python - 集成模型不采用与其子模型相同的输入

Question

为了减少方差，我预测在训练/测试期间使用 3 个模型

test_reshaped = test_reshaped.reshape(len(test_reshaped), 1, 1)

forecast1 = model1.predict(test_reshaped, batch_size=batch_size)
forecast2 = model2.predict(test_reshaped, batch_size=batch_size)
forecast3 = model3.predict(test_reshaped, batch_size=batch_size)

model1.save('lstm_model1.h5')
model2.save('lstm_model2.h5')
model3.save('lstm_model3.h5')

然后，为了将来的使用，我使用以下函数创建了一个平均 3 的集成模型：

models = list()
nb_models = 3

for i in range(nb_models):
    model_tmp = load_model("lstm_model"+str(i+1)+".h5")
    model_tmp.name = "model_"+str(i+1)
    models.append(model_tmp)

def create_ensemble(models,model_input):

    # take-in all outputs fro all models
    outModels = [model(model_input) for model in models]

    # calculate average of all results
    outAvg = layers.average(outModels)

    # merge into one model

    modelMerge = Model(inputs=model_input,outputs=outAvg,name='ensemble')

    return modelMerge

model_input = Input(shape=models[0].input_shape[1:])
modelEns = create_ensemble(models,model_input)
modelEns.summary()

modelEns.save('ensemble_model.h5')

当我加载集成模型并将其用于具有相同形状的相同数据时：

test_reshaped = test_reshaped.reshape(len(test_reshaped), 1, 1)
forecast1 = model.predict(test_reshaped,batch_size = batch_size)

我收到错误消息：您必须使用 dtype float 和 shape [1,1,1] [[{{node lstm_2_input}}]] 为占位符张量“lstm_2_input”提供一个值

生成 3 个子模型的函数：

# model
def fit_lstm(train, batch_size, nb_epoch, nb_neurons):
    X, y = train[:, 0:-1], train[:, -1]
    X = X.reshape(X.shape[0], 1, X.shape[1])
    model = Sequential()
    model.add(LSTM(nb_neurons, batch_input_shape=(batch_size, X.shape[1], X.shape[2]), stateful=True))
    model.add(Dense(1))
    model.compile(loss='mean_squared_error', optimizer='adam')
    for i in range(nb_epoch):
        model.fit(X, y, epochs=1, batch_size=batch_size, verbose=0, shuffle=False)
        model.reset_states()
    return model

3款车型总结：

Layer (type)                 Output Shape              Param #   
=================================================================
lstm_1 (LSTM)                (1, 1)                    12        
_________________________________________________________________
dense_1 (Dense)              (1, 1)                    2         
=================================================================

集成模型总结：

_________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input_1 (InputLayer)            (None, 1, 1)         0                                            
__________________________________________________________________________________________________
model_1 (Sequential)            multiple             14          input_1[0][0]                    
__________________________________________________________________________________________________
model_2 (Sequential)            multiple             14          input_1[0][0]                    
__________________________________________________________________________________________________
model_3 (Sequential)            multiple             14          input_1[0][0]                    
__________________________________________________________________________________________________
average_1 (Average)             (None, 1)            0           model_1[1][0]                    
                                                                 model_2[1][0]                    
                                                                 model_3[1][0]                    
==================================================================================================

标签： pythontensorflowkeraslstmensemble-learning