yes thanks a lot.
I was confused.

Are you aware of any default metric to measure how well two classes are
separated?


On Tue, Sep 8, 2015 at 6:27 PM Nicolas Goix <[email protected]> wrote:

> Hi Luca,
> The AUC score is 1 as soon as all the samples with label 0 have a score
> less than the minimum score of the samples with label 1.
> Hope this helps
> Nicolas
> On 8 Sep 2015 5:10 pm, "Luca Puggini" <[email protected]> wrote:
>
>> Hi,
>> I have a doubt regarding the AUC score.
>>
>> I would say that AUC should be 1 only if all the samples in class 0 have
>> score 0 and all the samples in class 1 have score 1.
>>
>> With the roc_auc_score function I get the value 1 for separable classes.
>> Isn't this wrong? Or maybe I am confused?
>>
>> x = np.arange(0, 1, .1)
>> y = np.array([0] * 7 + [1] * 3)
>> roc_auc_score(y, x) # = 1
>>
>> In this example if I classify 1 x>.3 than I do not have a 0 error. So I
>> think that auc should not be 1.
>>
>> Let me know.
>> thanks,
>> Luca
>> --
>>
>> Sent by mobile phone
>>
>>
>> ------------------------------------------------------------------------------
>>
>> _______________________________________________
>> Scikit-learn-general mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>>
>>
> ------------------------------------------------------------------------------
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
-- 

Sent by mobile phone
------------------------------------------------------------------------------
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to