Multivariable time series prediction using LSTM -- problem summary

Rigid contact LSTM, Reference resources Use LSTM Carry out multivariable time series prediction This article , Write very well , But there are many problems when running , May be tensorflow Version issue for , There are various problems when running , I changed it for a while this morning , Finally, I got through , There are also many bloggers with similar problems under this blog , Let's record it here , I hope it helps .

My package version information

python3.7
keras2.3.1
tensorflow2.2.0

ps: The original title is also used tensorflow+keras, although tensorflow2.X When to bring it keras, But I did it again keras!

problem 1

grid_search.fit(trainX,trainY) Report errors ,Cannot clone object <tensorflow.python.keras.wrappers.scikit_learn.KerasRegressor object at 0x7fc38c260f70>, as the constructor either does not set or modifies parameter validation_data

This problem , hold grid_model = KerasRegresso r(build_fn=build_model, verbose=1, validation_data=(testX,testY))
there validation_data=(testX,testY) Move to below grid_search.fit(trainX,trainY) in , The changed code is :

def build_model(optimizer):
    grid_model = Sequential()
    grid_model.add(LSTM(50,return_sequences=True,input_shape=(30,5)))
    grid_model.add(LSTM(50))
    grid_model.add(Dropout(0.2))
    grid_model.add(Dense(1))

    grid_model.compile(loss = 'mse',optimizer = optimizer)
    return grid_model

grid_model = KerasRegressor(build_fn=build_model,verbose=1)
parameters = {
    'batch_size' : [16,20],
              'epochs' : [8,10],
              'optimizer' : ['adam','Adadelta'] }

grid_search  = GridSearchCV(estimator = grid_model,
                            param_grid = parameters,
                            cv = 2)
grid_search = grid_search.fit(trainX,trainY,validation_data=(testX,testY))

There should be no big problems after the change here ,

problem 1.5

This is not a problem , Just look at the comment area and say that the program keeps running , It won't stop , This is actually looking for the optimal parameters in a loop , About a dozen rounds stopped

The following possible problems ,( Additional problems I encountered

problem 2

Cannot convert a symbolic Tensor (lstm/strided_slice:0) to a numpy array.

This is numpy The version does not correspond to the problem , Use 1.19.2 Just the version

pip install -U numpy==1.19.2

problem 3

TypeError: Descriptors cannot not be created directly
Downgrade protobuf that will do

pip install 'protobuf~=3.19.0'

It will probably be solved here , If you have other questions, welcome to the message area to discuss .