"""
EUR/USD Direction Prediction — Inference
=========================================
Usage:
    python predict.py
    
Requires: yfinance, lightgbm, xgboost, ta, scikit-learn, joblib, numpy, pandas
"""
import json
import numpy as np
import pandas as pd
import yfinance as yf
import ta
import joblib

# Optimal ensemble weights (grid-optimised over 198 walk-forward folds)
LGB_WEIGHT = 0.39
XGB_WEIGHT = 0.61


def compute_features(df):
    """Compute all features from OHLCV data."""
    LOOKBACK_WINDOWS = [5, 10, 21, 63, 126, 252]
    
    for w in LOOKBACK_WINDOWS:
        df[f"log_return_{w}d"] = np.log(df["Close"] / df["Close"].shift(w))
        df[f"momentum_{w}d"] = df["Close"] / df["Close"].shift(w) - 1
    
    df["log_return_1d"] = np.log(df["Close"] / df["Close"].shift(1))
    
    for w in [5, 10, 21, 63]:
        df[f"volatility_{w}d"] = df["log_return_1d"].rolling(w).std()
    
    for w in [5, 10, 20, 50, 100, 200]:
        sma = df["Close"].rolling(w).mean()
        df[f"price_to_sma_{w}"] = df["Close"] / sma - 1
    
    df["ema_12"] = df["Close"].ewm(span=12).mean()
    df["ema_26"] = df["Close"].ewm(span=26).mean()
    df["ema_cross"] = df["ema_12"] / df["ema_26"] - 1
    
    for w in [7, 14, 21]:
        df[f"rsi_{w}"] = ta.momentum.RSIIndicator(df["Close"], window=w).rsi()
    
    macd = ta.trend.MACD(df["Close"])
    df["macd"] = macd.macd()
    df["macd_signal"] = macd.macd_signal()
    df["macd_diff"] = macd.macd_diff()
    
    bb = ta.volatility.BollingerBands(df["Close"], window=20, window_dev=2)
    df["bb_pband"] = bb.bollinger_pband()
    df["bb_wband"] = bb.bollinger_wband()
    
    atr = ta.volatility.AverageTrueRange(df["High"], df["Low"], df["Close"], window=14)
    df["atr_14"] = atr.average_true_range()
    df["atr_14_pct"] = df["atr_14"] / df["Close"]
    
    stoch = ta.momentum.StochasticOscillator(df["High"], df["Low"], df["Close"])
    df["stoch_k"] = stoch.stoch()
    df["stoch_d"] = stoch.stoch_signal()
    
    adx = ta.trend.ADXIndicator(df["High"], df["Low"], df["Close"], window=14)
    df["adx"] = adx.adx()
    df["adx_pos"] = adx.adx_pos()
    df["adx_neg"] = adx.adx_neg()
    
    df["williams_r"] = ta.momentum.WilliamsRIndicator(df["High"], df["Low"], df["Close"]).williams_r()
    df["cci"] = ta.trend.CCIIndicator(df["High"], df["Low"], df["Close"]).cci()
    
    df["obv"] = ta.volume.OnBalanceVolumeIndicator(df["Close"], df["Volume"]).on_balance_volume()
    df["obv_pct"] = df["obv"].pct_change(5)
    
    df["day_of_week"] = df.index.dayofweek
    df["month"] = df.index.month
    df["quarter"] = df.index.quarter
    
    df["hl_range"] = (df["High"] - df["Low"]) / df["Close"]
    df["oc_range"] = (df["Close"] - df["Open"]) / df["Close"]
    
    for w in [10, 20, 50]:
        rolling_high = df["High"].rolling(w).max()
        rolling_low = df["Low"].rolling(w).min()
        df[f"channel_pos_{w}"] = (df["Close"] - rolling_low) / (rolling_high - rolling_low + 1e-10)
    
    return df


def predict():
    # Load models
    lgb_model = joblib.load("lgb_model.joblib")
    xgb_model = joblib.load("xgb_model.joblib")
    scaler = joblib.load("scaler.joblib")
    feature_cols = json.load(open("feature_columns.json"))
    
    # Fetch recent data (need 300 days for indicators to warm up)
    df = yf.download("EURUSD=X", period="2y", progress=False)
    if isinstance(df.columns, pd.MultiIndex):
        df.columns = df.columns.get_level_values(0)
    df["Volume"] = df["Volume"].replace(0, 1)
    
    df = compute_features(df)
    df = df.dropna(subset=feature_cols)
    
    # Predict latest
    latest = df.iloc[[-1]]
    X_latest = scaler.transform(latest[feature_cols].values)
    
    lgb_prob = lgb_model.predict_proba(X_latest)[:, 1][0]
    xgb_prob = xgb_model.predict_proba(X_latest)[:, 1][0]
    ensemble_prob = LGB_WEIGHT * lgb_prob + XGB_WEIGHT * xgb_prob
    
    direction = "UP" if ensemble_prob >= 0.5 else "DOWN"
    confidence = max(ensemble_prob, 1 - ensemble_prob)
    
    print(f"Date: {latest.index[0].date()}")
    print(f"EUR/USD Close: {latest['Close'].values[0]:.5f}")
    print(f"Prediction for next day: {direction}")
    print(f"Confidence: {confidence:.1%}")
    print(f"  LightGBM prob(UP): {lgb_prob:.3f} (weight: {LGB_WEIGHT:.0%})")
    print(f"  XGBoost prob(UP):  {xgb_prob:.3f} (weight: {XGB_WEIGHT:.0%})")
    print(f"  Ensemble prob(UP): {ensemble_prob:.3f}")

if __name__ == "__main__":
    predict()