Page 2 of 3

Re: Classification error

Posted: Tue Mar 19, 2013 8:09 am
by jtlindgren
Hi Lorena,

we could debug the crash, but it'd be helpful to have all the files to reproduce it - whats missing is the CSP config, the classifier and the data set. Could you put them up somewhere?


Cheers,
Jussi

Re: Classification error

Posted: Tue Mar 19, 2013 4:01 pm
by Lorena
That would be great for me Jussi!

You can download the data to test my scenario from this link: http://rapidshare.com/files/3717526564/files.rar

Please let me know if you find where I am making a mistake.

Thanks!

Lorena

Re: Classification error

Posted: Tue Mar 19, 2013 4:26 pm
by jtlindgren
There seems to be a problem with the file, unfortunately I get

"Download not available
Download permission denied by uploader. (0b67c2f5)"


Best,
Jussi

Re: Classification error

Posted: Tue Mar 19, 2013 4:46 pm
by Lorena
Ok, can you try here: https://www.dropbox.com/s/elyzguom7uzvilo/files.rar

Let me know if there is any problem please.

Lorena

Re: Classification error

Posted: Wed Mar 20, 2013 8:54 am
by jtlindgren
Ok, got the files, thank you.

There definitely is a crash. But while I can fix the crash rather easily, the real problem seems to be that the Signal Display does not seem to work with the kind of data that the Classifier Processor outputs, so the crash fix still wouldn't give you a nice display that you want. Supporting this data seems a bigger task, esp. since the human resources we can utilize for this at the moment are not familiar with the signal display box code. But I can add it to our todo-list.

Meanwhile, perhaps someone has an idea what else you could do with the outputs instead? You can try to forward the data to Matrix Display for example, but it doesn't take the stimuli part.


Happy hacking,
Jussi

Re: Classification error

Posted: Wed Mar 20, 2013 10:47 am
by yrenard
For debugging purpose, the EBML stream spy will perform an all types of stream and decode their content to the log manager (be it in the console or in the embedded widget of the designer)

I hope this helps,
Best regards,
Yann

Re: Classification error

Posted: Wed Mar 20, 2013 3:51 pm
by Lorena
Ok, so it's more complicated than I thought...

Actually my problem is that I am not getting feedback during the online session and I suspect it is because the classifier is not doing anything. This is why I used the signal display conected to the classifier processor. Before I used the stimulation listener but I only got alternated left-right stimulus (the task had random, not organized, stimulus).

Is there other way to see what the classifier procesor is doing?

Thanks!

Lorena

Re: Classification error

Posted: Wed Mar 20, 2013 4:19 pm
by yrenard
Dear Lorena,

for debugging purpose, the EBML stream spy and the stimulation listener boxes are always a good choice. If something goes out of the preceeding box, you will know for sure. Once you're sure something goes out, we can investigate why the Graz viz box does not display it. As far as I remember, this box requires that the stimulation sequence is sent according to the Graz protocol so that the feedback appears - the OVTK_GDF_FeedbackContinuous may be involved here - you might want to double check the original Lua script to be sure about the stim sequence.

Best regards,
Yann

Re: Classification error

Posted: Wed Mar 20, 2013 4:40 pm
by Lorena
Thanks Yann, I will try again with the stimulation listener and check the Lua script then.

I will post here how it worked after some tests.

Lorena

Re: Classification error

Posted: Wed Mar 20, 2013 6:23 pm
by uahmed
Hi all,
I am a bit busy, and also not feeling well, so sorry for the late reply. I saw your algorithm I guess what you basically did wrong was with the no. of input channels, and also the location of cfg file. I am sending you my successful algorithms, use them with the input of 14 channels, don't use the gyroscope channels. I hope it will work.

thanks

Re: Classification error

Posted: Thu Mar 21, 2013 3:12 pm
by Lorena
Hi Uahmed, I hope you get better soon!

You say I have a wrong number of input channels and the location of the cfg file, but where? What settings should I correct? During the acquisition I only use 14 channel, and don't check the box of the gyroscope channels, so I am not using them.

Thanks!

Lorena

Re: Classification error

Posted: Mon Mar 25, 2013 4:56 pm
by uahmed
Hi Lorena,
you got my reply on email? Please share the files here also if you did.

Thanks

Re: Classification error

Posted: Mon Mar 25, 2013 5:22 pm
by Lorena
Hi Uahmed,

I'm sorry I couldn't answer before. I have been really busy and couldn't check the files you sent me, but now I will look at them.

Anyway, I found a big mistake in the scenarios I was working with, and that was the issue I was having with the feedback. The problem was that I was using a .lua file modified for me for the Graz Stimulator, instead of the original one. Apparently (unintentionally, of course) I had made there some modification in the feedback function besides the modification I wanted to obtain.

Now I have the time to see your files, so I will let you know how they worked for me.

Thank you very much for your help!!

Lorena

Re: Classification error

Posted: Mon Mar 25, 2013 6:29 pm
by uahmed
No problem :)

Re: Classification error

Posted: Tue Mar 26, 2013 4:12 pm
by Lorena
Hi all!

I am still having troubles trying to "see" in some way the output of my classifier processor. I have used the stimulation listener and, setting the log level to "information" I get stimulations all the time (I suppose this is correct, since the classifier processor should work all the time), is this correct? Should I consider this information as reliable?

I actually need to see values more than labels, but the signal display is not working connected to the output of the classifier processor. Is there any way of getting the output values of the classifier processor as a matrix? (something like a signal matrix).

Yann suggested me to use the EBML stream spy, but I am lost on what should I expect from it, what would be right and what would be wrong.

Thanks for the help!

(Note: I wasn't sure if I should post this issue here or in a new topic)

Lorena