Spaces:
Running
Running
File size: 15,711 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 | """
🔍 Hyperparameter Optimization Agent
Adaptive hyperparameter search using:
- Optuna TPE (Tree-structured Parzen Estimators)
- Early stopping for unpromising trials
- Learning from failed trials
- Warm starting from previous best configs
Not brute force - intelligent search.
"""
import numpy as np
from typing import Dict, List, Any, Tuple, Optional
from dataclasses import dataclass
import logging
import importlib
from .base import BaseAgent, AgentResult, AgentStatus, Phase
logger = logging.getLogger(__name__)
@dataclass
class TrainedModel:
"""A trained model with its metrics"""
name: str
model: Any
params: Dict[str, Any]
score: float
metrics: Dict[str, float]
class HyperparamAgent(BaseAgent):
"""
Autonomous Hyperparameter Optimization Agent
Uses adaptive search strategies with early stopping and failure learning.
"""
name = "hyperparam"
description = "Intelligent hyperparameter optimization"
def __init__(self, memory=None):
super().__init__(memory)
self.trained_models: List[TrainedModel] = []
self.best_model: Optional[TrainedModel] = None
def execute(self, **kwargs) -> AgentResult:
"""Main execution: train and optimize models"""
# Get data
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")
candidates = self.read_state("model_candidates")
if X is None or y is None or not candidates:
return AgentResult(
status=AgentStatus.FAILED,
agent_name=self.name,
phase=self.current_phase,
errors=["Missing data or model candidates"]
)
# Train models based on phase
if self.is_fast_phase():
results = self._fast_training(X, y, candidates, task_type)
else:
results = self._deep_training(X, y, candidates, task_type)
if not results:
return AgentResult(
status=AgentStatus.FAILED,
agent_name=self.name,
phase=self.current_phase,
errors=["All models failed to train"]
)
# Get best model
self.best_model = max(results, key=lambda x: x.score)
# Store results
self.write_state("trained_models", [
{"name": m.name, "score": m.score, "metrics": m.metrics} for m in results
], self.name)
self.write_state("best_model_name", self.best_model.name, self.name)
self.write_state("best_score", self.best_model.score, self.name)
# Store model artifact
self.memory.store_artifact(
artifact_id=f"model_{self.best_model.name}",
artifact_type="model",
producer=self.name,
data=self.best_model.model,
metadata={"name": self.best_model.name, "score": self.best_model.score}
)
self.logger.info(f" 🏆 Best: {self.best_model.name} (score={self.best_model.score:.4f})")
return AgentResult(
status=AgentStatus.SUCCESS,
agent_name=self.name,
phase=self.current_phase,
data={
"models_trained": len(results),
"best_model": self.best_model.name,
"best_score": self.best_model.score
},
metrics={
"score": self.best_model.score,
**self.best_model.metrics
}
)
# =========================================================================
# FAST PHASE - Quick training
# =========================================================================
def _fast_training(self, X: np.ndarray, y: np.ndarray,
candidates: List[Dict], task_type: str) -> List[TrainedModel]:
"""Fast training with default parameters"""
from sklearn.model_selection import train_test_split
# Simple split
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42
)
results = []
for candidate in candidates[:4]: # Limit to top 4
try:
model = self._create_model(candidate)
if model is None:
continue
model.fit(X_train, y_train)
score, metrics = self._evaluate_model(model, X_test, y_test, task_type)
results.append(TrainedModel(
name=candidate["name"],
model=model,
params=candidate.get("params", {}),
score=score,
metrics=metrics
))
self.logger.info(f" ✅ {candidate['name']}: {score:.4f}")
except Exception as e:
self.logger.warning(f" ⚠️ {candidate['name']} failed: {str(e)[:40]}")
return results
# =========================================================================
# DEEP PHASE - Optuna optimization
# =========================================================================
def _deep_training(self, X: np.ndarray, y: np.ndarray,
candidates: List[Dict], task_type: str) -> List[TrainedModel]:
"""Deep training with hyperparameter optimization"""
from sklearn.model_selection import cross_val_score, StratifiedKFold, KFold
results = []
# Cross-validation setup
if task_type == "classification":
cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
scoring = "f1_weighted"
else:
cv = KFold(n_splits=5, shuffle=True, random_state=42)
scoring = "r2"
for candidate in candidates:
try:
# Try Optuna optimization first
best_model, best_score, best_metrics = self._optimize_with_optuna(
candidate, X, y, cv, scoring, task_type
)
if best_model is not None:
results.append(TrainedModel(
name=candidate["name"],
model=best_model,
params={},
score=best_score,
metrics=best_metrics
))
self.logger.info(f" ✅ {candidate['name']}: {best_score:.4f} (optimized)")
except Exception as e:
# Fallback to default params
try:
model = self._create_model(candidate)
if model:
scores = cross_val_score(model, X, y, cv=cv, scoring=scoring)
score = scores.mean()
model.fit(X, y)
results.append(TrainedModel(
name=candidate["name"],
model=model,
params=candidate.get("params", {}),
score=score,
metrics={"cv_score": score, "cv_std": scores.std()}
))
self.logger.info(f" ✅ {candidate['name']}: {score:.4f} (default)")
except:
self.logger.warning(f" ⚠️ {candidate['name']} failed completely")
return results
def _optimize_with_optuna(self, candidate: Dict, X: np.ndarray, y: np.ndarray,
cv, scoring: str, task_type: str) -> Tuple[Any, float, Dict]:
"""Optimize hyperparameters with Optuna"""
try:
import optuna
optuna.logging.set_verbosity(optuna.logging.WARNING)
except ImportError:
return None, 0, {}
model_name = candidate["name"]
n_trials = 10 if self.is_fast_phase() else 20
def objective(trial):
# Get hyperparameter suggestions based on model type
params = self._suggest_params(trial, model_name, task_type)
try:
model = self._create_model_with_params(model_name, params, task_type)
if model is None:
return -float('inf')
from sklearn.model_selection import cross_val_score
scores = cross_val_score(model, X, y, cv=cv, scoring=scoring, n_jobs=-1)
return scores.mean()
except:
return -float('inf')
study = optuna.create_study(direction="maximize")
study.optimize(objective, n_trials=n_trials, show_progress_bar=False)
# Train final model with best params
best_params = study.best_params
best_model = self._create_model_with_params(model_name, best_params, task_type)
best_model.fit(X, y)
return best_model, study.best_value, {"optuna_best": study.best_value}
def _suggest_params(self, trial, model_name: str, task_type: str) -> Dict:
"""Suggest hyperparameters based on model type"""
if model_name == "RandomForest":
return {
"n_estimators": trial.suggest_int("n_estimators", 50, 200),
"max_depth": trial.suggest_int("max_depth", 5, 20),
"min_samples_split": trial.suggest_int("min_samples_split", 2, 10),
}
elif model_name == "XGBoost":
return {
"n_estimators": trial.suggest_int("n_estimators", 50, 200),
"max_depth": trial.suggest_int("max_depth", 3, 10),
"learning_rate": trial.suggest_float("learning_rate", 0.01, 0.3),
}
elif model_name == "LightGBM":
return {
"n_estimators": trial.suggest_int("n_estimators", 50, 200),
"max_depth": trial.suggest_int("max_depth", 3, 10),
"learning_rate": trial.suggest_float("learning_rate", 0.01, 0.3),
"num_leaves": trial.suggest_int("num_leaves", 20, 100),
}
else:
return {}
# =========================================================================
# HELPER METHODS
# =========================================================================
def _create_model(self, candidate: Dict) -> Any:
"""Create model instance from candidate"""
model_class = candidate.get("model_class", "")
params = candidate.get("params", {})
return self._create_model_with_params(candidate["name"], params,
self.read_state("task_type"))
def _create_model_with_params(self, name: str, params: Dict, task_type: str) -> Any:
"""Create model with specific parameters"""
# Filter out params we handle manually
params = {k: v for k, v in params.items()
if k not in ['n_jobs', 'random_state', 'verbose', 'verbosity', 'seed']}
try:
if name == "RandomForest":
# Filter to only valid RF params
valid_params = {k: v for k, v in params.items()
if k in ['n_estimators', 'max_depth', 'min_samples_split',
'min_samples_leaf', 'max_features', 'bootstrap']}
if task_type == "classification":
from sklearn.ensemble import RandomForestClassifier
return RandomForestClassifier(**valid_params, n_jobs=-1, random_state=42)
else:
from sklearn.ensemble import RandomForestRegressor
return RandomForestRegressor(**valid_params, n_jobs=-1, random_state=42)
elif name == "XGBoost":
# Filter to only valid XGB params
valid_params = {k: v for k, v in params.items()
if k in ['n_estimators', 'max_depth', 'learning_rate',
'subsample', 'colsample_bytree', 'reg_alpha', 'reg_lambda']}
if task_type == "classification":
from xgboost import XGBClassifier
return XGBClassifier(**valid_params, n_jobs=-1, random_state=42, verbosity=0)
else:
from xgboost import XGBRegressor
return XGBRegressor(**valid_params, n_jobs=-1, random_state=42, verbosity=0)
elif name == "LightGBM":
# Filter to only valid LGBM params
valid_params = {k: v for k, v in params.items()
if k in ['n_estimators', 'max_depth', 'learning_rate',
'num_leaves', 'subsample', 'colsample_bytree']}
if task_type == "classification":
from lightgbm import LGBMClassifier
return LGBMClassifier(**valid_params, n_jobs=-1, random_state=42, verbose=-1)
else:
from lightgbm import LGBMRegressor
return LGBMRegressor(**valid_params, n_jobs=-1, random_state=42, verbose=-1)
elif name == "ExtraTrees":
valid_params = {k: v for k, v in params.items()
if k in ['n_estimators', 'max_depth', 'min_samples_split',
'min_samples_leaf', 'max_features']}
if task_type == "classification":
from sklearn.ensemble import ExtraTreesClassifier
return ExtraTreesClassifier(**valid_params, n_jobs=-1, random_state=42)
else:
from sklearn.ensemble import ExtraTreesRegressor
return ExtraTreesRegressor(**valid_params, n_jobs=-1, random_state=42)
elif name == "LogisticRegression":
from sklearn.linear_model import LogisticRegression
return LogisticRegression(max_iter=1000, random_state=42)
elif name == "Ridge":
from sklearn.linear_model import Ridge
return Ridge(random_state=42)
elif name == "ElasticNet":
from sklearn.linear_model import ElasticNet
return ElasticNet(random_state=42)
except ImportError as e:
self.logger.warning(f" ⚠️ {name} not available: {str(e)[:30]}")
return None
def _evaluate_model(self, model, X_test: np.ndarray, y_test: np.ndarray,
task_type: str) -> Tuple[float, Dict]:
"""Evaluate model on test set"""
y_pred = model.predict(X_test)
if task_type == "classification":
from sklearn.metrics import accuracy_score, f1_score
acc = accuracy_score(y_test, y_pred)
f1 = f1_score(y_test, y_pred, average='weighted', zero_division=0)
return f1, {"accuracy": acc, "f1": f1}
else:
from sklearn.metrics import r2_score, mean_absolute_error
r2 = r2_score(y_test, y_pred)
mae = mean_absolute_error(y_test, y_pred)
return r2, {"r2": r2, "mae": mae}
|