json.dump(history, f)
with open(cls.STAGING_FILE, "w") as f:
json.dump(staging, f)
logger.info(f"Resolved {resolved_count} predictions with feedback.")
cls._calculate_running_metrics(history)
@classmethod
def _calculate_running_metrics(cls, history: dict) -> None:
"""Calculates F1/Accuracy over the continuous time window to track degradation."""
try:
y_true = np.array(history["truths"])
y_pred = np.array(history["preds"])
if len(y_true) < 10: # Avoid noise on small samples
return
acc = accuracy_score(y_true, y_pred)
f1 = f1_score(y_true, y_pred, average='weighted')
logger.info(f"Running Performance Metrics (N={len(y_true)}): Accuracy={acc:.4f}, F1={f1:.4f}")
# Log correlation between drift and performance
cls._log_correlation(float(acc), float(f1))
if f1 < 0.5:
msg = f"Running F1-Score crashed below 0.5 (Current={f1:.4f}). Model retraining strongly advised!"
AlertManager.send_alert("Model Performance Alert", msg, level="CRITICAL")
except Exception as e:
logger.error(f"Failed to calculate running metrics: {e}")
@classmethod
def _log_correlation(cls, acc: float, f1: float) -> None:
"""Saves unified snapshot of system intelligence (Drift vs. Performance)."""
snapshot = {
"timestamp": time.time(),
"performance": {"accuracy": acc, "f1": f1},
"drift_p_values": DriftDetector.latest_p_values
}
history = []
if os.path.exists(cls.ANALYSIS_LOG):
with open(cls.ANALYSIS_LOG, "r") as f:
history = json.load(f)
history.append(snapshot)
history = history[-100:] # Keep last 100 snapshots
with open(cls.ANALYSIS_LOG, "w") as f:
json.dump(history, f, indent=4)
logger.info(f"Drift-Performance correlation snapshot saved to {cls.ANALYSIS_LOG}")
|