Add more feature selection methods

[sarthakpati]

Is your feature request related to a problem? Please describe. More sophisticated feature selection is required (per CD).

Describe the solution you'd like Add more feature selection methods that make the training module more robust.

Describe alternatives you've considered Use FE and then a separate feature selection.

Additional context

• Current implementations in CaPTk:
  • Effect-Size based
  • FFS based
• New stuff to explore:
  • OpenCV HFS: https://docs.opencv.org/master/dc/d29/group__hfs.html
  • TUM Vision group: https://github.com/tum-vision/afs
  • In scikit-learn: https://scikit-learn.org/stable/modules/feature_selection.html
  • CMICOT: https://github.com/yandex/CMICOT
  • mRMR: http://home.penglab.com/proj/mRMR/

[AlexanderGetka-cbica]

@sarthakpati Thanks for writing this up. I'm going to be taking a look at what other feature selection methods are out there, but who should I refer to first about which methods to implement/expose to the user?

[sbakas]

Some methods we should be considering: "Forward Feature Selection" "Recursive Feature Elimination" "Hierarchical feature selection" "LASSO regression"

[sarthakpati]

Some methods we should be considering: "Forward Feature Selection" "Recursive Feature Elimination" "Hierarchical feature selection" "LASSO regression"

Found some implementations:

• Forward feature selection:
  • https://github.com/rcmayhew/Naive_feature_selection/blob/master/main.cpp#L173
  • https://github.com/mfauvel/FFFS/blob/master/npfs.py#L364
• Recursive: https://github.com/prabathabey/svm-rfe/blob/master/svm-rfe.py
• Lasso: https://github.com/wangkuiyi/lasso