Uniform distribution instead of random permutation about shapeffects

[CCaribe9]

Hi,

First of all, thank you for noticing. Indeed, what the paper says is not being done and should be changed (and will be changed) as you propose. Still, the effect of this modification should be minimal. The intention of that whole fragment is to introduce a random feature that has no relationship with the target variable. That way you can measure the importance of a feature that you know in advance that has no real relationship with the target to be predicted. The construction of this feature as indicated in the paper or with any other criteria should not have any impact, since all procedures would result in an unnecessary explanatory variable.

As I say, we will change the code and the results of the paper if they change considerably, which would surprise us.

Carlos

[wilmar-thomas]

hi Carlos, thanks for your reply. Indeed I would assume it's not material at all. Thanks for bringing an answer nonetheless