Running MANY files through CSP/classifier training

Working with OpenViBE signal processing scenarios and doing scenario/BCI design
Post Reply
lfrajmu
Posts: 2
Joined: Mon Jun 04, 2018 5:31 pm

Running MANY files through CSP/classifier training

Post by lfrajmu »

Hey everyone,

I'm trying to find out how different electrode configurations change the classifier accuracy, and to do so I need to run 200 ov files through the CSP and the classifier quite a few times (once for each electrode configuration I have). Since doing this manually would take far too long, does anyone know of a way to automatically run files through the CSP and classifier scenarios? Ideally I would run them in batches of 200, changing the electrode configuration in between them, and the log would be recorded as usual (the accuracy is printed to the log, and I already have an extraction code set up to handle getting the necessary info). I have MatLab2015a installed, so any MatLab solutions would also work for me.

Thanks for the help.
lfrajmu

jtlindgren
Posts: 775
Joined: Tue Dec 04, 2012 3:53 pm
Location: INRIA Rennes, FRANCE

Re: Running MANY files through CSP/classifier training

Post by jtlindgren »

Hi,

its not super comfy, but what you can do is to have some external process launch designer multiple times with different configurations. For this you can use Matlab, Python, R or even a shell script if you're comfy with it.

Basically, you should loop over your electrode configurations with a block like this (pseudocode)

Code: Select all

trythese=1;2;4
call openvibe-designer.cmd --define electrodes $trythese$ --no-gui --play-fast scenario.xml
read the result somehow.
and then in the scenario.xml, configure the channel selector to use token ${electrodes} as the channels to pick.

ps. since you're likely doing science and want to compare results, make sure the estimate you're looking at is not optimistic. My recommendation is to use separate train and test sets. For some more detail, see http://openvibe.inria.fr/tutorial-how-t ... te-better/ .

Cheers,
Jussi

Post Reply