Does custom mini-batch datastore support Background Dispatch

I foloow This MATLAB tutorial to create my custom mini-batch datastore, and I want to accelerate the training process.

My question is Does custom mini-batch datastore support Background Dispatch?

Answer 1

[Venu]

Hi @Xie Shipley,

To enable background dispatch for your custom mini-batch datastore, you must ensure that it adheres to the requirements of the "Background Dispatch" interface. Your custom datastore should inherit from "matlab.io.datastore.BackgroundDispatchable" and implement the method to start reading background data. The datastore should support being divided into smaller parts that can be read independently, which is required for background dispatch.

For built-in training functions like "trainNetwork", set the "DispatchInBackground" option to "true" in the "trainingOptions" function. For custom training loops, use a "minibatchqueue" with your custom datastore and set the "DispatchInBackground" property to "true".

You must have MATLAB's Parallel Computing Toolbox because background dispatch relies on parallel workers.

Find this documentation and MATLAB Answer below for your reference:

https://in.mathworks.com/help/deeplearning/ug/datastores-for-deep-learning.html#mw_4e35721b-f50a-412a-af91-c35e606f52cc

https://in.mathworks.com/matlabcentral/answers/1884387-dispatchinbackgrounddatastore