Batch learning for deep learning lstm time series

Is it possible to train a LSTM network by retrain/update the LSTM network by offering different data sets in different batches?

To come one trained LSTM network, trained by using different data sets.

Many thanks for your support

Answer 1

[Priyank Sharma]

We train the network using mini-batches of data (if you want to change the size of the mini-batches they can do that in trainingOptions).

If this is what you are following and what you are asking is if you can retrain on multiple datasets, then the answer is yes. You can follow the same workflow for transfer learning that is mentioned in the documentation. Take the trained network, get the layerGraph(net), modify layer graph if needed, and then retrain using trainNetwork and the new dataset.

Please follow the below links for more reference:

https://www.mathworks.com/help/deeplearning/examples/train-deep-learning-network-to-classify-new-images.html

https://www.mathworks.com/help/deeplearning/examples/transfer-learning-using-alexnet.html

Hope this helps!

[Leon]

Thx for your answer Priyank