Currently, the group-wise regularisation coefficients are scaled by the square root of the group size, so the influence of each variable is the same. However, this does not make sense for dummy-encoded variables.
There should therefore be added an option upon init to disable the group-wise scaling so the user doesn't need to feed in a vector of regularisation coeficients.
yngvem
commented
4 years ago
Currently, the group-wise regularisation coefficients are scaled by the square root of the group size, so the influence of each variable is the same. However, this does not make sense for dummy-encoded variables.
There should therefore be added an option upon init to disable the group-wise scaling so the user doesn't need to feed in a vector of regularisation coeficients.