💡 [FEAT] - Scikit-Lexicographical-Trees into a Longitudinal Splitters Easier Library

[simonprovost]

Before Submitting

• I have checked if this feature already exists or has been requested before.
• This feature aligns with the goals and scope of Scikit-Longitudinal.

What's the Problem?

Following the great contribution by @MathiasValla to Sklong/Scikit-Lexicographical-Trees, I believe we should entirely rework Scikit-Lexicographical Trees so that it becomes a sklearn-style fork where new splitters and their parameters can be added easily without affecting the others.

For instance, right now, threshold_gain is used as gamma for TpT — can you confirm this, @MathiasValla? This may have made development more straightforward initially, but adding new hyperparameters should be much easier after a full refactor.

As a result, the name could become something along the lines of Scikit-Learn-Longitudinal-Splitters, and Scikit-Lexicographical-Trees might be removed in favour of this new package.

• Simplification of build
• Add unit tests to PRs
• Automation of release tags
• Easy to add new splitters without new hyper parameters
• Easy to add new splitters with new hyper parameters
• Brand new documentation to present each and one of the splitterrs
• //TO ADD MORE REQUEST IN COMMENT

Cheers

The Dream Solution

N/A

Alternatives

N/A

Context & Relevance

N/A

Screenshots or Additional Material

N/A

Willing to Contribute?

Yes

[MathiasValla]

That is just great !

I confirm that right now, threshold_gain is used as gamma for TpT. We can easily make that change in the future, since I intend to add new penalty forms for the splitting criterion, some of which requires multiple parameters. So many things that could be done, but so little time to work on it.

It may or may not be of interest in the end, but I'm working on 2 new splitters for longitudinal tree-based methods, I'll be happy to try to implement it once Scikit-Lexicographical Trees is reworked !

Cheers

[simonprovost]

Of course! More than needed! I'm trying to finish writing my thesis and over my post-doc which will be in Longitudinal ML, I'll have to work on that so that we can welcome all of your nice discoveries! Let's make Sklong a hub for us all 🙏

BTW @MathiasValla — I am unsure if you decided to, but I did not receive any email to potential co-author on one draft? Once again, no need if not possible, I am just curious as you mentioned that at some point :)

Cheers

[MathiasValla]

Simon,

Yes, I did work on an article, including you as co-author ! I was waiting for TpT's integration to go on with that, but I can already share emelents with you.

I'll do it tomorrow, I'll have some time for that ! It's very (very) preliminary, but since it's a software paper, most of the work is the implementation, so I'm not too worried about the fact that we'll be able to make it better rather quickly.

[simonprovost]

Simon,

Yes, I did work on an article, including you as co-author ! I was waiting for TpT's integration to go on with that, but I can already share emelents with you.

I'll do it tomorrow, I'll have some time for that ! It's very (very) preliminary, but since it's a software paper, most of the work is the implementation, so I'm not too worried about the fact that we'll be able to make it better rather quickly.

magical! count on me then :)