matplotlib-1.2.0: regression in histogram with barstacked drawing?

[mmokrejs]

To: Matplotlib-users@lists.sourceforge.net
Message-ID: 50DF638E.80904@fold.natur.cuni.cz

Hi,
I recently updated to matplotlib-1.2.0 from 1.1.1 and my figures have rotated
order of color bars stacked upon each other. Here is a small testcase I sketched
now and two generated png files are attached, generated by the two different
matplotlib versions. The colornames are just bad in the legend, just to show that
the order of input data is same in both cases.

I would be grateful for any comments on this.
Thank you,
Martin

! /usr/bin/python

import pylab

F = pylab.gcf()

F.set_size_inches(11.2, 8.4)

_e = [[100, 100, 100, 100, 200, 320, 450, 80, 20, 600, 310, 800], [20, 23, 50, 11, 100, 420], [120, 120, 120, 140, 140, 150, 180], [60, 60, 60, 60, 300, 300, 5, 5, 5, 5, 10, 300], [555, 555, 555, 30, 30, 30, 30, 30, 100, 100, 100, 100, 30, 30], [30, 30, 30, 30, 400, 400, 400, 400, 400, 400, 400, 400]]
_ee = [(0.5759849696758961, 1.0, 0.0), (0.0, 1.0, 0.350624650815206), (0.0, 1.0, 0.6549834156005998), (0.0, 0.6569064625276622, 1.0), (0.28302699607823545, 0.0, 1.0), (0.6849123462299822, 0.0, 1.0)]
_eee = ['green', 'orange', ' yellow', 'magenta', 'black']

pylab.hist(_e, bins=10, histtype='barstacked', align='mid', color=_ee, log=False, label=_eee)

pylab.legend(loc='upper right', bbox_to_anchor = (1.0, 1.0), ncol=1)

F.savefig("hist_barstacked_bug.png", dpi=100)

[mmokrejs]

[dmcdougall]

v1.1.1:

v1.2.0

[mdboom]

git bisect tells me the culprit is 7ba1743, but not sure exactly what the fix is yet.

[neggert]

I suspect you're correct. I'll take a look at it this weekend

[mmokrejs]

Michael Droettboom wrote:

git bisect tells me the culprit is 7ba1743 7ba1743, but not sure exactly what the fix is yet.

Hi,
thanks for your work on this. BTW, I also notice that also a barstacked chart with log Y-axis
is also affected. Probably same underlying issue.
Martin

[neggert]

Alright, I think I remember now what I did. Since the stacked histogram is done by putting later histograms over the top of previous one, the upper-most one needed to be plotted first.

This approach of layering the patches seems to be causing other problems as well (#1631), so I may need to revisit how this is implemented. I seem to remember having problems with it when using fill_between, as suggested there. I'll take another stab at it and if I get stuck I'll submit the PR so others can take a look.

[dmcdougall]

@neggert I think histograms can take a bottom= kwarg. Can you stack them that way instead?

[dmcdougall]

@neggert I think histograms can take a bottom= kwarg. Can you stack them that way instead?

Scratch that. I think I was thinking of Axes.bar().

[mdboom]

Closing since the relevant pull requests have been merged. Thanks!