Series.row_index/1

[iurimateus]

Interesting. So what if we:

1. Add a Series.row_index()
2. Make it raise for the default backend saying it is only supported in dataframes
3. Implement it properly for expressions

Would that work? Perhaps we try that as a separate PR so we can compare them side by side?

Originally posted by @josevalim in #833 (comment)

@josevalim is this what you had in mind? I think it's the first /0 function in Series; I'm also not sure on the implementation.

Should we rename len() to size() for consistency?

cc @cigrainger

[josevalim]

Thank you @iurimateus. Quick question: do we need to use expr_len? couldn't we use size(...) and expect the series as argument for row_index?

[billylanchantin]

@iurimateus I think expr_len is already present, it's just named expr_size:

https://github.com/elixir-explorer/explorer/blob/main/native/explorer/src/expressions.rs#L576-L580

[iurimateus]

For the record, these changes were before we had size.

@iurimateus I think expr_len is already present, it's just named expr_size:

Thanks for pointing it out, I was following py-polars implementation of using dsl::len(), but that doesn't seem necessary (and not really compatible with Explorer's API, on which it doesn't expose 'expressions').

Thank you @iurimateus. Quick question: do we need to use expr_len? couldn't we use size(...) and expect the series as argument for row_index?

That works, I've pushed a new commit.

Thoughts on int_range?

[josevalim]

I would prefer to postpone int_range for now. So let's remove it from the Elixir side but we can still keep it on the Rust side.

We also need to list row_index as a callback here:

explorer/lib/explorer/backend/series.ex

Line 41 in b901d97

@callback categories(s) :: s

And then provide an implementation for regular series (which will basically return a list of u32).

[josevalim]

Let's add docs here. :)

[josevalim]

I wonder if we should expose this function in Rust instead? https://docs.rs/polars/latest/polars/datatypes/struct.UInt32Type.html#method.new 🤔 would it have any performance difference?

[iurimateus]

(And perhaps new(lazy: true), but I'm not sure about allocations...)

Thanks for the reference, I've pushed a commit exposing it in Rust.
I'll rebase after review.

[josevalim]

Awesome job. I have pushed one last comment and I think we are good to go!

[josevalim]

💚 💙 💜 💛 ❤️