Classifier: Incremental Learning

About Brain Computer Interfaces, EEG, neurofeedback, related event announcements, jobs, etc.
Post Reply
etc
Posts: 16
Joined: Sat Mar 07, 2020 4:33 pm

Classifier: Incremental Learning

Post by etc »

Before to ask my question, congratulation for the new versione 3.0.0!!!

I have a question to try to train better the classifier. I think that the argument is known as “incremental learning". Now:
-I am on Openvibe 2.2.0

-I run the scenario P300-speller-1-acquisition; the Generic-stream-writer writes a file .ov for the scenario P300-speller-2-train-classifier. The name is P300-train[2020.10.29-14.10.37].ov

-Then, I run the scenario P300-speller-2-train-classifier; the classifier saves the result of the training in the file p300-classifier.cfg.

- I run the scenario P300-speller-3-online; the results are not good! I think that I have to train better the classifier.

- Then, I run again the scenario P300-speller-1-acquisition; the Generic-stream-writer writes a new file .ov for the scenario P300-speller-2-train-classifier. The name is P300-train[2020.10.31-17.25.45].ov

Now the question: I would like to get an “incremental learning”: in other words, I would like to train the classifier with the feature vectors in the first file P300-train[2020.10.29-14.10.37].ov AND (+) the feature vectors in the second file P300-train[2020.10.31-17.25.45].ov. Not only with the feature vectors in the first OR in the second file. I would like that the configuration (p300-classifier.cfg) can be built with the feature vectors of BOTH the files P300-train[ ].ov.

How can I do to get this opportunity? What scenario is good to get this opportunity? Have I to modify something (filenames, box or other) in the this scenario?
I hope that it is possible.
Thank you very much!
Ettore Coppola

Thibaut
Posts: 171
Joined: Wed Oct 31, 2018 9:14 am

Re: Classifier: Incremental Learning

Post by Thibaut »

HI,
You can use a little scenario as the one in link. to concatenate files. (You must just remove the stimulation who run the train on all scenario except the last.
Thibaut
Attachments
3-file-concatenation.xml
(18.81 KiB) Downloaded 8 times

etc
Posts: 16
Joined: Sat Mar 07, 2020 4:33 pm

Re: Classifier: Incremental Learning

Post by etc »

Dear Thibaut, I would say to you: THANK YOU!!!
Now, I didnt'understand what I have to do when you wrote "You must just remove the stimulation who run the train on all scenario except the last".
And I tested the new little scenario, but I didn't remove anything. I only wrote, in your new scenario, the name of the 2 input.
I made this test:
1) I have 2 file.ov: the file A (May 2020; 3,2 MB)) and the file B (October 2020; 7,6 MB);
2) I used your little-scenario and I got the concatened file A+B (10.4 MB);
3) I got 3 warnings; I attach the picture;
4) I put in input to the p300-speller-2-train-classifier the file A+B; I started the scenario and the scenario stopped at the end of the first part of A+B, in other words, at the end of the feature vectors of A.
5) Second test: I used again the your new little scenario and I built a new concatenated file: B+B (the file B concatenated with itsellf);
6) I got the same 3 warnings at point 3);
7) I put in input to the p300-speller-2-train-classifier the file B+B; I started the scenario and the scenario stopped at the end of the first part of B+B, in other words, at the end of the feature vectors of the first file B.
What can you say to me? Or perhaps the problem is the stimolation that I have to remove? But where and how? And what does"except the last" mean?"
Thank you very much!!!
Ettore
Attachments
2020-11-03-xopenvibe.jpg
2020-11-03-xopenvibe.jpg (188.14 KiB) Viewed 439 times

Thibaut
Posts: 171
Joined: Wed Oct 31, 2018 9:14 am

Re: Classifier: Incremental Learning

Post by Thibaut »

Hi, Sorry I have make a mistake. It's not scenario but file the word.
Your train file have a stimulation at the end to launch the train but if you concatenate file he launch the train when he receive this stimulation. so when you concatenate you must remove this stimulation on all file except the last (you have a stimulation filter box.
The warning is not important the concatenation make an inconsistent chunk date. OpenViBE say "Oh guys it's weird your date".^^

etc
Posts: 16
Joined: Sat Mar 07, 2020 4:33 pm

Re: Classifier: Incremental Learning

Post by etc »

Sorry, dear Thibaut! Give a look to the next my answer!
Last edited by etc on Thu Nov 05, 2020 10:58 am, edited 4 times in total.

etc
Posts: 16
Joined: Sat Mar 07, 2020 4:33 pm

Re: Classifier: Incremental Learning

Post by etc »

Dear Thibaut,
I put on Filter Box the parametr that you can see in the attached image. The filter box is only for the first file.
Now, the train classifier worked only for the second part of the concatened file... .
Where am I wrong?
Thank you
Ettore
Attachments
2020-11-05-stimulation-filter.jpg
2020-11-05-stimulation-filter.jpg (124.11 KiB) Viewed 420 times
2020-11-05-stimulation-filter-2.jpg
2020-11-05-stimulation-filter-2.jpg (66.62 KiB) Viewed 420 times

Thibaut
Posts: 171
Joined: Wed Oct 31, 2018 9:14 am

Re: Classifier: Incremental Learning

Post by Thibaut »

No No, it starts training once he has read both parts not just the second part. The classifier starts training on everything he received before this stimulation if it is deleted from the first file, he waits to receive it thanks to the second. (and if he does not appear in the second, he waits indefinitely for this stimulation.

etc
Posts: 16
Joined: Sat Mar 07, 2020 4:33 pm

Re: Classifier: Incremental Learning

Post by etc »

Dear Thibaut,
after a lot of tests, I succeeded!!!
Thank you very much for the scenario that you built to get this other option to concatenate 2 files!
Ettore

Post Reply