Apriori/one hot speedup

[jaksmid]

On my computer, the Apriori algorithm could not finish in reasonable time with the document containing millions of rows.
I optimized the code so the algorithm runs much faster

• Transform: One hot is much faster
• Apriori is much faster - some invalid combinations omitted

[pep8speaks]

Hello @jaksmid! Thanks for updating the PR.

Cheers ! There are no PEP8 issues in this Pull Request. 🍻

Comment last updated on February 12, 2018 at 14:18 Hours UTC

[coveralls]

Coverage increased (+0.02%) to 91.562% when pulling 3f0ae0d on jaksmid:master into ebc692b on rasbt:master.

[rasbt]

This is great, thanks a lot!

Yeah, omitting items/itemsets without enough support in the previous step actually makes sense.

Not sure, but do you have "allow maintainer edits" enabled on this PR? (I wish GitHub would add a visual clue for that). I wanted to adjust the documentation a bit for consistency.

[jaksmid]

Hello @rasbt,
glad that you like my PR.

I see allow edits from maintainers as checked. But if you send me what you want to have changed I will be happy to incorporate the amendments for you.

[rasbt]

Thanks! There are actually only two little things that come to mind:

• it's not included in the API doc because it's an internal function, but could you change the docstring of "generate_new_combinations" to the NumPy-style docstring format for consistency with other docstrings?

• An entry in the Changelog (mlxtend/docs/sources/CHANGELOG.md) would be nice. I just see that there's no category for "improvements" but I think the "Changes" category should suffice :)

[jaksmid]

Done. Is this something you had in mind @rasbt?

[rasbt]

Thanks for the update and the PR. Looks good to me now and I am happy to merge this :)