As the following questions, LSTM do time series prediction, the training is according to the time before and after validation set differentiate? Multiple sequence multivariable step to do then, each sequence modeling training? If the data is too little can't use LSTM