andreassot10

Thanks @pfistfl. I think that I could help with this- or at least I could try. I'm doing this stuff for work, which means that I cannot allocate all of...

Sounds reasonable. I'll take a look this week.

@pfistfl , I need a little bit of support on this please, as I'm pretty new to the R6 stuff. So I'm trying to understand what the backends are and...

Thanks so much for the detailed response and ideas, @pfistfl. I'm almost convinced that the only way forward would be to break completely free from `data.table`/`data.frame`/`Matrix` formats and create a...

So a `dfm` is, according to its authors: _"[...] a type of Matrix-class object with additional slots, described below [in `dfm-class {quanteda}`]. quanteda uses two subclasses of the dfm class,...

Thanks, things are much clearer now. As it turns out, it's the conversion from `dfm` to `matrix` with `quanteda::convert` that slows things down in `PipeOpTextVectorizer`: https://github.com/mlr-org/mlr3pipelines/blob/6427f5e9377d7c3d7e1e1aac063c410cffb351b9/R/PipeOpTextVectorizer.R#L239 Converting the `matrix` to...

Apologies for the long silence. I'm working on [solution (2)](https://github.com/andreassot10/mlr3extralearners), i.e. build a `mlr3extralearners` version of `quanteda`'s Multinomial NB model that directly incorporates `mlr3pipelines::PipeOpTextVectorizer` in it, to avoid the unnecessary...

Also on [StackOverflow](https://stackoverflow.com/questions/61772147/tuning-smotes-k-with-a-trafo-fails-warningk-should-be-less-than-sample-size)

> > > I think this can be seen from two perspectives: > > * If you specify an invalid parameter, the method **should** fail, and a backup learner might...

> > > 1. Ok, a couple of comments. I think this is somewhat of a "fringe issue"; sorry @andreassot10. > Currently, you might have to live with this not...