2012/5/18 Mathieu Blondel <[email protected]>:
> On Fri, May 18, 2012 at 6:39 AM, Ian Goodfellow <[email protected]>
> wrote:
>>
>> If the class label vector passed to the "fit" method of
>> OneVsRestClassifier has shape (n,1) instead of shape (n,),
>> then really weird results happen. In this example script here, when
>> the fit SVM is asked to predict the labels of
>> 1000 examples, the resulting prediction vector has shape (45000,1),
>> not (1000,) as expected.
>
> Thanks for reporting. I could reproduce the problem. So the "issue" seems to
> be in LabelBinarizer. To fix this, I suggest that we just raise an error in
> LabelBinarizer when the input to fit has 2d shape. Personaly, I'm not in
> favor of converting an array of shape (n_samples, 1) to (n_samples, ) behind
> the scenes. The reason is because those shapes have different semantics. If
> we start accepting arrays of shape (n_samples, 1), I would expect arrays of
> shape (n_samples, n_tasks) to work too (just like multivariate regression).

LabelBinarizer has the (undocumented) feature that it accepts an
indicator matrix as well as an array-like of labels and a list of
lists of labels, so checking for 1-d input might break things
elsewhere.

-- 
Lars Buitinck
Scientific programmer, ILPS
University of Amsterdam

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to