Weiyuan Wu

I see. So it seems DataPrep automatically identified your columns as categorical. May I ask what is the output of `dataframe.dtypes`?

Hi @AndrejIring thanks for the bug report and the detailed analysis of the reason! I'll take a look into the fix.

@DiptenduIDEAS Yeah I agree that this is very important. We do have this in design (e.g. `eda.missing.compute_missing`.) but the format is not standardized yet. Having a standardized format will be...

It also panics when the df is empty.

I tried this one and it gives me 9% speed up on the `basic_batched` benchmark.

I think one of the reasons is that `ParquetFiles.jl` doesn't have the interface `Tables.columns` implemented, which makes `DataFrame(...)` go to the fallback solution, that is, row by row appending.

@logaritmisk @duelinmarkers using jimmycuadra/rust solves the issue. Basically, scorpil/rust lacks of `make` installed

Can we have Default derived for Level?

> Added. The maintainers doesn't seem to be interested in this PR though. Thank you! Yeah I hope this PR can be merged soon. I've been using your branch for...

> Would `#[cfg_attr(not(feature = "bytemuck), forbid(unsafe))]` work? Yeah this should work. But is it too verbose because you also need `#[cfg_attr(feature = "bytemuck), deny(unsafe)]`? I just feel deny is sufficient...