bci2000.org BBS

Hello All

I was trying to control 1D and 2D cursor movement using Emotiv EPOC. I am getting good data as well but the problem is coming with Classifiers. I was getting best r2 values from FC5 and FC6 so I am using them as follows:

FC5 10Hz 1 0.5
FC6 10Hz 1 -0.5

ERROR: Normalizer::Preflight: The number of columns in the BufferConditions parameter may not exceed the number of input channels.

WARNING: Warning: LinearClassifier::Preflight: Specification in physical units:
Classifier(1,input element (bin))=10Hz
does not exactly meet a single element.

Kindly suggest me how to resolve this problem.

Thanks

The first error is because your normalizer is trying to normalize two control signals even though you've modified your classifier to only output one control channel (for 1D control). You should resize your buffer condition matrix to only have one column.

The warning I understand only partially. The spectral processing filter you're using isn't returning a frequency bin that is exactly centered at 10 Hz. If you are using the AR signal processing module with the default settings then your frequencies are 0, 3, 6, 9, 12, 15, etc. I suspect if you specify 9Hz instead of 10Hz then you won't have a problem. I'm not sure if the warning means that more than 1 matching element was identified or only that the 1 element identified doesn't match exactly. In any case, you might prefer to modify the spectral processing parameters so that it returns frequency bins exactly as you desire.

Hi Chris,

thanks for answering the question, and especially for pointing out that the warning message is difficult to make sense of. I changed the wording to "is not an exact match", in the hope to be more precise. If you have a better suggestion, please let me know!

Regards,
Juergen

It's Chad, not Chris, but I can't expect you to remember that unless I sign my posts.
-Chad

The new warning is less ambiguous.

How about, "is not an exact match. Using nearest match."
The advantage is that users will not have to worry about whether the inexactly matched element was dropped from the classifier or if it was replaced with the nearest match.

This might be improved upon slightly with, " matching element not found. Using %s instead." % matched_element + " " + units
That way if someone used bins at 8Hz and 10Hz (again using the AR default settings with 6,9,12 Hz bin-centers) then they'd get a hint that both classifier entries are returning the exact same element.

-Chad

Hi Chad,

very sorry about your name, and thanks for the additional suggestion.
Also, thank you so much for all the helpful answers on this forum!

Cheers,
Juergen