mb706

- don't assert on function with only one argument, we can't set `trafo = log` currently. - give the head of the task to allow conversion between tasks.

Use [PipeOpStratify](https://github.com/mlr-org/smashy/blob/master/R/PipeOpStratify.R). Things to consider: * Explicit output with defined channels as an option. * Empty multiplicities don't work yet I believe, this needs to be repaired. * option to...

this is because S3 printers have precedence before R6-printers.

All `Task`-`PipeOp`s should automatically call `$predict` on the `"test"`-rows, so operations (or learners) that come later in the stream can rely on the `"test"` rows being valid. Ideally this should...

... and not the datatypes it gets from `$data()` (https://github.com/mlr-org/mlr3/issues/685). Otherwise subsequent PipeOpTaskPreprocs can give errors about incongruent train/predict-tasks.

Another POFU issue... apparently `DataBackendRename$missings` breaks when queried for no columns. This happens because `cbind()` overwrites all columns in this case. ```r > gr >!% po("featureunion", innum = c("a", ""))...

```r as_task_classif(data.frame(x = factor(c("a", "b", "a", "b"), levels = c("a", "b")), y = factor(c("a", "b", "a", "b"), levels = c("a", "b"))), target = "y") #> Error in .__Task__initialize(self = self,...

Apparently we get some overhead when mlr3pipelines builds tasks with many BackendCbinds. One way to fix this would be if there were an option to "flatten" cbinded tasks. Suggested interface:...

This way we could correctly query a pipeline that selects features first and gives the result to a learner. The GraphLearner could then ask the learner at the end how...

When I have a dataset that I want to convert to a task, my first intuition would be to look up the help text of `as_task`. However, it does not...