File size: 11,339 Bytes
0162f5e |
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 |
"""
ML Model Trainer for Schedule Optimization
Handles model training and retraining with multiple models and ensemble
"""
import os
import pickle
import json
from datetime import datetime, timedelta
from pathlib import Path
from typing import Optional, Dict, Tuple
import numpy as np
from sklearn.ensemble import GradientBoostingRegressor, RandomForestRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, r2_score
import xgboost as xgb
import catboost as cb
import lightgbm as lgb
from .config import CONFIG
from .data_store import ScheduleDataStore
from .feature_extractor import FeatureExtractor
class ModelTrainer:
"""Train and manage ML models for schedule optimization"""
def __init__(self, model_dir: Optional[str] = None):
self.model_dir = Path(model_dir or CONFIG.MODEL_DIR)
self.model_dir.mkdir(parents=True, exist_ok=True)
self.data_store = ScheduleDataStore()
self.feature_extractor = FeatureExtractor()
self.models = {} # Dictionary of trained models
self.model_scores = {} # Performance scores for each model
self.ensemble_weights = {} # Weights for ensemble
self.best_model_name = None
self.last_trained = None
self.training_history = []
def _get_model(self, model_name: str):
"""Get model instance by name"""
if model_name == "gradient_boosting":
return GradientBoostingRegressor(
n_estimators=CONFIG.EPOCHS,
learning_rate=CONFIG.LEARNING_RATE,
random_state=42
)
elif model_name == "random_forest":
return RandomForestRegressor(
n_estimators=CONFIG.EPOCHS,
random_state=42,
n_jobs=-1
)
elif model_name == "xgboost":
return xgb.XGBRegressor(
n_estimators=CONFIG.EPOCHS,
learning_rate=CONFIG.LEARNING_RATE,
random_state=42,
verbosity=0
)
elif model_name == "lightgbm":
return lgb.LGBMRegressor(
n_estimators=CONFIG.EPOCHS,
learning_rate=CONFIG.LEARNING_RATE,
random_state=42,
verbose=-1
)
elif model_name == "catboost":
return cb.CatBoostRegressor(
iterations=CONFIG.EPOCHS,
learning_rate=CONFIG.LEARNING_RATE,
random_state=42,
verbose=False
)
return None
def should_retrain(self) -> bool:
"""Check if model should be retrained"""
if not self.last_trained:
# Never trained
return True
# Check time since last training
hours_since_training = (
datetime.now() - self.last_trained
).total_seconds() / 3600
if hours_since_training >= CONFIG.RETRAIN_INTERVAL_HOURS:
# Check if enough new data
new_schedules = self.data_store.get_schedules_since(self.last_trained)
if len(new_schedules) >= CONFIG.MIN_SCHEDULES_FOR_RETRAIN:
return True
return False
def train(self, force: bool = False) -> Dict:
"""Train or retrain all models"""
if not force and not self.should_retrain():
return {
"success": False,
"reason": "Retraining not needed yet"
}
# Load data
schedules = self.data_store.load_schedules()
if len(schedules) < CONFIG.MIN_SCHEDULES_FOR_TRAINING:
return {
"success": False,
"reason": f"Not enough data. Need {CONFIG.MIN_SCHEDULES_FOR_TRAINING}, have {len(schedules)}"
}
# Prepare dataset
X, y = self.feature_extractor.prepare_dataset(schedules)
if len(X) == 0:
return {
"success": False,
"error": "No valid features extracted"
}
# Split data
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=CONFIG.TRAIN_TEST_SPLIT, random_state=42
)
# Train all models
self.models = {}
self.model_scores = {}
all_metrics = {}
for model_name in CONFIG.MODEL_TYPES:
print(f"Training {model_name}...")
model = self._get_model(model_name)
if model is None:
print(f"Skipping {model_name} - not available")
continue
# Train model
model.fit(X_train, y_train)
# Evaluate
train_pred = model.predict(X_train)
test_pred = model.predict(X_test)
train_r2 = r2_score(y_train, train_pred) # type: ignore
test_r2 = r2_score(y_test, test_pred) # type: ignore
test_rmse = np.sqrt(mean_squared_error(y_test, test_pred)) # type: ignore
self.models[model_name] = model
self.model_scores[model_name] = test_r2
all_metrics[model_name] = {
"train_r2": train_r2,
"test_r2": test_r2,
"train_rmse": np.sqrt(mean_squared_error(y_train, train_pred)), # type: ignore
"test_rmse": test_rmse
}
print(f" {model_name}: R² = {test_r2:.4f}, RMSE = {test_rmse:.4f}")
# Compute ensemble weights based on performance
if CONFIG.USE_ENSEMBLE and len(self.models) > 1:
total_score = sum(self.model_scores.values())
self.ensemble_weights = {
name: score / total_score
for name, score in self.model_scores.items()
}
else:
self.ensemble_weights = {}
# Find best model
if self.model_scores:
self.best_model_name = max(self.model_scores.items(), key=lambda x: x[1])[0]
# Save model
self.last_trained = datetime.now()
self.save_model()
# Record training history
history_entry = {
"timestamp": self.last_trained.isoformat(),
"metrics": all_metrics,
"best_model": self.best_model_name,
"ensemble_weights": self.ensemble_weights,
"config": {
"models_trained": list(self.models.keys()),
"version": CONFIG.MODEL_VERSION
}
}
self.training_history.append(history_entry)
self._save_history()
return {
"success": True,
"models_trained": list(self.models.keys()),
"best_model": self.best_model_name,
"metrics": all_metrics,
"ensemble_weights": self.ensemble_weights,
"samples_used": len(X),
"timestamp": self.last_trained.isoformat()
}
def predict(self, features: Dict[str, float], use_ensemble: bool = True) -> Tuple[float, float]:
"""Predict schedule quality and confidence"""
if not self.models:
self.load_model()
if not self.models:
return 0.0, 0.0
# Convert features to vector
feature_vector = np.array([
[features.get(f, 0.0) for f in CONFIG.FEATURES]
])
if use_ensemble and CONFIG.USE_ENSEMBLE and self.ensemble_weights:
# Ensemble prediction
prediction = 0.0
for model_name, weight in self.ensemble_weights.items():
if model_name in self.models:
pred = self.models[model_name].predict(feature_vector)[0]
prediction += weight * pred
# Confidence based on ensemble agreement
predictions = [
self.models[name].predict(feature_vector)[0]
for name in self.models.keys()
]
std_dev = np.std(predictions)
confidence = max(0.5, min(1.0, 1.0 - (std_dev / 50))) # Higher agreement = higher confidence
else:
# Use best single model
best_model = self.models.get(self.best_model_name)
if best_model is None:
best_model = list(self.models.values())[0]
prediction = best_model.predict(feature_vector)[0]
confidence = min(1.0, 0.8 + (prediction / 100) * 0.2)
return float(prediction), float(confidence)
def save_model(self):
"""Save all models to disk"""
if not self.models:
return
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
model_path = self.model_dir / f"models_{timestamp}.pkl"
latest_path = self.model_dir / "models_latest.pkl"
model_data = {
"models": self.models,
"ensemble_weights": self.ensemble_weights,
"best_model_name": self.best_model_name,
"last_trained": self.last_trained,
"config": {
"version": CONFIG.MODEL_VERSION,
"features": CONFIG.FEATURES,
"models_trained": list(self.models.keys())
}
}
with open(model_path, 'wb') as f:
pickle.dump(model_data, f)
with open(latest_path, 'wb') as f:
pickle.dump(model_data, f)
def load_model(self) -> bool:
"""Load models from disk"""
latest_path = self.model_dir / "models_latest.pkl"
if not latest_path.exists():
return False
try:
with open(latest_path, 'rb') as f:
model_data = pickle.load(f)
self.models = model_data["models"]
self.ensemble_weights = model_data.get("ensemble_weights", {})
self.best_model_name = model_data.get("best_model_name")
self.last_trained = model_data.get("last_trained")
return True
except Exception as e:
print(f"Error loading models: {e}")
return False
def _save_history(self):
"""Save training history"""
history_path = self.model_dir / "training_history.json"
with open(history_path, 'w') as f:
json.dump(self.training_history, f, indent=2, default=str)
def get_model_info(self) -> Dict:
"""Get information about current models"""
if not self.models:
self.load_model()
return {
"models_loaded": list(self.models.keys()) if self.models else [],
"best_model": self.best_model_name,
"ensemble_enabled": CONFIG.USE_ENSEMBLE,
"ensemble_weights": self.ensemble_weights,
"last_trained": self.last_trained.isoformat() if self.last_trained else None,
"should_retrain": self.should_retrain(),
"schedules_available": self.data_store.count_schedules(),
"training_runs": len(self.training_history)
}
|