l] Use SciPy optimization instead of
> brute force GridSearch
> To: [email protected]
> Message-ID:
> [email protected]>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Interesting to see this
Interesting to see this issue get some attention again.
I'm hoping to develop sklearn wrappers for hyperopt this summer. If anyone
wants to be part of that development process, by all means let me know.
I'll keep this list posted on progress and listen to feedback on usability.
The quick wrapper I
I've hacked together something like this on top of a refactored
BaseSearchCV (building on my PR#1787's param_search_struct_arrays branch
rather than master):
https://github.com/jnothman/scikit-learn/tree/param_search_callback.
Unlike the current parameter searches require the entire sequence of
pa
I have made a few attempts in these directions (for instance using a
Nelder Mead optimizer). However, it is quite hard to get an optimizer
that does not get stuck in local minima, given that there is a lot of
noise, and flat regions.
Jame Bergstra has put a lot of intelligence in his HyperOpt. Whi
Currently BaseSearchCV expects a predetermined sequence of parameter
settings, which is unideal for some cases. SciPy opts for a callback
approach. I've not used that interface, but I gather something like this
might work:
class MinimizeCV(BaseEstimator):
def __init__(self, minimiser, clf, par
Oops. I figured someone must have thought of this already.
On Sun, Apr 7, 2013 at 6:40 PM, Ronnie Ghose wrote:
> ... can we have an open issue on this? this has now been addressed quite a
> few times ~ one of the repetitions being by me i confess.
>
> roman -- hyperopt: https://github.com/jaber
... can we have an open issue on this? this has now been addressed quite a
few times ~ one of the repetitions being by me i confess.
roman -- hyperopt: https://github.com/jaberg/hyperopt
On Sun, Apr 7, 2013 at 6:35 PM, Roman Sinayev wrote:
> It seems like brute force Grid Search takes forever
It seems like brute force Grid Search takes forever when attempting to
determine best parameters with many classifiers. Let's say the
parameter space looks something like this
http://i.imgur.com/AiBl8Wt.png . Why not use the SciPy simulated
annealing or some simple genetic algorithm instead of se
