import pandas as pd
import numpy as np
from statsmodels.tsa.arima.model import ARIMA
from sklearn.preprocessing import MinMaxScaler
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import LSTM, Dense

def arima_forecast(ts_data, order=(5,1,0), steps=5):
    """
    ts_data: list of historical stock prices
    steps: number of future steps to forecast
    """
    ts_series = pd.Series(ts_data)
    model = ARIMA(ts_series, order=order)
    model_fit = model.fit()
    forecast = model_fit.forecast(steps=steps)
    return forecast.tolist()

def lstm_forecast(ts_data, look_back=60, steps=5, epochs=20):
    """
    ts_data: list of historical stock prices
    steps: number of future steps to forecast
    Automatically adjusts look_back if input is shorter than look_back.
    """
    # Adjust look_back if input is too short
    if len(ts_data) < look_back + 1:
        look_back = max(1, len(ts_data) - 1)

    # Normalize
    scaler = MinMaxScaler(feature_range=(0, 1))
    scaled_data = scaler.fit_transform(np.array(ts_data).reshape(-1,1))

    # Create sequences
    def create_sequences(dataset, look_back):
        X, Y = [], []
        for i in range(len(dataset) - look_back):
            X.append(dataset[i:(i+look_back), 0])
            Y.append(dataset[i + look_back, 0])
        return np.array(X), np.array(Y)

    X, y = create_sequences(scaled_data, look_back)

    # Train/test split (use all for training in deployment)
    X = np.reshape(X, (X.shape[0], X.shape[1], 1))
    y = y

    # Build LSTM
    model = Sequential()
    model.add(LSTM(50, return_sequences=True, input_shape=(look_back,1)))
    model.add(LSTM(50))
    model.add(Dense(1))
    model.compile(optimizer='adam', loss='mean_squared_error')

    # Train
    model.fit(X, y, epochs=epochs, batch_size=32, verbose=0)

    # Forecast future steps
    last_seq = scaled_data[-look_back:].reshape(1, look_back, 1)
    predictions = []

    for _ in range(steps):
        pred = model.predict(last_seq, verbose=0)
        predictions.append(pred[0,0])
        # Fix: reshape pred to 3D and append to sequence
        pred_reshaped = pred.reshape(1, 1, 1)
        last_seq = np.concatenate([last_seq[:,1:,:], pred_reshaped], axis=1)

    predictions = scaler.inverse_transform(np.array(predictions).reshape(-1,1))
    return predictions.flatten().tolist()

def infer(model_type: str, input_data: list, steps: int = 5):
    """
    model_type: 'arima' or 'lstm'
    input_data: list of recent stock prices
    steps: number of future days to forecast
    """
    if model_type.lower() == 'arima':
        return arima_forecast(input_data, steps=steps)
    elif model_type.lower() == 'lstm':
        return lstm_forecast(input_data, steps=steps)
    else:
        return {"error": "Invalid model_type. Use 'arima' or 'lstm'."}