Spaces:
Running
Running
File size: 12,354 Bytes
09801ca | 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 | """
✅ Evaluation & Generalization Agent
Final approval gate with comprehensive checks:
- Cross-validation consistency
- Robustness (perturbation tests)
- Drift sensitivity
- Overfitting detection
Model only approved if ALL checks pass.
"""
import numpy as np
from typing import Dict, List, Any, Tuple, Optional
from dataclasses import dataclass
import logging
from .base import BaseAgent, AgentResult, AgentStatus, Phase, MessageType
logger = logging.getLogger(__name__)
@dataclass
class EvaluationCheck:
"""Result of an evaluation check"""
name: str
passed: bool
score: float
threshold: float
details: str
class EvaluationAgent(BaseAgent):
"""
Evaluation & Generalization Agent
Comprehensive final checks → Committee approval → Release to production
"""
name = "evaluation"
description = "Final approval gate with robustness checks"
def __init__(self, memory=None):
super().__init__(memory)
self.checks: List[EvaluationCheck] = []
def execute(self, **kwargs) -> AgentResult:
"""Main execution: run all checks, make approval decision"""
# Get data and model
X = self.read_state("features_engineered")
if X is None:
X = self.read_state("features")
y = self.read_state("target")
task_type = self.read_state("task_type")
model_artifact = self.memory.get_latest_artifact("model")
if model_artifact is None or X is None or y is None:
return AgentResult(
status=AgentStatus.FAILED,
agent_name=self.name,
phase=self.current_phase,
errors=["Missing model or data"]
)
model = model_artifact.data
self.logger.info(f"🔍 Running evaluation checks...")
# Run checks based on phase
if self.is_fast_phase():
checks = self._fast_evaluation(model, X, y, task_type)
else:
checks = self._deep_evaluation(model, X, y, task_type)
self.checks = checks
# Determine approval
passed = all(c.passed for c in checks)
critical_fails = [c for c in checks if not c.passed]
avg_score = np.mean([c.score for c in checks])
if not passed:
# Create retry messages
result = AgentResult(
status=AgentStatus.RETRY,
agent_name=self.name,
phase=self.current_phase,
data={"checks": [c.__dict__ for c in checks]},
recommendations=[f"{c.name} failed: {c.details}" for c in critical_fails],
metrics={"score": avg_score}
)
for check in critical_fails:
result.add_message(
receiver="model_strategy",
msg_type=MessageType.RETRY,
payload={"failed_check": check.name}
)
self.logger.warning(f" ⚠️ {len(critical_fails)} checks failed")
return result
# All checks passed
self.write_state("evaluation_approved", True, self.name)
self.write_state("evaluation_checks", [c.__dict__ for c in checks], self.name)
self.logger.info(f" ✅ All {len(checks)} checks passed")
return AgentResult(
status=AgentStatus.SUCCESS,
agent_name=self.name,
phase=self.current_phase,
data={
"approved": True,
"checks_passed": len(checks)
},
metrics={
"score": avg_score
}
)
# =========================================================================
# FAST EVALUATION
# =========================================================================
def _fast_evaluation(self, model, X: np.ndarray, y: np.ndarray,
task_type: str) -> List[EvaluationCheck]:
"""Quick evaluation checks"""
checks = []
# 1. Cross-validation check
cv_check = self._check_cross_validation(model, X, y, task_type, n_folds=3)
checks.append(cv_check)
# 2. Quick overfitting check
overfit_check = self._check_overfitting(model, X, y, task_type)
checks.append(overfit_check)
return checks
# =========================================================================
# DEEP EVALUATION
# =========================================================================
def _deep_evaluation(self, model, X: np.ndarray, y: np.ndarray,
task_type: str) -> List[EvaluationCheck]:
"""Comprehensive evaluation checks"""
checks = []
# 1. Robust cross-validation (5-fold)
cv_check = self._check_cross_validation(model, X, y, task_type, n_folds=5)
checks.append(cv_check)
# 2. Overfitting check
overfit_check = self._check_overfitting(model, X, y, task_type)
checks.append(overfit_check)
# 3. Robustness (perturbation test)
robust_check = self._check_robustness(model, X, y, task_type)
checks.append(robust_check)
# 4. CV stability (std check)
stability_check = self._check_cv_stability(model, X, y, task_type)
checks.append(stability_check)
return checks
# =========================================================================
# CHECK IMPLEMENTATIONS
# =========================================================================
def _check_cross_validation(self, model, X: np.ndarray, y: np.ndarray,
task_type: str, n_folds: int = 5) -> EvaluationCheck:
"""Check cross-validation performance"""
try:
from sklearn.model_selection import cross_val_score, StratifiedKFold, KFold
if task_type == "classification":
cv = StratifiedKFold(n_splits=n_folds, shuffle=True, random_state=42)
scoring = "f1_weighted"
threshold = 0.5
else:
cv = KFold(n_splits=n_folds, shuffle=True, random_state=42)
scoring = "r2"
threshold = 0.1
scores = cross_val_score(
model.__class__(**model.get_params()),
X, y, cv=cv, scoring=scoring
)
mean_score = scores.mean()
passed = mean_score >= threshold
self.logger.info(f" 📊 CV Score: {mean_score:.4f} (threshold: {threshold})")
return EvaluationCheck(
name="cross_validation",
passed=passed,
score=mean_score,
threshold=threshold,
details=f"CV mean={mean_score:.4f}, std={scores.std():.4f}"
)
except Exception as e:
return EvaluationCheck(
name="cross_validation",
passed=False,
score=0,
threshold=0,
details=f"Failed: {str(e)[:50]}"
)
def _check_overfitting(self, model, X: np.ndarray, y: np.ndarray,
task_type: str) -> EvaluationCheck:
"""Check for overfitting via train-test gap"""
try:
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42
)
# Clone and refit
new_model = model.__class__(**model.get_params())
new_model.fit(X_train, y_train)
train_pred = new_model.predict(X_train)
test_pred = new_model.predict(X_test)
if task_type == "classification":
from sklearn.metrics import accuracy_score
train_score = accuracy_score(y_train, train_pred)
test_score = accuracy_score(y_test, test_pred)
else:
from sklearn.metrics import r2_score
train_score = r2_score(y_train, train_pred)
test_score = r2_score(y_test, test_pred)
gap = train_score - test_score
threshold = 0.15 # Max acceptable gap
passed = gap <= threshold
self.logger.info(f" 📊 Train-Test Gap: {gap:.4f} (threshold: {threshold})")
return EvaluationCheck(
name="overfitting",
passed=passed,
score=1 - gap, # Higher is better
threshold=1 - threshold,
details=f"train={train_score:.4f}, test={test_score:.4f}, gap={gap:.4f}"
)
except Exception as e:
return EvaluationCheck(
name="overfitting",
passed=True, # Pass by default if can't check
score=0.5,
threshold=0,
details=f"Could not check: {str(e)[:50]}"
)
def _check_robustness(self, model, X: np.ndarray, y: np.ndarray,
task_type: str) -> EvaluationCheck:
"""Check robustness via feature perturbation"""
try:
# Add small noise to features
noise = np.random.normal(0, 0.1, X.shape)
X_perturbed = X + noise
original_pred = model.predict(X[:1000])
perturbed_pred = model.predict(X_perturbed[:1000])
if task_type == "classification":
# Check prediction stability
stability = (original_pred == perturbed_pred).mean()
else:
# Check prediction correlation
stability = np.corrcoef(original_pred.flatten(), perturbed_pred.flatten())[0, 1]
stability = max(0, stability) # Ensure non-negative
threshold = 0.7
passed = stability >= threshold
self.logger.info(f" 📊 Robustness: {stability:.4f} (threshold: {threshold})")
return EvaluationCheck(
name="robustness",
passed=passed,
score=stability,
threshold=threshold,
details=f"Prediction stability under noise: {stability:.4f}"
)
except Exception as e:
return EvaluationCheck(
name="robustness",
passed=True,
score=0.8,
threshold=0.7,
details=f"Could not check: {str(e)[:50]}"
)
def _check_cv_stability(self, model, X: np.ndarray, y: np.ndarray,
task_type: str) -> EvaluationCheck:
"""Check cross-validation stability (low variance)"""
try:
from sklearn.model_selection import cross_val_score, KFold
cv = KFold(n_splits=5, shuffle=True, random_state=42)
scoring = "f1_weighted" if task_type == "classification" else "r2"
scores = cross_val_score(
model.__class__(**model.get_params()),
X, y, cv=cv, scoring=scoring
)
std = scores.std()
threshold = 0.05 # Max acceptable std
passed = std <= threshold
self.logger.info(f" 📊 CV Stability: std={std:.4f} (threshold: {threshold})")
return EvaluationCheck(
name="cv_stability",
passed=passed,
score=1 - std, # Higher is better
threshold=1 - threshold,
details=f"CV std={std:.4f}"
)
except Exception as e:
return EvaluationCheck(
name="cv_stability",
passed=True,
score=0.95,
threshold=0.95,
details=f"Could not check: {str(e)[:50]}"
)
|