neil-gallagher / csfa Goto Github PK

Thank you for the nice repository! I would like to ask about the prediction scores of each electome factor:
As I understand it, the scores that are saved at the end of trainCSFA represent the activity of each electome factor in each trial. But is there a way to rank them regarding their predictability (i.e., the higher the score the more important for classification?). Basically my goal would be to get to the same figure as in Hultman et al., Cell, 2018 in Figure 1F to get a ranking of factors regarding their contribution to classifying between two conditions.

When trying to run the trainCSFA function to obtain the dCSFA, I stumbled upon the following question:
As suggested in the description of trainCSFA, the target information for the classifier is stored in labels.windows as an 60x1 array containing ones for a positive and zeros for a negative event. However, initModel transforms the target variable into a 60x2 by adding another array below with ones and zeros swapped. Subsequently balanceClasses bails out an error because upsampling is not possible because sum(labels==1) and sum(labels==0) are both an array of the same length. Also, fitcsvm requires the response variable to be a vector, not a 2-row matrix.
Most likely I'm missing something here, but I would appreciate Clarification on this.
Thanks and best regards
Daniel