OpenViBE forum

Hello all,

I am trying to run a scenario that must discriminate between left and right movements. I train the classifier but have no results in the console, but have not error messages or warnings either. Then, when I run the online scenario, I get a warning "Feature vector size 2 and hyperplane parameter size 0 does not match"

I used the motor imagery-CSP scenario, but as I am using emotiv, couldn't use CSP filter (right??). So, I replaced CSP for 2 boxes: Channel Selector and Spatial Filter (with coefficients to obtain a Laplacian filter), could the problem be caused by this change?

I see in the motor imagery scenario using Laplacian filter that there is a Reference Channel box that I am not using, is this box mandatory in this configuration? What is its purpose?

Thanks for any help!

Lorena

Hi Lorena,
Good to see you again :)Actually emotiv does not really fit for the CSP scenario. Since I have personally encountered the same kind of problems, I guess I can help you now. The CSP filter won`t work for emotiv. And you should not input the gyroscope channel while online acquisition scenario.
Thanks

Thanks Uahmed,

Actually I didn't use the CSP, I replaced it with the Laplacian filter.

Which is the gyroscope channel? Shoul I remove it for the online acquisition AND the classifier trainer scenario as well?

Lorena

Dear Lorena,

it is true that the Emotiv device is not very suitable for motor imagery as it does not really cover motor cortices. This said, the CSP algorithm generates a spatial filters that maximizes the detection of the two classes. At worse, it will perform as good as a Laplacian that is entered by hand, based on your own knowledge. So knowing motor imagery will be difficult to achieve with Emotiv, if you really want to do it, go for the CSP, it can't be anything but better than the Laplacian based scenarios.

As to the reference, it is not mandatory, the device that was used to acquire the demo signals needed this as it natively delivered "common average referenced" signals, which is not the case of the Emotiv device.

Hope this helps,
Best regards,
Yann

Thanks Yann, I will try with the CSP saenario then

Hi,

Well, I tried the CSP scenario. When I train the CSP, I don't get any kind of info message in the status bar, which makes me think I am not getting a valid configuration file. Then, when I want to train the classifier I get a message that says configuration file can't be found (or something like that, but that's the message ).

I can´t see where is my problem, because I have used these same scenarios before. The only thing I changed is the band pass filter cut frequencies, I use the 0.2-40Hz band, but I don't think this generates an error.

Thanks!

Lorena

These are the messages I am getting:

[WARNING] Could not override [CSP Spatial Filter] settings because configuration file [../share/openvibe-scenarios/bci/eog-bci/csp-config-1.cfg] could not be opened

[ ERROR ] At time 0.063 sec <Box algorithm::CSP Spatial Filter> Bad matrix size - Expected 16 input channels and received 14 input channels

[WARNING] Box algorithm <CSP Spatial Filter> has been deactivated because process phase returned bad status

Apparently, there is something with te number of channels I am using.

Thanks for any help!

Lorena

Dear Lorena,

I suspect the train stimulation is never sent to the CSP Filter Training box (do you get the stimulations from the last recorded session ?). You might want to tune your stimulation choice in the training box settings - or - switch to a manual triggering of the stimulation with a Keyboard Stimulator for instance.

Yann

Yes Yann, that was the problem, I had to modify the train trigger and it worked perfectly.

Thanks!

Lorena

Well, now I see a new problem: when I perform the online classification, I can't make work the feedback. I used a stimulation listener and all it showed was stimulations right in a row followed by left stimulations in a row. I think classifier the processor is not really classifying anything.

What could this be caused for?

Thanks!

Lorena

Hi all,
Don`t use the stimulation listener. You have to work with the SIGNAL out of the classifier processor, not the STIMULATION part. for left signals, a negative value is output, and for right signals, a positive value is output.

And one thing more, I did not use any filter in my training of classifier trainer, and then it gave me the best results.

Thanks

Thanks Uahmed, I will see the signal now then.

When you did your tests, could you see the blue bar due to the feedback while you were performing the online acquisition? Because I am not getting any feedback.

Thanks!

Lorena

I tried using the signal display to see the classification results, but I don't see any signal there. I get the following error:

[WARNING] At time 1.297 sec <Box algorithm::Signal display> Error dimmension count isn't 2!

I suspect my classifier processor isn't working, could this be possible?

Lorena

Yes, I did get it, and that feedback bar actually shows the accuracy of results. I guess you are making some mistake in your input of channels, and the settings. You can send me the Xml files,I CAN CHECK THEM.

Thanks

Hi all,

I am still trying to make work my online classification, and for one of the tests I prepared the scenario I attach in this message. There must be a big problem with it because every time I run it, Openvibe crashes. I attach also the image with the error messages I get before the crash. Where could the problem be?

By the way, thanks Uahmed for your offer! I already sent you my scenario

Thanks for any help!

Lorena