Type of datastore for deep learning

Hi everyone, I'm trying to step inside the world of neural networks.

I would like to build a neural network (using Deep Network Designer) and it should have n>1 input values and m>1 output values.

I am really struggling to build a datastore or whatever that can be accepted as input dataset for training and testing.

I just create a csv from Excel, import it in Matlab and build the datastore, but when I try to start the training process of the network I always get the same problem: it always recognize my datastore as a 1x1 variable, even if it was built from a (n+m) x n_observations matrix.

Any suggestion?

[Mahesh Taparia]

Hi

Can you share the sample code which you are using?

Answer 1

[Jayanti]

Hi RobertC,

It seems like the issue you're facing is due to how the datastore is being created and imported into Deep Network Designer. Assuming your "input" and "target" is in matrix format you can create “arraydatastore” as shown below:

inputDS = arrayDatastore(inputs);  
targetDS = arrayDatastore(targets);  
ds = combine(inputDS, targetDS);

"arrayDatastore(inputs)" creates a datastore for the input data assuming inputs is a matrix. Similarly create a datastore for the target data. “combine” function combines these two individual datastores into a single datastore.After creating and combining the datastores, import the combined data into "Deep Network Designer".

You can also refer to below offical Mathworks documentation link for more information:

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