Giuseppe Casalicchio
Giuseppe Casalicchio
We are a group of researchers working on an open data and open machine learning plattform OpenML.org which already has ~20.000 datasets for which you can filter for specific properties...
```r library(datasets) library(margins) mtcars$cyl = as.factor(mtcars$cyl) mod = lm(mpg ~ cyl + hp + (cyl + hp):carb, data = mtcars) ame.effects = summary(margins(mod))$AME ame.effects # Your plots seem to have...
When we use `mlr:::makeTuneControlMBO` in mlr, the default uses 4*d for initial design, this should be configurable also the type of the initial design.
If one wants to use mbo for hyperparameter tuning, one can define the objective function as: ``` r obj.fun = function(x) { lrn2 = setHyperPars(lrn, par.vals = x) cv =...
As far as I can see, the [autoxgboost ](https://github.com/ja-thomas/autoxgboost/blob/b64048e603751bcba9b6e212c775baff8ababccb/R/autoxgboost.R#L158) function internally uses holdout for the objective function within the mbo tuning (it is hard-coded). 1) Wouldn't it be cool if...
If people use `early.stopping.rounds = 500` or some other high value, this line : https://github.com/ja-thomas/autoxgboost/blob/b64048e603751bcba9b6e212c775baff8ababccb/R/autoxgboost.R#L171 always measures the performance of the last iteration, right? Shouldn't the "best iteration" be used...
I feel that the dependency on R v3.6 is too strict. I have seen many companies still having only R 3.5 (or earlier) installed so that I had to remove...
For doing chunked requests, the listOMLRunEvaluation function should be in long format not in wide format. See also here https://github.com/openml/openml-r/issues/375 for some other possible improvements.
and store this in an extra slot so that it can be uploaded
@ja-thomas suggested to introduce a `convertMlrBMRtoOMLRun` function so that studies created with `mlr::benchmark` or the new `mlr::batchmark` function can simply be uploaded to openml.