Closed GregoryMorse closed 4 years ago
I do not think that adding a classifier calls for a SLEP: SLEPs are for modifications that touch many classifiers.
It should rather be discussed in an issue in the scikit-learn repo.
However, I am not sure that the proposed algorithm meet the criteria for inclusion (listed on https://scikit-learn.org/stable/faq.html#what-are-the-inclusion-criteria-for-new-algorithms), indeed, a Google search shows only a few dozen citations: https://scholar.google.ca/scholar?q=Accelerated+algorithm+for+pattern+detection+in+logical+analysis+of+data&hl=en&as_sdt=0&as_vis=1&oi=scholart
A scikit-learn-contrib package maybe.
Okay there is definetely good information here. But this algorithm might be worth it. Take for example a more influential paper. That paper I cited is particularly useful for finding patterns efficiently but for the overall process: https://scholar.google.ca/scholar?hl=en&as_sdt=0%2C5&as_vis=1&q=An+implementation+of+logical+analysis+of+data&btnG=
An implementation of logical analysis of data E Boros, PL Hammer, T Ibaraki, A Kogan Cited by 430
At least I should focus on making a supporting argument based on the right paper, along with other requirements from the FAQ. I was also unsure if a SLEP was the right place for the discussion. For now probably a contrib package is the right direction and it can be considered for the main library thereafter. Thanks to both of you for your guidance on this.
The only real question is wide use after studying the FAQ. But I would consider that its wide use is partly because public open source implementations are not available. Research demonstrates very good results with it.
Which sounds like a good reason for it to be implemented in contrib initially
Also, it may be easier to have it in sklearn-extra instead of a stand-alone package.
See the SLEP for details.