train_gridsearch.py

"""
10-hour hyperparameter search with extensive grid.

This will train many models with different hyperparameters
and select the best one.
"""

from algo_trader.research.train import train_single_split
from algo_trader.research.dataset import build_ml_table, load_features_labels
from algo_trader.research.splits import make_walk_forward_splits
from algo_trader.common.logger import setup_logger, logger
from algo_trader.common.config import settings
from sklearn.model_selection import GridSearchCV
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from datetime import datetime
import numpy as np
import pandas as pd

setup_logger(level="INFO", json=False)

print("=" * 60)
print("🔍 10-HOUR HYPERPARAMETER GRID SEARCH")
print("=" * 60)
print(f"Start time: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
print("=" * 60)
print()

# Load data
features, labels = load_features_labels(timeframe="1h")
X, y, meta = build_ml_table(features, labels, target="direction_4")

print(f"Dataset: {len(X):,} samples, {len(X.columns)} features")
print()

# Define extensive parameter grid
param_grid = {
    'classifier__C': [0.001, 0.01, 0.1, 0.5, 1.0, 2.0, 5.0, 10.0, 50.0, 100.0],  # 10 values
    'classifier__class_weight': ['balanced', None],  # 2 values
    'classifier__max_iter': [5000, 10000, 20000, 50000],  # 4 values
    'classifier__solver': ['lbfgs', 'saga'],  # 2 values
}

# Total combinations: 10 * 2 * 4 * 2 = 160 models
# With 5-fold CV: 160 * 5 = 800 model fits!

print(f"Parameter grid:")
for param, values in param_grid.items():
    print(f"  {param}: {values}")
print()
print(f"Total combinations: {10 * 2 * 4 * 2} models")
print(f"With 5-fold CV: {10 * 2 * 4 * 2 * 5} fits")
print()

# Create pipeline
pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('classifier', LogisticRegression(random_state=42))
])

# Grid search with cross-validation
grid_search = GridSearchCV(
    estimator=pipeline,
    param_grid=param_grid,
    cv=5,  # 5-fold cross-validation
    scoring='roc_auc',
    n_jobs=-1,  # Use all CPU cores
    verbose=3,  # Show progress
    return_train_score=True,
)

print("🚀 Starting grid search...")
print("This will take approximately 10 hours...")
print()

start_time = datetime.now()

# Fit grid search
grid_search.fit(X.values, y.values)

end_time = datetime.now()
duration = (end_time - start_time).total_seconds() / 3600

print()
print("=" * 60)
print("✅ GRID SEARCH COMPLETED!")
print("=" * 60)
print(f"Duration: {duration:.2f} hours")
print(f"Best score (AUC): {grid_search.best_score_:.4f}")
print(f"Best parameters:")
for param, value in grid_search.best_params_.items():
    print(f"  {param}: {value}")
print()

# Save results
results_df = pd.DataFrame(grid_search.cv_results_)
results_path = settings.artifacts_dir / "gridsearch_results.csv"
results_df.to_csv(results_path, index=False)
print(f"Results saved to: {results_path}")

# Save best model
from algo_trader.models.sklearn_baseline import SklearnBaseline
from algo_trader.research.registry import save_run

best_params = grid_search.best_params_
model = SklearnBaseline(
    balanced=(best_params['classifier__class_weight'] == 'balanced'),
    C=best_params['classifier__C'],
    max_iter=best_params['classifier__max_iter'],
    random_state=42,
)
model.fit(X, y)

model_path = save_run(
    model=model,
    calibrator=None,
    metrics={'best_cv_auc': grid_search.best_score_},
    config={
        'timeframe': '1h',
        'target': 'direction_4',
        'search_type': 'gridsearch',
        'best_params': best_params,
    },
    feature_names=list(X.columns),
)

print(f"Best model saved to: {model_path}")
print("=" * 60)