Hi,
I am currently working on an application which uses the P300speller with xDAWN.
When the parameters of the xDAWN is at 3 (i mean that it will reduce the data to 3 dimension), it works. But when I try it with a different value, the replay cannot manage to give any answer (even a false one).
The xDAWN is generated, it seems to work for the classifier trainer but if there is something I should have modified, I guess it is either the classifier or the replay.
Do you know where would be my mistake?
Nicolas Dennewald
Problem with xDAWN on the P300speller
Re: Problem with xDAWN on the P300speller
Dear Nicolas,
you probably missed the voting classifier stage, didn't you ?
Yann
you probably missed the voting classifier stage, didn't you ?
Yann
my blog
my tweets
my linkedinRe: Problem with xDAWN on the P300speller
hi,
I am sorry but I don't really see what should be modified on the voting classifier
It does not seem to depend on the dimension of the filtered data. And as it works with 3 dimension, i don't really see why it would not with other values.
Does it depend on the dimension?
Nicolas Dennewald
I am sorry but I don't really see what should be modified on the voting classifier
It does not seem to depend on the dimension of the filtered data. And as it works with 3 dimension, i don't really see why it would not with other values.
Does it depend on the dimension?
Nicolas Dennewald
Re: Problem with xDAWN on the P300speller
Hello dennewan,
what value did you use exactly?
Note that the example signal has three channels, which means that you can use an xDAWN filter with at most 3 values. If you have tried to calculate 4 or more values then the trainer would fail (without a warning, which we will fix) but still provide a broken filter.
Cheers
Jozef
what value did you use exactly?
Note that the example signal has three channels, which means that you can use an xDAWN filter with at most 3 values. If you have tried to calculate 4 or more values then the trainer would fail (without a warning, which we will fix) but still provide a broken filter.
Cheers
Jozef
Re: Problem with xDAWN on the P300speller
Dear Nicolas,
sorry, I misunderstood your question, forget my first reply. As Jozef suggested, there is a small issue in the xDAWN box when the requested filter dimension is higher than the actual number of input channels - it reduces the number of output channels but still claims the filter has higher dimension. This can be fixed quickly. Meanwhile, try to always use a filter dimension lower or equal to the number of input channels and tell us if it solves your problem.
Yann
sorry, I misunderstood your question, forget my first reply. As Jozef suggested, there is a small issue in the xDAWN box when the requested filter dimension is higher than the actual number of input channels - it reduces the number of output channels but still claims the filter has higher dimension. This can be fixed quickly. Meanwhile, try to always use a filter dimension lower or equal to the number of input channels and tell us if it solves your problem.
Yann
Re: Problem with xDAWN on the P300speller
hi jlegeny,
I am using a BCI from gtec. I made the experiment with 10 electrodes. i used the 8 electrodes placement from "An effcient P300-based brain-computer interface for disabled subjects"
at http://infoscience.epfl.ch/record/10109 ... script.pdf
I added the O1 and O2 to get more data in the visual cortex.
Nicolas Dennewald
I am using a BCI from gtec. I made the experiment with 10 electrodes. i used the 8 electrodes placement from "An effcient P300-based brain-computer interface for disabled subjects"
at http://infoscience.epfl.ch/record/10109 ... script.pdf
I added the O1 and O2 to get more data in the visual cortex.
Nicolas Dennewald
Re: Problem with xDAWN on the P300speller
Dear Nicolas,
could you please post the log from OpenViBE to see if there were any issues? Also, make sure that you are using the good file in the replay scenario File Reader box - the file you recorded during the training session or during an online session.
-
Jozef
could you please post the log from OpenViBE to see if there were any issues? Also, make sure that you are using the good file in the replay scenario File Reader box - the file you recorded during the training session or during an online session.
-
Jozef
Re: Problem with xDAWN on the P300speller
hi,
I don't really know why but it works now.
However, i get a complete recognition of the caracters. No error whereas the classifier is said to be 64% accurate. Did I make a mistake?
Nicolas Dennewald
I don't really know why but it works now.
However, i get a complete recognition of the caracters. No error whereas the classifier is said to be 64% accurate. Did I make a mistake?
Nicolas Dennewald
- Attachments
-
- replay.png (48.06 KiB) Viewed 7659 times
Re: Problem with xDAWN on the P300speller
Dear Nicolas,
the classifier accuracy is computed for single trial. The p300 detection usually uses several repetitions so the final detection is usually better than the announced performance at training phase. For instance, if the single trial detection is 70%, after 12 repetitions, the accuracy is 100% as you'll have detected the right target sufficient number of times to be pretty sure this is the correct one.
Yann
the classifier accuracy is computed for single trial. The p300 detection usually uses several repetitions so the final detection is usually better than the announced performance at training phase. For instance, if the single trial detection is 70%, after 12 repetitions, the accuracy is 100% as you'll have detected the right target sufficient number of times to be pretty sure this is the correct one.
Yann