CODEWITHSUNDEEP
cross-validationmlr3
ds_col
ds_col

Reputation: 139

Leave-One-Out Cross-Validation in mlr3

I like to use Leave-One-Out Cross-Validation in mlr3 (as part of a pipeline).

I could specify the number of folds (=number of instances) e.g. via

resampling = rsmp("cv", folds=task$nrow)

But this refers explicitly to "task" which may not work in the pipeline.

How can I proceed here?

Upvotes: 3

Views: 198

Answers (1)

mb706
mb706

Reputation: 672

There is a specific "leave-one-out" Resampling object, named "loo". It can be used like any Resampling object, for example:

rs <- rsmp("loo")
rr <- resample(tsk("iris"), lrn("classif.rpart"), resampling = rs)
rr$aggregate()
#> classif.ce
#> 0.06666667

It determines the number of iterations directly from the Task being used, and has no configuration parameters (i.e. has an empty $param_set).

Upvotes: 6

Related Questions