How is 'score' calculated in fscmrmr?

I am trying to use fscmrmr for my dataset. I understand from the help page on how rank is calculated for each feature. However, it is mentioned that score is calculated using a heuristic algorithm. Could you please share how score is calculated? Thanks.

Answer 1

[Shashank Gupta]

Hi Cordelia,

The scores here represents the scores of individual features in respect to previously selected feature (also mentioned in standard MRMR Procedure). In simple words, the scores can be defined as the Mutual Information(MI) between target and the considered feature minus the average MI of previously selected feature and the considered feature.

If you are looking for more details, you can try looking at the source code of this function and understand the workflow. Neverthless I hope my insight helped you in some way.

Cheers

[Cordelia Ezhilarasu]

Hi Shashank,

Thanks for the explanation. It clarifies my doubt on the score algorithm.

Kind regards

[Yezi Kadhim]

Dear Shashank,

I was just searching about the scores in fscmrmr and I found your post. Here is my setting, I used CNN to extract features from a given dataset (tumor images). And I was trying to minimize them through hybrid approach of a set of feature selection algorithm. The final features file is a N_of_images * 1024 in dimension. How do I submit the scores array to fscmrmr?. I checked so many and further I found (even in mathwork tutorials) that the score is a column in the dataset/features file. What I am missing. Any help would be appreciated.

Thanks in advance.