checkpoint/python_version/model.py

"""
ML-3m-trader Model
===================
LightGBM multi-class classifier for trade signal prediction.
Handles training, validation, prediction, and persistence.
"""

import os
import sys

import numpy as np
import pandas as pd
import joblib

import config as cfg
from features import get_feature_columns

try:
    import lightgbm as lgb
except ImportError:
    print("[ERROR] lightgbm not installed. Run: pip install lightgbm")
    sys.exit(1)


def _ensure_dirs():
    os.makedirs(cfg.MODEL_DIR, exist_ok=True)


def train(df: pd.DataFrame, labels: np.ndarray) -> lgb.LGBMClassifier:
    """
    Train a LightGBM classifier on the provided feature DataFrame.

    Uses chronological train/validation split (no shuffle).

    Parameters
    ----------
    df : pd.DataFrame
        DataFrame with feature columns present.
    labels : np.ndarray
        Integer labels aligned with df index.

    Returns
    -------
    lgb.LGBMClassifier
        Trained model.
    """
    feature_cols = get_feature_columns()
    X = df[feature_cols].values.astype(np.float32)
    y = labels.astype(np.int32)

    split_idx = int(len(X) * cfg.TRAIN_SPLIT_RATIO)
    X_train, X_val = X[:split_idx], X[split_idx:]
    y_train, y_val = y[:split_idx], y[split_idx:]

    print(f"[INFO] Training set : {X_train.shape[0]:,} samples")
    print(f"[INFO] Validation set: {X_val.shape[0]:,} samples")

    params = dict(cfg.LGBM_PARAMS)
    n_est = params.pop("n_estimators", 500)

    model = lgb.LGBMClassifier(n_estimators=n_est, **params)

    model.fit(
        X_train, y_train,
        eval_set=[(X_val, y_val)],
        eval_metric="multi_logloss",
        callbacks=[
            lgb.early_stopping(cfg.EARLY_STOPPING_ROUNDS, verbose=True),
            lgb.log_evaluation(period=50),
        ],
    )

    # Feature importance
    imp = pd.DataFrame({
        "feature": feature_cols,
        "importance": model.feature_importances_,
    }).sort_values("importance", ascending=False)
    print("\n[INFO] Feature importance (top 10):")
    print(imp.head(10).to_string(index=False))

    return model


def predict(model: lgb.LGBMClassifier, df: pd.DataFrame) -> np.ndarray:
    """
    Generate predictions for the given DataFrame.

    Returns
    -------
    np.ndarray of int
        Predicted labels.
    """
    feature_cols = get_feature_columns()
    X = df[feature_cols].values.astype(np.float32)
    return model.predict(X)


def predict_proba(model: lgb.LGBMClassifier, df: pd.DataFrame) -> np.ndarray:
    """Return class probabilities."""
    feature_cols = get_feature_columns()
    X = df[feature_cols].values.astype(np.float32)
    return model.predict_proba(X)


def save_model(model: lgb.LGBMClassifier, filename: str = "lgbm_model.pkl"):
    """Persist model to disk via joblib."""
    _ensure_dirs()
    path = os.path.join(cfg.MODEL_DIR, filename)
    joblib.dump(model, path)
    print(f"[INFO] Model saved to {path}")


def load_model(filename: str = "lgbm_model.pkl") -> lgb.LGBMClassifier:
    """Load a previously saved model."""
    path = os.path.join(cfg.MODEL_DIR, filename)
    if not os.path.exists(path):
        print(f"[ERROR] Model file not found: {path}")
        print("        Run 'python main.py train' first.")
        sys.exit(1)
    model = joblib.load(path)
    print(f"[INFO] Model loaded from {path}")
    return model