creating a custom Box

[KJeffrey]

Hello,

We have been trying to create a custom box that does these things:

1. transforms the EEG data to a type of Z-score (requires all of the data in order to calculate means and stddevs over all trials)
2. selects the top P percent of the data (P entered by the user) that matches the top P percent of the Z-scores
3. sends only the selected portion on to the next step in the scenario (the classifier)

The point is to select only the more significant features to send to the classifier. We are starting with 192,000 features (64 channels x 500 time frames x 6 frequency bands) and want to select the more significant 10-15% to send to the classifier. We have proven this works well by using the Matlab version of libsvm, as libsvm is the package used by OpenViBE.

We were advised that we would be able to do this using the Python Scripting box, but have discovered that this is not possible because of limitations in the data types that that box can export. The same limitations apply to the Matlab Scripting Box. Now we are wondering if the standard Python box can be used for the above procedure? If so, we will need some guidance on creating it.

-Jeff

[KJeffrey]

This is the scenario we are using. The proposed Box would fit after each of the two Feature Aggregator Boxes and the SVM Classifier Box.
smok_clas_svm_train.xml attached.
-Jeff

smok_clas_svm_train.png (33.97 KiB) Viewed 4930 times