Data Complexity Library (DCoL) Metafeatures

[bjschoenfeld]

Taken from the README:

The data complexity library (DCoL) is a library implemented in C++ that provides the implementation of a set of measures designed to characterize the apparent complexity of data sets for supervised learning, which were originally proposed by Ho and Basu (2002).

https://github.com/nmacia/dcol

Do we wrap this C++ library or re-implement them in Python (or not use them at all)?

These metafeatures/library were suggested by @MichaelMMeskhi

[joaquinvanschoren]

Cool. If possible, it would be best to reimplement them in Python to allow future extensions.
I know a PhD student working on data complexity, I can ask whether he is interested.

[MichaelMMeskhi]

@joaquinvanschoren That would be better yes. There won't be any problem just the reimplementation would require good speed optimization to compensate what C++ offers thats all.

[bjschoenfeld]

There is also an R implementation called ECol by Luis Paolo which has some bug fixes from the original implementation.

[bjschoenfeld]

Here is the link to ECol. Luis Paulo, one of the authors is willing to discuss any bug fixes or updates.

It will be up to us to wrap or re-implement these metafeatures. It is not yet clear to me whether we should wrap or re-implement. Re-implementing is preferred in many ways, but is simply a matter of dedicating hours to it.

[MichaelMMeskhi]

Re-implementing is the way to go. Think long term.