Using trainNetwork on a remote headless machine

I am trying to use the deep learning toolbox to train a CNN network. Matlab is installed on a remote machine and I am accessing it via jupyter using the matlab supplied python bindings (Installed from /extern/engines/python). Following along [1]. I can succesfully train however the problem is calling trainNetwork works but since there is no gui when using jupyter or when running repl through ssh there is no way to see the progress. Is it possible to instead of a GUI the output can be printed to the repl so that jupyter can pick up the output and show the progress.

[1] https://www.mathworks.com/help/deeplearning/examples/create-simple-deep-learning-network-for-classification.html

Answer 1

[Steven Lord]

I haven't tried this but I think you want to use trainingOptions to tell MATLAB not to create the training progress Plots. If you want to see iterative output either set the Verbose and VerboseFrequency options (if the information MATLAB provides by default is sufficient) and/or specify an OutputFcn. The section about OutputFcn on that documentation page lists the information trainNetwork passes into the OutputFcn.

Remember you'll need to pass the options object trainingOptions returns into trainNetwork. I say this because in similar situations in the past I've seen people assume the options are set globally and express confusion when MATLAB "ignored" their options.

[Hamza Yerlikaya]

Thanks. That solved the problem.