Inclusion in scikit-mine

[remiadon]

Hi there, very nice and rich implementation of these 3 algorithms

The INRIA center at Rennes is creating a new python library, namely scikit-mine, to centralise pattern mining methods, and improve inter-operability and consistency with other fields, such as Machine Learning.

Your API already has similarities with what scikit-mine provides, being:

• an implementation of LCM for closed frequent itemsets (equivalent to BIDE, but for regular itemsets)
• custom callbacks for richer exploration

In the context of scikit-mine, only BIDE and FEAT would be nice to have, as PrefixSpan mines too many patterns, and we encourage concise representations.

I also plan to try FEAT as a candidate generator for SQS-candidates, an algorithm based on MDL.
To this purpose handling gaps would be required, as SQS natively accounts them when running its optimization process

Anyone to provide support for integration into scikit-mine ?

[chuanconggao]

Hi, it is an interesting idea.

As the maintainer, I may provide some help but it requires some effort estimation. If there any planning or scope so far?

[remiadon]

@chuanconggao thanks for responding,

I think we can start with something "simple"

1. integrating BIDE, w.r.t functional definition in skmine
2. unit tests for BIDE
3. adding relevant example / doc for it

Rough estimation for this would be 2 weeks (my side) with daily feedback from you

Once this is done, I would add another task

• make sure the skmine.callback API is consistent with your callback API (we can treat similar use cases with both)

[chuanconggao]

Sounds good.

On my side, I can start some refactoring to make it more robust.

[remiadon]

@chuanconggao I wrapped your code and added a few unit tests in the PR mentioned above

What sort of refactoring were you thinking of ?

My implem is still missing the top-k part though ... which is interesting to have

[chuanconggao]

The refactoring part is mostly docstring, typing, and bug fixes.