train a deep learning model (resnet-50 network) on a remote HPC cluster

I am trying to run a code, which uses a pre-trained ResNet-50 network, on a remote HPC cluster by submitting batch GPU jobs. I get the following error at this line:

net = resnet50

Error using resnet50

resnet50 requires the Deep Learning Toolbox Model for ResNet-50 Network support

package for the pretrained weights. To install this support package, use the <a

href="matlab:

matlab.addons.supportpackage.internal.explorer.showSupportPackages('RESNET50',

'tripwire')">Add-On Explorer</a>. To obtain the untrained layers, use

resnet50('Weights','none'), which does not require the support package.

It seems the Deep Learning Toolbox Model for ResNet-50 Network add-on is not installed on the cluster. How can I install this add-on on it?

Thanks

Answer 1

[David Willingham]

Just to confirm, you're sending batch jobs to a HPC cluster that has MATLAB parallel server installed?

If so, one option to try would be:

1. save resnet50 as as MAT file
2. attach the MAT file when submitting the job
3. have a load MAT file command in the function you're submitting.

[EK_47]

Brilliant! Thank you for your answer. It solved my problem.

Yes, the HPC cluster has MATLAB paraller server installed.

In your point 1, you said "save resnet50 as a MAT file". I was not sure what you mean by "save resnet50". What I did was just I called it in MATLAB on my local machine

basenet = resnet50;

then saved it as

save('basenet.mat','basenet');

and then transferred this MAT file into the remote cluster and loaded it there.

Thanks