P300Classifier issues

Locked
alex
Posts: 2
Joined: 13 May 2009, 05:01

P300Classifier issues

Post by alex » 21 May 2010, 07:38

Hi,
we performed 2 P300 calibration sessions (using the same parameters) strictly following the protocol suggested in BCI2000 documentation, using 8 channels sampled at 128 Hz, 4 NumberOfSequences.

Then we used P300Classifier separately on the 2 file sets (4+4 files corresponding respectively to the words 'THE QUICK BROWN FOX' and 'CHIARE FRESCHE DOLCI ACQUE') obtaining 2 different feature weights sets which generated 2 *.prm files.
We could get respectively only 79% and 70% accuracy with 4 flashes, decimation frequency=64, spatial filter=CAR.

More we observed that it was impossible to set the decimation frequency at 128 Hz in order to use full data.

Then we tried to use P300 spelling sessions using separately the 2 *.prm files obtained in the previous step, varying also the NumberOfSequences/EpochsToAverage from 4 to 10 without obtaining significant results (trying to spell the word 'TOCAI' we hit just 2 letters in 1 trial, 1 letter in 2 trials and no letters in 3 more trials).

So we tried then to use P300Classifier using the 2 calibration file sets together (8 files), but after having loaded without problems all the 8 Training Data Files, while processing the data to generate Feature Weights, there appeared an error window and the application stopped.
We tried to vary the decimation frequency and spatial filter parameters, but we could not avoid the error: as a matter of fact it seems impossible also to work with joint subsets of the 2 original calibration sets, even if they are accepted without problems in the loading phase.

It was also impossible to load the 2 *.prm files sequentially in the Classifier matrix to be used in the P300 spelling sessions. Couldn't it be useful to improve the overall accuracy?

Is there any information available about the algorithms used in P300Classifier?

thanks
Alex

cmpotes
Posts: 13
Joined: 03 Oct 2009, 11:24

Post by cmpotes » 22 May 2010, 07:49

Alex,

1. Could you send me the data that makes the P300Classifier crash?

2. The classification in the P300Classifier is performed by using the stepwise linear discriminant analysis algorithm.

my email is cmpotes@miners.utep.edu

Thank you,

Locked

Who is online

Users browsing this forum: No registered users and 0 guests