2014-04-28 18:56 GMT+02:00 Jacob Schreiber <[email protected]>:
> I understand that HMMs do not perform classification in the same manner as
> SVMs or Random Forest, but why is it not desirable to create a new section
> to handle HMMs and possibly other graphical models? They seem like an
> extremely useful and widespread part of machine learning, and I know from
> personal experience that I'd prefer to have all my machine learning from the
> same source if possible.

Because we try to provide a unified API for the basic tasks in machine
learning, with pipelines and meta-algorithms like grid search to tie
everything together. The required concepts, APIs, algorithms and
expertise required for stuctured learning are different from what
scikit-learn has to offer. If we started doing arbitrary structured
learning, we'd need to redesign the whole package and the project
would likely collapse under its own weight.

That said, there are two project from scikit-learn contributors that
do structured prediction:

* pystruct [1] by Andreas, Vlad et al. handles general structured
learning (focuses on SSVMs on arbitrary graph structures with
approximate inference; defines the notion of sample as an instance of
the graph structure)
* seqlearn [2] by myself and others handles sequences only (focuses on
exact inference; has HMMs, but mostly for the sake of completeness;
treats a feature vector as a sample and uses an offset encoding for
the dependencies between feature vectors)

AFAIK, neither has solved the problem of putting a structured output
learner at the end of a Pipeline, or putting one inside GridSearchCV.

[1] http://pystruct.github.io/
[2] http://larsmans.github.io/seqlearn/

------------------------------------------------------------------------------
"Accelerate Dev Cycles with Automated Cross-Browser Testing - For FREE
Instantly run your Selenium tests across 300+ browser/OS combos.  Get 
unparalleled scalability from the best Selenium testing platform available.
Simple to use. Nothing to install. Get started now for free."
http://p.sf.net/sfu/SauceLabs
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to