Use knitr or the minted package for highlighted and easy copyable R code.

General

• Cheatsheets usually contain no hyperlinks as one of their main purposes is the printed version (i.e. the "mlr3book" link and "mlr-org" link)

• Use grey background for inline code formatting, e.g. mlr task is hard to distinguish between code and normal text

• replace www.mlr-org.com by https://mlr-org.com

• reduce header height for more content space?

• Change font to roboto

Packages

• The "Packages" section should only contain the "core" packages and also name them like that. I.e. no survival and ordinal. And possibly provide a link to the overview of all extension packages

Intro & Workflow

• "The mlr3 package provides R6 classes for the essential
  building blocks of this machine learning workflow:"

  • R6 is only the system behind, so better say "builds on R6 classes" or similar
  • "this machine learning workflow" -> "a common machine learning workflow"

• Rather than starting all listings with "A [...]" just do "Task: Encapsulates [...]"

• maybe "encapsulates" is too techy here, rephrase?

• color the workflow boxes? Replace workflow figure

Task

• "If you simply provide the dataset, it is automatically
  converted to a DataBackendDataTable." -> The default backend is "DataBackendDataTable".
  Then, list all other available backends in a list? We do not have space for this. I removed this part completely. I think this is not of interest for the normal user.

• Before starting with code, add a heading like "examples"

• TaskRegr$newt(backend, target) <- typo

• "For classification, the name of the target column is a label with only few distinct values." -> sounds more special than it is. Maybe just write something like "for classif, target is a character vectors with representing class labels"?

• Rather than showing all methods, show the output of an example task? There are too many methods for a task and it gets boring quite quickly. Also all of this is in more detail in the help page whereas the output of a created task is not.

Learner

• Show mlr_learners dictionary? Replace all dictionaries with sugar

• Mention and link to mlr3learners org and custom learners?

• "Get learner by key and construct the learner with specific
  hyperparameter and settings (...) in one go." -> Do not show a generic way but rather a concrete example.

Train & predict

• Show a train-predict workflow example rather than a generic example

Resampling

• "Resampling is used to assess the performance of a learning algorithm." -> mention "cross-validation" somewhere here

• shortly explain the difference/similarity between resample/cross-validation

• mention that multiple train&predict calls are done during resampling

Benchmarking

• Remove expand_grid()

• I know, not much space left, but maybe also show the output of a BMR?