Hi,
few clarifications.. in order to clarify what I have done!
1)For the Screening test: I used the Stimulus Presentation Task. I was wondering what I have to set in the Interpret Mode.
- none;
- on line or free mode;
- copy mode;
I answered myself that as for the screening task BCI2000 doesn’t need to do an online classification I should set this parameter to NONE, Am I right?
2) SpatialFilter matrix: I tried to load the 16Car.prm found in the examples directory, and I noticed that all the 16channels on the main diagonal are set to one (..0,93?), it is just an example isn't it?. If in my case the control channels are only three I should set to 1 only those channels, shouldn’t I??
3) Related to the above question..while the BCI2000 is running, in the SpatialFilter Window what I should see? Of course, only channels I want to filter, but I mean how the signal should be? I noticed that in my experiments the signal of each channel are flat?.is it correct?or something is wrong?
4) If in the BCI2000 I use CAR filter I should use the same filter in the Offline Analysis? Because for a strange reason some data gave more interesting results with NONE filter. ?! Is the answer related to the ground and reference positions??
5) Can you address me where I can find more explanation of How the classifier acts in the BCI2000, I don’t understand well how I should choose the sign of weight (-1, +1). I read on Wiki that the spectrum involves taking the absolutes value (or its square) which is a nonlinear operation: |A-B| is different from |A|-|B|, ? Then I answered, as the offline perform the second operation..if this operation is positive (i.e. the difference between the absolute value of the different class signal ) I put +1 if it is negative I will put -1. In this way I can balance the cursor equation? y=offset(gain- differential potential)), is it correct?!?
Thank you in advance for your great help.
Cheers,
Maria Laura
setting online and offline parameters
-
gschalk
- Posts: 615
- Joined: 28 Jan 2003, 12:37
questions ....
Maria Laura,
Thanks for the good questions.
Regarding the weight, this is very simple. After spatial filtering and spectral analyses, you have one value of the spectrum at each channel and frequency bin. Then, the classifier can produce a weighted linear combination of amplitudes at particular channels and frequencies for each output control signal. Larger output control signal values make the cursor go right or up. Thus, depending on where you want the cursor to go with what subject instruction, you will need to chose a negative or positive weight.
Gerv
Thanks for the good questions.
None is correct for your purpose. StimulusPresentation does not need to interpret results.1)For the Screening test: I used the Stimulus Presentation Task. I was wondering what I have to set in the Interpret Mode.
I assume that the values off the diagonal are a small negative value. What this does is to produce 16 output channels where each output channel is a linear combination of all input channels, such that it subtracts from the center electrode the mean of all the other electrodes (i.e., a CAR filter; see BCI2000 documentation on SpatialFilter). You could definitely use this, thus doing spectral analyses on all 16 channels and, e.g., only select 3 of these channels in the classifier.2) SpatialFilter matrix: I tried to load the 16Car.prm found in the examples directory, and I noticed that all the 16channels on the main diagonal are set to one (..0,93?), it is just an example isn't it?. If in my case the control channels are only three I should set to 1 only those channels, shouldn’t I??
Yes, use the filter that gave the best answer offline. However, this should typically be a CAR filter and not NONE.4) If in the BCI2000 I use CAR filter I should use the same filter in the Offline Analysis? Because for a strange reason some data gave more interesting results with NONE filter. ?! Is the answer related to the ground and reference positions??
Regarding the weight, this is very simple. After spatial filtering and spectral analyses, you have one value of the spectrum at each channel and frequency bin. Then, the classifier can produce a weighted linear combination of amplitudes at particular channels and frequencies for each output control signal. Larger output control signal values make the cursor go right or up. Thus, depending on where you want the cursor to go with what subject instruction, you will need to chose a negative or positive weight.
Gerv
Who is online
Users browsing this forum: No registered users and 0 guests
