Parallel Computing to Train NARX

Hi,

I am using Matlab 2015a with license for Neural Networks and Parallel Computing toolboxs.

I have been using NARX for a few years. I want to use parallel computing to training my NARX. But, when I run the train function with "useParallel", "yes", there is only 1 worker on MEX WIN64. All other 11 workers are "not used". I have a server with 12 cores (2xCPUs).

How I can put all the workers to the MEX to have parallel training with all CPU cores for NARX?

Really looking forward to your answer! Thanks. Baker Huang (Yu)

[Yu Huang]

I think this is due to very large data set that I use to train and simulate the NARX. Could you please help me on how to divide the data sets into Composites? I tried, failed. Could you please give me some examples?

Answer 1

[Diogo Goncalves]

I think you should check the parallel pool set in your matlab. If you have it default you should change manually like: http://www.mathworks.com/help/distcomp/parpool.html