mlr3pipelines
mlr3pipelines copied to clipboard
mice package
mice.mids now supports newdata, so could be a pipeop
mice() function: we should have all hyperparameters that are scalar, fixed length, length(ncol) but not length(nrow), since that should always be possible to change. The 'blocks' argument should also allow Selector functions.
Prediction seems to be difficult, see
- https://github.com/amices/mice/discussions/336
- https://github.com/amices/mice/discussions/346
I am not entirely convinced this is all there is, since predictions on the training set give different imputed values than the imputation from training. Is this an RNG thing?