如何在python中的kers中添加LSTM中的漏失和注意

import numpy as np
from keras.models import Model
from keras.layers import Input, LSTM, Dense, Dropout, Masking, TimeDistributed
from keras.layers import Bidirectional, Concatenate, Permute, Dot, Multiply
from keras.callbacks import EarlyStopping

def build_lstm_model(input_shape, output_shape, hidden_units, dropout_rate):
    X_input = Input(input_shape)

    # LSTM层
    lstm = LSTM(hidden_units, return_sequences=True)(X_input)
    dropout = Dropout(dropout_rate)(lstm)

    # 漏失层
    attention = TimeDistributed(Dense(1, activation='tanh'))(dropout)
    attention = Flatten()(attention)
    attention = Activation('softmax')(attention)
    attention = RepeatVector(hidden_units)(attention)
    attention = Permute([2, 1])(attention)

    # 注意力机制
    attention_mul = Multiply()([dropout, attention])
    attention_mul = Lambda(lambda x: K.sum(x, axis=1))(attention_mul)

    # 全连接层
    output = Dense(output_shape, activation='softmax')(attention_mul)

    model = Model(inputs=X_input, outputs=output)

    return model

# 定义输入和输出的形状
input_shape = (input_sequence_length, input_dimension)
output_shape = output_dimension

# 定义超参数
hidden_units = 64
dropout_rate = 0.2

# 构建模型
model = build_lstm_model(input_shape, output_shape, hidden_units, dropout_rate)

# 编译模型
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

# 定义早停策略
early_stopping = EarlyStopping(monitor='val_loss', patience=3)

# 训练模型
model.fit(X_train, Y_train, validation_data=(X_val, Y_val), epochs=10, batch_size=32, callbacks=[early_stopping])