> Also David experienced poor performance compared to other
> implementation when using the CD models in a sparse coding. Would be
>
> You mean that the data matrix X has a lot of zero entries? There is a
> comment
> on this case in section 2.3 ( www.stanford.edu/~hastie/Papers/glmnet.pdf  ).

section 2.3 is for sparse X. There a sparse coordinate descent too which
btw does not support intercept if you feel brave :)

the poor performance can be improved by adding strong rules and simply
profiling eg making sure that data are Fortran-ordered when it's faster.

> great to ensure comparable performance with state of the art for this
> use case as well. Investigating with OpenMP via cython prange might be
> possible solution.
>
> I'm not sure if the algorithm is good to parallelize, I think there a other
> speed up tricks
> not used yet but I will look into it too.

good. Parallel computation can be particularly useful when dealing with many y's
ie Y with many regression tasks run in parallel as in sparse_encode
used in the dict learning code.

> Thanks for the suggestions, I start drafting the proposal as soon as I'm
> done with
> the patch.

perfect

Cheers,
A

------------------------------------------------------------------------------
This SF email is sponsosred by:
Try Windows Azure free for 90 days Click Here 
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to