Sebastian Fischer

See e.g. [this issue](https://github.com/mlr-org/mlr3/issues/961) where things go wrong.

``` r library(mlr3) task = tsk("iris") task$select(c("Petal.Length")) task$col_info #> id type levels label fix_factor_levels #> 1: ..row_id integer FALSE #> 2: Petal.Length numeric FALSE #> 3: Petal.Width numeric FALSE #>...

Tests / Checks for flipped factors

1

I can imagine that some learner's don't handle factors with flipped levels perfectly. A proposal would be to add an autotest that checks whether predictions are equivalent irrespective of the...

Is this intentional? ``` r library(mlr3) learner = lrn("regr.rpart") task = tsk("mtcars") learner$train(task) learner$state$train_task #> (32 x 11): Motor Trends #> * Target: mpg #> * Properties: - #> *...

It can happen that custom resamplings have exactly the same instance but are instantiated on different tasks. I would still expect these resamplings to have a different hash. This case...

One thing that is hugely annoying I think is that we do not have autocompletion for the names of parameters. Maybe we can somehow add a method to a learner...

In some cases, users might want to fit large models on large datasets using `resample()`. In this case [here](https://stackoverflow.com/questions/75405217/no-model-stored-with-mlr3), the person was unable to fit the models into RAM when...

[Suggestion] Serialization property for learners

2

Some learners (especiall torch learners) but also e.g. lightgbm break when saved and reloaded. One situation where this occurs is if one calls `benchmark(..., store_models = TRUE)`. To avoid this...

Some learners require (such as priority lass from mlr3extralearners) require to set a parameter to a specific value, that depends on properties of the tasks (in this case, the number...

We recommend to add r-universe to the repos, and then mention how to install the dev version. If they add r-universe to the repo as suggested, they always get the...