On 02/17/2015 03:50 PM, Pagliari, Roberto wrote:
> I see. But in my case I have things like
>
> for i in xrange(0, num_experiments):
> estimators = # some pipeline
> etc...
>
> so estimators should be "re-created" everytime.
What do you mean by re-created? And where? if you construct your
pipeline for each experiment,
they are clearly independent. But according to the sklearn API you don't
need to. "fit" will always forget any previous model.
What happens inside GridSearchCV and cross_val_score is:
for train, test in cv:
this_estimator = clone(estimator)
this_estimator.set_params(**params)
this_estimator.fit(....)
Here "clone" takes all the parameters given by "get_params" and creates
a new object that has the same parameters.
What is the issue you are having?
Also, did you read through the developer documentation on how to roll
your own estimator
(http://scikit-learn.org/dev/developers/index.html#rolling-your-own-estimator)?
If we don't explain this clearly there, we have to work on that.
>
> In any case, I recently changed my code and I am no longer using lists.
>
> With scalar numbers I think it is working and I see the correct number all
> the time and from both __init__ and fit().
What do you mean by "the correct number"?
>
> I'm not sure if sklearn is supposed to support lists as parameters. But for
> now I'm no longer using lists .
Yes, it does.
------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=190641631&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
