Page 1 of 1

Predicted Character

Posted: 24 Feb 2004, 01:33
by M_Wentz
In looking at the matrix data sets, when you open the log file of each person, it gives predicted characters in a number of runs. The number of runs listed in this log file do not match up with the number of runs that exist for that person in the file folder. Why is that? Also, is the "predicted character" the desired letter that is known before the run begins or is the "predicted character" the letter the online program determined the person was thinking of after the run occurred?

Predicted character: SUSPECTED BUG

Posted: 28 Feb 2004, 11:31
by Brendan

I think you have hit on a bug that we found here as well.

The reason they don't match up is that there are really two different parameters in BCI2000 for controlling the number of flashes that occur and the number of flashes used for the pattern recognition system. The first is controlled by "NumberofSequences" under the P3Speller tab. The second is controlled under "NumERPstoAverage" under the P3 SignalProcessing tab. These two numbers must match.

This will fix it as long as you are using an application with 12 events, such as the 6 x 6 speller. If you are using a different number of events, you'll have another problem. Namely, the "NumberofSequences" always assumes you will be flashing through 12 events, and the "NumERPstoAverage" does not. We ran into this bug testing our P3 robot arm, which uses 6 icons instead of 12, and it works fine as long as the NumERPstoAverage is twice as high as NumberofSequences. This is a bug - both of these parameters should be dynamic to account for P3 applications with more or less than 12 events.