In situations when I am imputing data as a recipe step, I would like to be able to tune the hyperparameters of the imputation method - which I anticipate is going to be a nightmare to implement, especially when parameters are passed between packages. For example, imputing with step_bagimpute uses ipred, which has its own hyperparameters but also passes on hyperparameters to rpart!
# the ipred hyperparameters, e.g. nbagg, can be set:
step_bagimpute(options = list(nbagg = 25))
# I'd like to be able to do something like:
step_bagimpute(options = list(nbagg = tune()))
But there's another level of HPs that ipred passes on to rpart, and it would be nice to be able to tune them as well.
# HPs can be passed through ipred to rpart
step_bagimpute(options = list(control = rpart::rpart.control(minsplit = 10))
# but it would be great to tune even these "next level" parameters:
step_bagimpute(options = list(control = rpart::rpart.control(minsplit = tune()))
Feature
In situations when I am imputing data as a recipe step, I would like to be able to tune the hyperparameters of the imputation method - which I anticipate is going to be a nightmare to implement, especially when parameters are passed between packages. For example, imputing with step_bagimpute uses ipred, which has its own hyperparameters but also passes on hyperparameters to rpart!
But there's another level of HPs that ipred passes on to rpart, and it would be nice to be able to tune them as well.
Good luck!