Datastore is not PartitionableByIndex

Hello I am trying to train a CNN using Matlab 2023a.

I create a signal data store using the following code. The function @readData returns 2 variables. Variable 1 is 1xN time-series data, and Variable 2 is a categorical variable and is T1 or T2.

CODE for creating data store

for j = [1:i-i i+1:length(pts)] % For parsing patient folder exluding one patient

k = k + 1;

folder_train{1,2*k-1} = fullfile(data_folder,'T1',sprintf('T%.3d',pts(j))); %Rd data corresponding to Task 1

folder_train{1,2*k} = fullfile(data_folder,'T2',sprintf('T%.3d',pts(j))); %Rd data corresponding to Task 2

end

% CREATE data store

eds_train = signalDatastore(folder_train,'IncludeSubfolders',true,'FileExtensions','.mat','ReadFcn',@readData);

When I train the algorithm using parallel environment option it returns me an error

"Input datastore is not PartitionableByIndex and does not support parallel operations"

I tested whether my datastore is partiationable using isPartitionable and it returned 1. I am unsure what the issue is at this point. Can you please help me?

Thanks

Answer 1

[Divyanshu]

Hi N/A,

I understand that you are creating a data store for training a CNN model using MATLAB. You can go through the following pointers:

• Though the function “isPartionable” outputs 1 for the datastore, a possible reason for the error is the way datastore is created.
• As you have specified a custom Read function “readData” as an argument to “signalDataStore” function it is possible that the read function does not support parallel processing.
• Ensure that the “readData” function is implemented correctly to support reading data in chunks & subsets.

Refer the following documentation for examples & better understanding:

https://www.mathworks.com/help/signal/ref/signaldatastore.html#mw_1158dbc4-7c7f-47c6-8a54-8b24b23b69df