how to run classifier online and with pre=recorded EEG

[KJeffrey]

Hi,

We have a classification scenario with two classes. we have trained it using SVM, which created the SVM model in a file. We now want to use that model during our neurofeedback sessions, and on pre-recorded data. we tried to follow the motor-imagery example, but are stuck. Here is what we have for the training scenario:

and this is for the playback:

smok_proc.png (24.67 KiB) Viewed 7604 times

Our situation varies from the motor-imagery scenario, so we are not quite sure how to keep the stimulation-based epoching in the pipeline. That is where we put the 1.0 second epoch with 0 offset. If we try to insert a stim-based epoching box, we can only specify one of the two classes. If we insert a time-based epoch we cannot control the proper time and offset.

Please advise

-Jeff

[KJeffrey]

had problem with this attachment, trying again
-Jeff

[Thomas]

Hi Jeff,

Indeed you don't want to use the StimulationBasedEpoching since it would mean that you already know the class, and therefore don't need the classifier...
The StimulationBasedEpoching is only needed for training, to filter each class in the corresponding input the classifier trainer.

If you use the Time Base Epoching, with epochs of 1 second at 0.0625s interval (as in the motor imagery scenario):
It acts as a moving window, allowing your pipeline to process and detect one of the two classes anywhere it would appear in the signal.
What doesn't work for you with this approch and how would you like to control the time and offset of the epoching ?

Thnaks,
Thomas

[KJeffrey]

Hi Thomas,

I did not see your response until tonight. Is there any way to be notified by email when there are new replies to a subject?

I will have to go back and try the Time-based epoching as you suggest here and see what happens. As I recall it failed for some reason.

More important now is our inability to perform feature selection. Please read the next post I make on this topic.

-Jeff