To make it clearer, here's the current implementation's take on the example
from the other day:
>>> from sklearn import grid_search, linear_model, datasets
>>> iris = datasets.load_iris()
>>> clf = grid_search.GridSearchCV(linear_model.LogisticRegression(),
{'C': [1, 10]}, scoring='prf1')
>>> clf.fit(iris.data, iris.target == 1) # binary classification
...
>>> clf.cv_folds_
{'n_test_samples': array([[50, 50, 50],
[50, 50, 50]]),
'precision': array([[ 0.66666667, 0.54545455, 0. ],
[ 0.69230769, 0.53333333, 0.5 ]]),
'recall': array([[ 0.375 , 0.35294118, 0. ],
[ 0.5625 , 0.47058824, 0.23529412]]),
'score': array([[ 0.48 , 0.42857143, 0. ],
[ 0.62068966, 0.5 , 0.32 ]])}
>>> clf.cv_scores_
array([ 0.30285714, 0.48022989])
Actually, we might as well replace 'f1' scorer with 'prf1'...
The other thing I note is missing from the implementation is the index of
the best params within these arrays.
- Joel
On Tue, Mar 12, 2013 at 5:05 PM, Joel Nothman
<[email protected]>wrote:
> An implementation, without backwards compatibility or new tests (but with
> tests.test_grid_search modified to pass) at
> https://github.com/jnothman/scikit-learn/tree/cv-enhanced-results (
> c9d45a3444<https://github.com/jnothman/scikit-learn/commit/c9d45a3444e6474901dca13eaeaef83b708bd969>).
> I hope you like the general idea; I'm not sure you'll like the current
> attribute names...
>
------------------------------------------------------------------------------
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
