MRG Class weight refactor #1464
Conversation
|
|
|
|
||
|
|
||
| def compute_class_weight(class_weight, classes, y): | ||
| """Estimate class weights for unbalanced datasets.""" |
There was a problem hiding this comment.
This docstring deserves to better describe what's happening for the different values of the class_weight argument.
There was a problem hiding this comment.
Good point. Docstring coming tonight.
|
Renamed to MRG. Should be good now. |
|
More lines added then deleted :( Well, at least it is consistent with the other estimators now. that has been bothering me for a while. |
| X, y = iris.data[:, :2], iris.target | ||
| unbalanced = np.delete(np.arange(y.size), np.where(y > 1)[0][::2]) | ||
|
|
||
| assert_true(np.argmax(_get_class_weight('auto', y[unbalanced])[0]) == 2) | ||
|
|
|
I built this branch and the tests pass. +1 for merging once my last comment is addressed. |
| List of the classes occuring in the data, as given by | ||
| ``np.unique(y_org)`` with ``y_org`` the original class labels. | ||
| y : array-like, shape=(n_samples,), dtype=int | ||
| Array of class-indices for 0 to n_classes. |
There was a problem hiding this comment.
This docstring is unclear. I think it should read
Array of class indices per sample; 0 <= y[i] < n_classes for i in range(n_samples).
|
@erg maybe you want to have a look as you complained so much about this ;) |
|
Rebased, changed the docstring. |
|
whoops, messed that one up... merged the wrong branch... hope no-one saw that ;) |
|
Ok so this is not merged but I can not reopen it. Great. |
Refactor
class_weightsfrom SGDClassifier and SVC, which enables the use ofclasses_in SVC.I put it in
utils.__init__, since I didn't know a where else to put it. Ideas welcome.Closes #745, #1037.