bci2000.org BBS

We're having some issues with parameters working accurately; we're investigating the idea of using Subject B's parameter during a session with Subject A.

If we load the default, which includes all of our settings for our experiment, and then a Subject A's parameter created in the P300Classifier, run a few trials yielding high accuracy, and then load Subject B's parameter, the accuracy decreases substantially.
However, when we load our default parameter and then immediately afterwards load a Subject B-specific parameter, the subject parameter yields accurate spelling results for Subject A.

I've tried looking at the Classifier Matrix in the Config window for Subject B's parameter in both situations, and it doesn't seem to be any different. The only issue I can think of is that the parameter isn't completely overwriting the last one, but since the matrix doesn't appear different in the two situations, I'm at a loss.

Please help!

I've tried looking at the Classifier Matrix in the Config window for Subject B's parameter in both situations, and it doesn't seem to be any different. The only issue I can think of is that the parameter isn't completely overwriting the last one, but since the matrix doesn't appear different in the two situations, I'm at a loss.

The two classifier matrices should be different when P3Classifier was used with different training files. When there is no difference between the two matrices, something else could have caused the drop in performance of subject A. E.g., you might have moved an electrode while loading subject B's parameter file.

Generally, spelling accuracy may vary independently of parameters, e.g. due to fluctuations in electrode impedance, alertness of the subject, ... Without doing statistics over a number of runs, you cannot draw any conclusions about how well subject A is able to spell with parameters from subject B.

Regards,
Juergen