Interesting!
I'm not sure if optimization algorithms are close enough to sklearn's focus
to be a GSoC. For Hyperopt, sure, but for sklearn I doubt it.
I'd be thrilled if key sklearn people would be open a GSoC project looking
at using e.g. hyperopt to do model selection in a way that's smart for
using various cross-validation schemes and regularization paths (where
available). The reason to put this sort of thing into sklearn rather than
leaving it as a separate project (e.g. hyperopt-sklearn) is that a model
search algorithm makes it possible to give a relatively simple interface
for new users to use ELM and NNet models. ELMs and NNet models are hard to
use without either hyperparameter expertise or an algorithm for taking care
of search.
After ELM and NNet models are offered with hyperparameter search as a
subroutine, then it would make sense to compare various search methods for
that job, such as the ones you listed.
- James
On Tue, Mar 11, 2014 at 11:50 AM, Marko Mitic <[email protected]> wrote:
> Hi James,
>
> Thanks for the quick response!
>
> I've already finished my writing, and is expected to defend the thesis in
> May. Therefore, my supervisor won't argue about the whole project, since
> the major part of it would be realized after the defense (and during the
> summer).
>
> I think that my first project in majority corresponds to your research. As
> far as I know, there exists only 2 evolutionary algorithms implemented in
> Python for black-box optimization (PSO and GA). I would like to extend that
> with other, more recent methods, such as Firefly Algorithm, Bat Algorithm,
> Differential Evolution, Harmony Search, etc. These algorithms could be used
> to train NNs, similarly to the two mentioned ones. Also, they can be used
> to solve different problems, highly possible in your hyperopt area. I don't
> have any knowledge on Bayesian optimization, or statistics, but I believe I
> could help.
>
> Thanks again,
> Marko
>
>
> 2014-03-11 15:49 GMT+01:00 James Bergstra <[email protected]>:
>
> Hi Marko,
>>
>> This is a good fit with the sort of thing I was hoping to mentor. I've
>> worked on Bayesian optimization / black-box optimization for neural
>> networks, and more recently, for optimizing over the algorithm
>> configuration of sklearn pipelines (
>> https://github.com/hyperopt/hyperopt-sklearn).
>>
>> What's your project idea?
>>
>> My one concern would be: shouldn't you be writing your thesis? (Is your
>> supervisor supportive of you taking 4 months to work on GSoC?)
>>
>> - James
>>
>>
>>
>>
>> On Tue, Mar 11, 2014 at 10:25 AM, Marko Mitic <[email protected]> wrote:
>>
>>> Hi all,
>>>
>>> My name is Marko Mitic, and I'm a final year PhD student at University
>>> of Belgrade, Serbia. I am willing to participate in GSoC, if I can find
>>> someone to mentor my work. My research interest are robotics and applied
>>> computational intelligence with emphasis on evolutionary algorithms. For
>>> purpose of finishing my PhD, I already implemented some
>>> evolutionary/metaheuristic algorithms in Python. Likewise, I have some
>>> experience using Extreme Learning Machines in Matlab, and would love to
>>> transfer the code to Python. I am proficient in Matlab, with solid
>>> knowledge of Python which I would like to improve further. I am willing to
>>> work 40+ hours a week on the GSoC project, starting 1st May 2014.
>>>
>>> Two main projects which I want to propose are:
>>>
>>> - Evolutionary (Metaheuristic) algorithms for function optimization
>>>
>>> - Extreme Learning Machines for Python users
>>>
>>> If there is someone finding these interesting, I would be delighted to
>>> explain in detail my ideas and project proposal. I am apologizing if its a
>>> little late for this whole procedure.
>>>
>>> Thanks in advance!
>>> Marko Mitic
>>>
>>> _______________________________________________
>>> Soc2014-general mailing list
>>> [email protected]
>>> https://mail.python.org/mailman/listinfo/soc2014-general
>>>
>>>
>>
>
------------------------------------------------------------------------------
Learn Graph Databases - Download FREE O'Reilly Book
"Graph Databases" is the definitive new guide to graph databases and their
applications. Written by three acclaimed leaders in the field,
this first edition is now available. Download your free book today!
http://p.sf.net/sfu/13534_NeoTech
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
