Add Lo splitter

[SteshinSS]

Hi, all. @cwognum suggested adding Lo-Hi splitter into Splito. Here is the first part of it.

Changelogs

• Add Lo splitter (from Lo-Hi splitter)

If the code seems ok, I'll add documentation and update this PR.

---

Checklist:

• [x] Was this PR discussed in an issue? It is recommended to first discuss a new feature into a GitHub issue before opening a PR.
• [x] Add tests to cover the fixed bug(s) or the newly introduced feature(s) (if appropriate).
• [x] Update the API documentation if a new function is added, or an existing one is deleted.
• [x] Write concise and explanatory changelogs above.
• [x] If possible, assign one of the following labels to the PR: feature, fix or test (or ask a maintainer to do it for you).

---

[SteshinSS]

Also, comment please on the code style. I didn't use any formatter yet.

[SteshinSS]

Thank you for the thorough review. The code is very complex, so any way to simplify it would be helpful. I'll return with the revision.

[SteshinSS]

Changes in the new commit:

• Added support for duplicates in the SMILES list.
• Changed the style of the code and docstrings.
• Used a logger instead of print statements.

Please review this version, and if it looks fine, I will include documentation in this PR.

[SteshinSS]

Thanks, guys, for the review! It seems that the PR has mostly converged. In this case, I'm going to add documentation to this PR.

Please check my old tutorial on the Lo splitter. I'm going to add more details, remove the k-fold examples, and rewrite the code using Splito and datamol datasets. What do you think about that? Is there anything that you would like to see in the documentation? Is there anything that could be clearer?

[cwognum]

Hi @SteshinSS, thanks for the nice work so far! As to the documentation, I would focus on the why rather than on the what. For a large part I find the code self-explanatory and won't need comments that explain me what's happening. Instead, I would like to understand why it has been implemented in a certain way and - zooming out - what the reasoning is behind this splitting method to begin with. You can of course always refer to your paper, but I would assume people have not read it.

[SteshinSS]

Done :)

[cwognum]

@SteshinSS Maybe worth saying that I didn't test the suggested changes above. It's from the top of my head.

[SteshinSS]

I just checked, and I'm not sure we want to add LoSplitter into train_test_split().

There are many splitters with a similar interface, accepting test_size and seed as parameters and returning an iterator. This function allows them to work in the same manner, which might be beneficial. However, LoSplitter:

• Does not return an iterator (because there are a lot of different clusters in the test set).
• Takes neither test_size nor seed.
• Accepts other arguments such as threshold, min_cluster_size, max_clusters, and std_threshold.

If we add LoSplitter to the train_test_split(), we will need:

• To raise an error if test_size or seed are passed, because users will expect these parameters to work.
• To handle LoSplitter as a special case in the return statement (because next(splitter.split(X, y)) won't work). In this case, the common interface isn't so common, and the utility of this umbrella function isn't clear.

As an alternative, train_test_split() could take **kwargs and pass them into splitters as _ = train_test_split(X, y, method, **kwargs). However, this approach kinda make the train_test_split() pointless because users could simply use splitters directly.

Have you considered a functional approach? For example, splito could offer not only classes but also pure functions for splitting, similar to torch.nn.functional

[cwognum]

I like the idea of a functional module! That's indeed maybe the better way to go. Thanks for the elaborate suggestion and considerations!

However, I don't want to scope creep this PR too much, so could you maybe open an issue with your idea and then we can merge this PR! Thanks again 🙏

[cwognum]

Please just make sure the code is properly formatted with Ruff!

[SteshinSS]

I believe it could be merged :)