cpappas18 / articulaite Goto Github PK

Should we delete audio_features_visualization.ipynb ? I feel like it's very hard to read and doesn't show any actual visualizations, and we already have feature_extraction.py, so it might be better to delete this one. Unless someone wants to clean it up ?

This is happening because the multi-class datasets (like all_disorders_remove_std_onehot.csv) don't have a 'Status' column but rather 4 columns (Control, PD, etc.)

There's an issue with line 25 of the smote.py file, it seems that it also oversamples the Gender column resulting in genders other than 1 and 0 (like 0.608).

I don't fully understand the SMOTE technique yet but could it be fixed by only oversampling on the other columns (not Status and Gender) and then either 1) "filling in" randomly the genders with a 50/50 split or 2) with the same split as in the original dataset ? Unless it's fine to have non-binary genders ?