mb706

I am personally not bothered by getting "bad" issues. If someone is on the edge between submitting an issue or not submitting it, I would prefer for them to submit...

potential problem: we don't want to tempt people into doing the wrong kind of preprocessing on train and test set

suggestion (bb): ```r outdata = preproc("encode", indata, paramlist, state = state) ```

maybe we want to have `preproc(po("encode", paramlist))` instead, more consistent and would also work with PipeOps that are not in `po()`.

> Is there a single learner that uses it during training? Grepping a bit: * [`LearnerClassifSvm`](https://github.com/mlr-org/mlr3learners/blob/59cc447d05a438a7d6d72f04863333161806683c/R/LearnerClassifSVM.R#L70) gives it as an argument to the model fit, idk how much it slows...

I'm assuming @badrigopalan means the [ask-tell-interface from skopt](https://scikit-optimize.github.io/stable/auto_examples/ask-and-tell.html), what another optimizer calls ["Human in the Loop" optimization](https://mlrmbo.mlr-org.com/articles/supplementary/human_in_the_loop_MBO.html)? So the idea would be to have the SMAC optimizer *return* the configuration(s)...

This one will replace the chapter in the master. I am currently a bit overwhelmed with other responsibilities, however, so will probably only be able to finish it in two...

It is there to make a point about pipeops with different input / output types. This can be replaced with another pipeop that has varying output types (back then we...

Maybe set formula hyperparameter of a learner through a previous pipeop

mlr3pipelines makes the assumption that the following works: ```r task