Incorrect behaviour of numpy.histogram

[Elby]

The behavior of numpy.histogram is not consistent with its doc string :

• doc string says that, with a range argument, values outside of this range are allocated to the closest bin.
• in fact values below the range are simply ignored.

There was a discussion on this subject on the scipy.user mailing list: http://groups.google.com/group/scipy-user/browse_frm/thread/3b3166e2200f846b/d6040fb6b659c6dd?hl=fr&lnk=gst&q=histogram#d6040fb6b659c6dd

IMHO, the current behavior of numpy.histogram, that is assuming that values below the range are outliers, is not what a neophyte is waiting for, and should be clearly stated.

Besides, the user should have the possibility to choose what to do with values outside the range : just ignoring them is not a good idea in most of the case I've seen.

[dhuard]

A new histogram function (breakage warning).

[dhuard]

The patch makes sure that:

• outliers are not counted.
• normalization is done correctly, that is, sum(hist*diff(bins))==1.

Also, I added support for weights for good measure.

It breaks the current behavior in the following ways:

• returns the bin edges (nbins+1) instead of the left bin edges.
• upper outliers are not put in the closest bin.

[pv]

I think this was fixed in r5085, r5086, r5087, r5088 by dhuard

[dhuard]

It's fixed, but I'd like someone to review what I've done.

[peridot]

Replying to dhuard:

It's fixed, but I'd like someone to review what I've done.

Looks good to me; the docstring doesn't mention that you can use a sequence of left bin edges and new=False, but maybe that's just as well.

[dhuard]

This can be closed. The current ticket for the follow up on the histogram change is #797.

[charris]

Closed as requested. See #797 for followup.