bci2000.org BBS

Hello,

I am unclear of the intended order of operations for the written parameter file that the P300Classifier outputs. From some train and test data I have what appears to be a potentially good classifier, and have output the .prm file. I'm a little unsure how this is intended to be used. My thought is that if I wanted to free spell, I would load the .prm file I was using for copy spelling and then load the .prm output by the P300Classifier, change the application settings for free spelling, and wallah, or maybe vice verse. This did not seem to give good results when free spelling.

Is this the right approach?

Am I supposed to do an "online classification" of some sort instead at this point, I think I've seen that term used a few times.

Perhaps I didn't create a robust enough weights matrix, what would be a recommended training data to testing data ratio?

I realize the quality of the data in the data files is important here, but I feel that my data was likely both consistent and relatively clean.

Thanks!

Ngubser,

to get familiar with the P300 speller, please see the documentation on the P300Classifier [1] and the tutorial from one of our workshops [2]. Please note that the P300GUI has since been retired. To ensure a good classification result, please record at least 15 characters with 15 repetitions each in copy spelling. You see form the tutorial, that you will need a electrode montage that can capture the occipital response. Analyze the dataset with the OfflineAnalysis tool within BCI2000. This should show a nice physiological pattern, just as in the tutorial. Next you calculate a good classifier using the P300Classifier tool. Verify this then in an copy spelling with showing the results session. Then reduce the number of sequences to a comfortable level for free spelling.

Regards, Peter

[1] http://www.bci2000.org/wiki/index.php/U ... Classifier
[2] https://db.tt/DDqCS5SD

Sorry for the delayed reply.

I was able to successfully set the configuration and am now onto my next endeavors and inevitably obstacles.

What I was unclear on was that I needed to first load my .par file which I had been using for copy spelling, and then load the .par file that was output from the P300Classifier. I'm learning that certain .par files seem to only update/overwrite some configuration parameters. I guess then, that means there are various instances where multiple .par files need to be loaded and perhaps even in a certain orders? Nonetheless, I was able to get this online, and like I said am happy to be moving forward.

Thanks again for your time and support!

Ngubser,

the idea of the parameter files is that they can be fragments, i.e., only contain a subset of the configuration. If you open a parameter file in a text editor you will see this more clearly. In your case that means, that you load the parameter fragment with the classifier at last, overwriting the standard settings for the classifier with the subject specific settings.

Regards, Peter