speeding up trainbr using more cpu cores or a gpu

trainbr is slow on my computer often taking an hour. I'm looking at ways to speed it up. I'm trying to figure out if the Parallel Toolbox would help. It seems to say it would allow me to use multiple cpu cores, but when I'm running trainbr training it already looks like it is using multiple cores. Anybody know about this?

I have a GPU available but I recall reading trainbr can't use a GPU, anybody know about this?

Thank you!

Answer 1

[Aditya Patil]

Bayesian regularization requires calculating the Jacobian. Jacobian calculation is not supported on GPU. Hence trainbr doesn't support GPU.

[Brent McWatters]

I understand re the GPU.

However when I run trainbr I see it using all the CPU cores. So does that mean the parallel toolbox wouldn't help?

[Aditya Patil]

parallel toolbox uses multiple processes. Even if trainbr uses multiple CPU, parallel toolbox might be useful, especially if your code waits for an external event, like reading a file from disk.

You can try to time the code with and without parallel toolbox, and see if there is any significant improvement.