@efiring @mdboom Comments on this? I would like to get feed back on the handling of mid + where before I write tests (which will probably just be the above examples).

Yes, the "where" kwarg is confusing and surprising. First, I would expect "where" to be used in a way that matches its use in step. Second, your example of a "where" kwarg has the surprising result that a single zero knocks out two blocks of fill; presumably this means there is no way to knock out a single block. Maybe "where" could replace "step_where", and then a "skip" could replace "where"; "skip" would be like a mask (True is masked, so skip the point). Then the "mid" case is easy--there is one color block per point. For "pre" and "post" probably the simplest thing would be to require and use only n-1 values, corresponding to the filled blocks.

Are masked arrays also supported?

How about kwarg "step" for "mid, pre, post", and "skip" as described above?

where has been in the API as long as I remember and knocking out two squares is consistent with the current fill_between behaviour.

import matplotlib.pyplot as plt
import numpy as np

x = y = np.arange(5)

fig, ax_list = plt. subplots(3, 1)

for ax, where in zip(ax_list, ['pre', 'post', 'mid']):
    ax.step(x, y, where=where, color='r', zorder=5, lw=5)
    ax.step(x, -y, where=where, color='k', zorder=5, lw=5)
    ax.fill_between(x, y, -y, where=[1, 1 ,0, 1, 1])

I suspect that ma works (I was thinking about that when I wrote it), but have not tested.

OK, we are stuck with "where", which is like the inverse of the mask if the y values were a masked array. Nevertheless, I'm not sure that keeping full consistency with the non-step behavior is a good idea. What will people actually want to do? Will people want to be able to knock out a single block? I've never used "where", so I don't have use cases in mind. I'm not necessarily opposed to your consistency argument--just trying to anticipate complaints.

As for "step_where": having the "where" in that kwarg name makes it unnecessarily confusing, given that there is a plain "where" kwarg as well. Is there any reason not to just call it "step", or use "linestyle" or "style"?

I am sold on changing step_where -> step. I will try to get to that
tonight.

I think the use-case for fill_between is skewed towards large numbers of
points ex ax.fill_between(x, y1, y2, where=y2 > .5) with len(x) >>100 in
which case taking out two bins is not a huge deal.

On Tue, May 19, 2015 at 1:08 PM Eric Firing [email protected]
wrote:

OK, we are stuck with "where", which is like the inverse of the mask if
the y values were a masked array. Nevertheless, I'm not sure that keeping
full consistency with the non-step behavior is a good idea. What will
people actually want to do? Will people want to be able to knock out a
single block? I've never used "where", so I don't have use cases in mind.
I'm not necessarily opposed to your consistency argument--just trying to
anticipate complaints.

As for "step_where": having the "where" in that kwarg name makes it
unnecessarily confusing, given that there is a plain "where" kwarg as well.
Is there any reason not to just call it "step", or use "linestyle" or
"style"?

—
Reply to this email directly or view it on GitHub
#4433 (comment)
.

This just needs a whats new entry before this gets merged, right?

Added what's new and example.

This also has some of my ideas on how to deal with labeled data and how to start fixing hist in general.

rebased to deal with conflicts in .travis.yaml

Apart from the docstring tweaks, I think this is ready to go, right?

Yes, I'll take care of that now (I wrote some fo those examples at stupid oclock in the morning due to my sleep schedule being very messed up after scipy, I am surprised any of it is in English)

closes #643

The test failure is from the doc build timing out which has been fixed on master, can this be merged?