Thank for the pull requestion.
That looks good.
+1 for merge apart from @mblondel's comment.

For the future I think it is good if the titile of the pull request is a bit more descriptive than just the issue number. These are a bit hard to remember ;)

Oh, and can you please also a test for the new functionality?

Is it possible to wait a little before merging?
The pr #1512 is almost finished. You would have the possibility to update the narrative doc.

By the way, an example would be nice in the docstring.

appart from above comments, 👍 for merge

@kyleabeauchamp As we discussed in #1512, don't wait for my pr . I'll update the narrative doc.

For the example, you can use:

Examples
--------
>>> from sklearn.metrics import zero_one
>>> y_pred = [0, 2, 1, 3, 4]
>>> y_true = [0, 1, 2, 3, 4]
>>> zero_one(y_true, y_pred)
2
>>> zero_one(y_true, y_pred, normalize=True)
0.4

For the testing part, you can add and adapt existing tests in test_metrics.py.

When you have finished, add your contribution in doc/whats_new.rst.

Thanks for the pr. :-)

OK I think I integrated everyone's comments. I think it's ready to go.

One note--the following doctest didn't work as written because of floating point representation:

zero_one(y_true, y_pred, normalize=True)
0.4

I did some rounding to avoid FP ambiguity.

instead of rounding, you should use ... in doctests and add the ellipsis flag (git grep DOCTESTS or something, I can never remember the exact syntax)

Are you sure this is the best approach? The python docs (http://docs.python.org/2/library/doctest.html) seem to suggest that rounding is the best way to go. My concern with ELLPISIS is that it's a general regex capability, while rounding allows us to directly control the level of precision required.

I would use rounding for tests and ellipsis for doctests, since in doctests the prime concern is for the example to be clear to the user, and using round might be confusing and someone unfamiliar might think you somehow need to round everytime, or something like that. Just my 2p.

This is the usual way it is done in sklearn.
We don't really rely on doctests for tests, they are just examples that we know that work.
And examples are easier to read without a call to round imho.
Btw, you can still directly controll the required precion.

@amueller we cross-posted... at least we agreed :)

@vene yeah, saw that. I take that as a good sign ;)

OK I agree that ELLIPSE is prettier for users.

Instead of having round or ellipse, it might be better to change the example.
Just a quick ipython session:

In [1]: import numpy as np

In [2]: y_true = np.array([1, 2, 3, 4])

In [3]: y_pred = np.array([1, 3, 2 , 4])

In [4]: np.mean(y_true != y_pred)
Out[4]: 0.5

In [5]: np.sum(y_true != y_pred)
Out[5]: 2

sure that also works ;)

I tend to agree with the comment of @vene on the sentence formulation.
But if @vene is ok, then it is ok for me. :-) Ok for you @vene?

Except from the above comment, +1 for merge.

Thanks for your quick responses @kyleabeauchamp !!!

Looks good. +1 for merging as well.

crazy amount of reviewers out there tonight. Sweet :) will merge by rebase! (pretty sure @vene won't mind ;)

merged

Thanks a lot @kyleabeauchamp, in particular for the immediate responses! 👍

@kyleabeauchamp When you added yourself to whatsnew.rst, you made your name a link, but didn't add a website. I removed the link for now. Give me a shout if you want to add your website.

Oops--sorry about that, I copy pasted a previous line on whatsnew.rst
and copied my name in there. I didn't fully realize what the formatting
was doing. Thanks.

On 01/16/2013 01:38 PM, Andreas Mueller wrote:

@kyleabeauchamp https://github.com/kyleabeauchamp When you added
yourself to whatsnew.rst, you made your name a link, but didn't add a
website. I removed the link for now. Give me a shout if you want to
add your website.

—
Reply to this email directly or view it on GitHub
#1534 (comment).

On 01/16/2013 10:41 PM, kyleabeauchamp wrote:

Oops--sorry about that, I copy pasted a previous line on whatsnew.rst
and copied my name in there. I didn't fully realize what the formatting
was doing. Thanks.
I guessed it was something like that ;)
No problem. Just wanted to give you the opportunity to add a link if you
like.