lstm_model.py

import numpy as np
import pandas as pd
import os
from tensorflow.keras.models import Sequential, load_model
from tensorflow.keras.layers import LSTM, Dense, Dropout
from tensorflow.keras.optimizers import Adam

MODEL_FILE = "lstm_model.keras"
SEQ_LEN = 30
FEATURES = ["close","RSI","EMA50","EMA200","ATR"]

def build_sequences(df):
    df = df.dropna()
    X, y = [], []
    for i in range(len(df) - SEQ_LEN - 1):
        seq = df[FEATURES].iloc[i:i+SEQ_LEN].values
        future = df["close"].iloc[i+SEQ_LEN]
        current = df["close"].iloc[i+SEQ_LEN-1]
        label = 1 if future > current else 0
        X.append(seq)
        y.append(label)
    return np.array(X), np.array(y)

def train_lstm(df):
    X, y = build_sequences(df)
    if len(X) < 200: return None

    model = Sequential([
        LSTM(64, return_sequences=True, input_shape=(X.shape[1], X.shape[2])),
        Dropout(0.2),
        LSTM(32),
        Dense(1, activation="sigmoid")
    ])

    model.compile(loss="binary_crossentropy", optimizer=Adam(learning_rate=0.001), metrics=["accuracy"])
    
    # Optimisation i7 : On double le temps d'apprentissage pour la précision
    model.fit(X, y, epochs=20, batch_size=32, verbose=0)
    model.save(MODEL_FILE)
    return model

def load_lstm():
    if os.path.exists(MODEL_FILE): return load_model(MODEL_FILE)
    return None

def predict_lstm(model, df):
    df = df.dropna()
    if len(df) < SEQ_LEN: return 0.5
    seq = df[FEATURES].iloc[-SEQ_LEN:].values
    seq = np.expand_dims(seq, axis=0)
    prob = model.predict(seq, verbose=0)[0][0]
    return float(prob)