Python 使用LSTM进行交叉验证的正确方法是什么？

我对LSTM和深度学习还比较陌生。我正在尝试训练多元LSTM时间序列预测，我想进行交叉验证。我尝试了两种不同的方法，发现了非常不同的结果

• 使用

  kfold.split

• 使用

  KerasRegressor

  和

  cross\u val\u分数

第一个选项的结果更好，RMSE约为3.5，而第二个代码的RMSE为5.7（反向归一化后）。我试图搜索使用KerasRegressionor包装器的LSTM示例，但没有找到很多，而且它们似乎没有遇到相同的问题（或者可能没有检查）。我想知道Keras回归者是不是搞乱了模型。或者如果我做错了什么，因为原则上这两个代码应该给出相同的结果？如果你有什么想法，请告诉我

    seed(1)
    tf.random.set_seed(2)

    kfold = KFold(n_splits=8, shuffle=False)
    epochs = 20
    batch_size=100

    model = Sequential()
    model.add(LSTM(200, activation="relu",
                   input_shape=(X_train.shape[1], X_train.shape[2])))
    model.add(Dense(1,activation="sigmoid"))

    #compile model
    model.compile(optimizer='adam', loss='mse', metrics=rmse)
    
    for trn, val in kfold.split(X_train, Y_train):     
        model.fit(X_train[trn], Y_train[trn], epochs=epochs,
                            batch_size=batch_size, verbose=1)
        
        # Generate generalization metrics
        scores = model.evaluate(X_train[val], Y_train[val], verbose=1)     

    #make prediction
    history = model.fit(X_train, Y_train, epochs=epochs, batch_size=batch_size,verbose=1)
    Y_pred = model.predict(X_test)

epochs = 20
batch_size=100

#model funtion
def lstm_model(unit=200):
    seed(1)
    tf.random.set_seed(2)
    
    #define model
    model = Sequential()
    model.add(LSTM(unit, activation="relu",input_shape=(X_train.shape[1], X_train.shape[2])))
    model.add(Dense(1,activation="sigmoid"))
    
    # compile model
    model.compile(optimizer="adam", loss="mse", metrics=rmse)
    
    model.summary()
    return model

model = KerasRegressor(build_fn=lstm_model, epochs=epochs, batch_size=batch_size, verbose=1)

kfold = KFold(n_splits=8, shuffle=False)
cv_scores = cross_val_score(model, X_train, Y_train, cv=kfold, scoring='neg_root_mean_squared_error')
    
#fit the model
model.fit(X_train, Y_train)
    
#test on test set
Y_pred = model.predict(X_test)