Fri.jl

This repository contains a simple Julia implementation of the feature relevance bounds method.

It exists a much more complete python version here. This is mostly a proof of concept and is missing features such as cross validation for hyper parameteres, regression and ordinal regression models and advanced statistics for feature classification.

Quickstart

A runnable example is included in the example notebook.

include("src/Fri.jl")

┌ Info: Precompiling JuMP [4076af6c-e467-56ae-b986-b466b2749572]
└ @ Base loading.jl:1273
┌ Info: Precompiling ECOS [e2685f51-7e38-5353-a97d-a921fd2c8199]
└ @ Base loading.jl:1273
Main.Fri

We generate dataset with 200 samples, 5 strongly relevant features, 4 weakly relevant features and 10 noise features (irrelevant).

X,y = Main.Fri.dataset.generate(200,d_rel=5,d_irrel=10,d_weak=4);

relev_bounds = Main.Fri.relevance_bounds(X,y)

17×2 Array{Float64,2}:
  2.55727      2.55727  
  2.01039      2.01039  
  2.32116      2.32116  
  2.09897      2.09897  
  2.35608      2.35608  
 -3.85029e-12  2.22748  
 -3.8504e-12   2.22748  
  0.0948441    0.0948441
  0.100589     0.100589 
  0.156153     0.156153 
  0.0760626    0.0760626
  0.11897      0.11897  
  0.19504      0.19504  
  0.09535      0.09535  
  0.0669823    0.0669823
  0.22617      0.22617  
  0.0860516    0.0860516

Minimal relevance for feature 1

relev_bounds[1,1]

2.5572730462733446

Maximal relevance for feature 1

relev_bounds[1,2]

2.557273051045558