How to prefetch data during network training ?

Is there any method to prefetch data during network training In MATLAB?

Answer 1

[Venu]

Hi @Xie Shipley,

Yes you can prefetch your data while training the network in MATLAB if you have a GPU. While the GPU is working on the current mini-batch of data, the CPU can simultaneously prepare the next mini-batch, which helps reduce the time the GPU spends waiting for data.

To implement data prefetching in MATLAB, you can use the "backgroundDispatchEnable" property of the "miniBatchQueue" object. This property enables the loading and processing of data in the background while the GPU is training on the current batch.

You can refer to these documentations:

https://www.mathworks.com/help/deeplearning/ref/minibatchqueue.html

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

hope this helps

[Xie Shipley]

Yes, i have GPU.

I use `trainNetwork` with custom-mini-batch-datastore, and enbale `DispatchInBackground` in trainingOptions. But the Parallel Pool will not start automatically, Are there any steps I'm missing ?@Venu

[Venu]

Hi @Xie Shipley,

Check with this MATLAB Answer and discussion.

https://www.mathworks.com/matlabcentral/answers/318950-fail-to-start-parpool

https://www.mathworks.com/matlabcentral/answers/92124-why-am-i-unable-to-use-parpool-or-validate-with-the-local-or-processes-profile-of-parallel-compu

Try checking Parallel Pooling preferences in Home Tab, "Automatically create a parallel pool" option should be checked.

You can also start a parallel pool manually before calling train network by using "parpool" function.