On 03/10/2013 12:01 PM, Joel Nothman wrote:
> Firstly thank you to the devs for a great toolkit.
>
> I am using sklearn's GridSearchCV for a classification with the F1 
> metric. GridSearchCV.fit() produces a .cv_scores_ attribute which 
> allows me to view the scores for each fold for each point in the grid. 
> But it does not let me view the precision and recall for each fold, or 
> even overall for a grid point, from which F1 is calculated.
>
Is there a particular reason why you don't use the f1 score for doing 
the grid search? If this is your ultimate goal, I don't see a reason to 
use accuracy.
In the current stable, you can use ``score_func=f1_score``, in the 
current developer version you can use ``scoring='f1'``.

> I can see two ways to make this data available:
> * allow the user to provide an arbitrary diagnostics function which is 
> run on each fold's predictions and whose output is stored with 
> cv_scores_ (or one could even store the learnt parameters for each fold)
This should be possible in the current developer version with the 
introduction of the ``scoring`` parameter.

We deemed storing all models infeasible because of the large memory 
requirement of some models.  We could optionally reintroduce it.

Hth,
Andy

------------------------------------------------------------------------------
Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester  
Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the  
endpoint security space. For insight on selecting the right partner to 
tackle endpoint security challenges, access the full report. 
http://p.sf.net/sfu/symantec-dev2dev
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to