Intelligence Artificielle BrainAmu

# Sleep Stage CNN/LSTM
import tensorflow as tf

model = tf.keras.Sequential([
    tf.keras.layers.Conv1D(32, kernel_size=3, activation='relu', input_shape=(256, 4)),  # EEG, 4 channels
    tf.keras.layers.MaxPooling1D(pool_size=2),
    tf.keras.layers.LSTM(64, return_sequences=True),
    tf.keras.layers.LSTM(64),
    tf.keras.layers.Dense(4, activation='softmax')  # 4 stages
])

# Quantize for ESP32
converter = tf.lite.TFLiteConverter.from_keras_model(model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
tflite_model = converter.convert()

# Fragmentation Prediction (Multi-Input)
model = tf.keras.Sequential([
    tf.keras.layers.Dense(128, activation='relu', input_shape=(64,)),  # EEG + temp features
    tf.keras.layers.Dense(1, activation='sigmoid')  # Probability
])

# Temperature Regulation (Decision Tree)
# ESP32 Arduino code
float eeg_features[64], temp_data[1];
if (predict_fragmentation(eeg_features, temp_data) > 0.5) {
    if (temp_data[0] > 37.0) cool_peltier(-0.5);  // Cool if >37°C
    else if (temp_data[0] < 36.0) heat_peltier(0.5);  // Heat if <36°C
}