How to create a LSTM model for multivariate time dependent time series

I want to forecast a time series using two dependent time series. How can I do this task with Matlab Deep Learning Toolbox? I used the following architecture:

Sequence Input Layer(2) LSTM Layer(100) Fully Connected Layer(1) Regression Layer

I shifted my training output by one time step and standardized the data. When I run my code I get the following error:

Invalid training data. For cell array input, responses must be a an N-by-1 array of sequences, where N is the number of sequences. …

Answer 1

[Venu]

Hi @Omid,

You can refer to this example to train a multi-output LSTM network using a custom training loop:

https://www.mathworks.com/help/deeplearning/ug/train-network-with-multiple-outputs.html

This example shows multi-step ahead (closed loop) predictions:

https://www.mathworks.com/help/deeplearning/ug/time-series-forecasting-using-deep-learning.html

This MATLAB answer can be relevant regarding your error:

https://www.mathworks.com/support/search.html/answers/1632610-predictors-must-be-a-n-by-1-cell-array-of-sequences-where-n-is-the-number-of-sequences-all-sequenc.html?fq%5B%5D=asset_type_name:answer&fq%5B%5D=category:deeplearning/image-data-workflows&page=1

Hope this helps!