Linear Classifier Matrix ambiguous

[omarz]

Hi,
I did an mu offline analysis of 1D control experiment, I have 3 conditions Left, Right, Rest, and I am recording from C3,C4,Cz. The analysis showed a strong component around 9Hz, and clear difference between Rest-Left, Rest-Right, however there is no clear difference between Left-Right. So I decided to use Rest-Right for example to perform another 1D control experiment. My question is how can I translate this into parameters of the Linear Classifier matrix, since they are two separate events. I used to use two input channels C3,C4 and one output channel 1 , freq is set to 10Hz, and the weights for C3 is 1 and for C4 is -1.
I still found the linear matrix ambiguous, and the way the control signal is generated as well. I know the control signal will be the summation of the input signal features at the specified frequency times the corresponding weights in the matrix, however, this is computed during the current event, is that right?

this is how my Linear Classifier matrix, I use an identity matrix for spatial filtering
================================
Filtering:LinearClassifier matrix Classifier= 2 { input%20channel input%20element%20(bin) output%20channel weight } 1 10Hz 1 -1 2 10Hz 1 1 // Linear classification matrix in sparse representation
=================================
your help is appreciated,

Omar,

[gschalk]

Omar,

I think there is a misunderstanding. You say you found a robust difference at 9 Hz for Rest vs. Right Hand. You do not say at which location. Let's say C3. In this case, all you have to tell the classifier to use C3 at 9 Hz. It will simply always calculate the amplitude at 9 Hz, i.e., during all "events," and will simply produce larger numbers for one condition than for the other, i.e., allowing the subject to move the cursor.

In other words, you do not have to have different classifiers for different "events." You simply tell BCI2000 which signal the subject can modulate.

Gerv

[omarz]

Hi Gerv,
I've got your point, I was a bit confused. I've got one more thing to clarify about the Linear classifier matrix. Suppose I have two input channels C3, C4, and I want to take the difference between C3 and C4 at say 10 Hz so as to control cursor movement, Now if I want to subtract C3 from C4 that is C4 - C3, would I set the weight of C3 to -1 or is it the opposite.
If you check the link and 2D cursor movement example http://www.bci2000.org/wiki/index.php/U ... Classifier
you would notice that the weight of CP4 is set to -0.5 and CP3 is 0.5, while the underlying statement says subtracting CP3 from CP4?!.

The reason I am asking such question is that I've seen some papers that show spectra at C3, C4 during Right, Left hand Imagery movement. The spectra of C3 is lower than of C4 during Right movement, while the spectra at C4 is lower than of C3 during Left Movement. obviously we want to use the difference in this case.
How can you translate this scenario of offline analysis into the Linear Classifier Matrix.

sorry for being long, your help is appreciated.

Omar,

[mellinger]

Omar,
the help page was wrong, and is fixed now. Thank you for your note.
Juergen