ModelOriented/fairmodels

Basic question about fairmodel

Closed this issue · 2 comments

Nehagupta90 commented

Hi Jakub and other members, I have two questions about the basic concept of fairmodels ..

(1) Can we use any variable of a dataset as a protected variable which is either binary or nominal?

(2) Is there any way we use a continuous variable (with as much as 1000 different values) as a protected variable?

(3) I have an input variable which has values 9 and 13.. I used the fairmodel and found that there is no unprivileged value i.e. almost all the metrics fall in the green region. Would it be a good idea to use different class-balancing techniques and see if they (inclusion/exclusion of instances) impact the fairness of the model.?

jakwisn commented

Hi, answering your questions:

  1. Yes
  2. Yes, but it will be treated as a discrete one and I am not sure that 1000 values will be visible on the plot. Better to make some levels like 0-100, 101-500, 501-1000.
  3. Why not, It would be an interesting piece of research.
Nehagupta90 commented