Doubts regarding pref-shap implementation

[elishaparhi]

Hi,

I have the following doubts regarding the implementation part of pref-shap. Can you please help me with this?

1. While estimating the Shapley values using the weighted linear regression, for the value function of the null set, we should put the value that results from the original function when all the features are treated as random variables. It might be zero as well. I am not sure, why you have used the mean-centred predictions for the test pairs in that part.

2. For the value function of the grand coalition, I think we have to put the particular function values or the kernel logistic regression values for those test pairs. In the code, why zero has been taken in that part?

3. In OLS_solve, why the weights have not been used while computing b = (Z^T)(W)v_x (in line 432)?

Thanks,
Elisha