mb706

```r rsmp("bootstrap", does_not_exist = 1) #> Error in x$public_methods : object of type 'closure' is not subsettable ``` instead it should say something like "Cannot set argument ... (not a...

xgboost default nrounds should not be set

4

I think many users will assume that the "default" values that are given for a `Learner` are relatively save values that lead to ok performance in many cases. However, the...

Fix some parameters after some optimization

1

Scenario: I optimize dimensions (a, b, c) for N steps, then determine that the value for c should be c*, but I want to go on optimizing over (a, b)....

Non-Uniform sampling / Restriction of infill-opt-domain

1

E.g. my optimization happens over a simplex. One possibility is to map from the hypercube to the simplex using a trafo, but that would mean the surrogate model fits the...

There is no `leanify_package()` in https://github.com/mlr-org/mlr3hyperband/blob/main/R/zzz.R yet, is that on purpose?

CPO documentation

6

- work in progress -

Multi-objective feature selection, number of selected features as one criterion

1

Would theoretically work (although not entirely correct) for learners where the `msr("selected_features")` works. Would be nice if the measure worked for all learners.

mice.mids now supports newdata, so could be a pipeop

Rest API: provide example for dataset with with row-id and ignored features.

3

Not sure where to put this issue, since it is documentation-related on the one hand but not part of the docs.openml.org page on the other. The json response to `/task/{id}`...

Both the untransformed as well as the transformed version of the result should be shown in the tuning instance printers.