How to import non-image data in Deep Network Designer

How do I import non-image, csv data to train LSTM or any deep neural network model in Deep Network Designer?

Answer 1

[Garmit Pant]

Hello Parashhyoti

My understanding of the problem is that you are trying to train an LSTM or any other non-image related deep neural network using the Deep Network Designer and thus want to import data for the same.

In Deep Network Designer, you can import data in the form of Image Data or a Datastore from the workspace as shown below.

To import non-image data, you'd first have to create a datastore object that wroks with the trainNetwork function (eg: TransformedDatastore, CombinedDatastore, Custom mini-batch datastore.)

You can consult the following resource for creating your custom mini-batch datastore:

lhttps://www.mathworks.com/help/deeplearning/ug/develop-custom-mini-batch-datastore.html

Alternatively, you can export your network using the Export option in the Designer tab and use the 'trainNetwork' method to train the network using training data in the form cell array of numeric arrays or numeric array created after reading the CSV file.

You can consult this tutorial as well: https://www.mathworks.com/help/deeplearning/gs/create-simple-sequence-classification-network.html

[PARASHJYOTI BORAH]

Thank you for your comment.

[David Ho]

To complement this great answer from @Garmit Pant, for more information on importing non-image data into Deep Netework Designer you can refer to the following documentation page:

https://uk.mathworks.com/help/deeplearning/ug/import-data-into-deep-network-designer.html

For sequences, as Garmit has said, you can either load the data into the MATLAB Workspace (if it fits in memory) and create a combinedDatastore, or use a custom datastore.