Allow multiple hatch colors in same figure

In version 2.0 the hatching defaults have changed. In particular, the color is always black by default and it is now inconvenient to restore the old way in which the colors are set (based on the artist line color at least for fill_between).

I'd like to request a hatchcolor parameter for the functions that accept hatches.

To make matters worse, it looks to me that the hatch color is hardcoded in the backend in such a way that it's not possible to have more than one hatch color. Therefore I don't know how to produce a figure like this with mpl2:
http://pcteserver.mi.infn.it/~nnpdf/validphys-reports/X-vnmR5BSRygX3qoK9NtFQ==/figures/norm_plot_pdfs_bard.pdf

#7059 is roughly related

I've also encountered the same problem. It seems that PR #7415 broke what I believe should be a supported functionality, i.e. using hatching with different colors within one figure. I want to add further support for the Zaharid's suggestion of adding a kw argument for addressing this issue.

Otherwise I think it might even be preferable to revert the PR. There is a slight gain of convenience from being able to set a global default hatching color, but a net loss in functionality. (Plotting with hatch color != edgecolor could also be achieved before by plotting the artist twice once with hatching and once without.)

That #7415 fixed #7166 which was hatches not showing up in the legend so it can not be reverted.

The fix for 2.0.1 could be:

add a set_hatch_color method to gc base with semantics of None -> get the rcParam, else use the color
add private _hatchcolor to the artists that know about hatches
when we set edge color also set _hatchcolor
ever place with call set_hatch we also call set_hatch_color

This will give the behavior of:

by default use the rcParam
if user explicitly set edge color use it for hatch

I want to hold off adding the hachcolor kwarg until 2.1, as that also implies adding get_hatchcolor/set_hatchcolor which is a bit much for a micro-release. There would also need to be a couple decisions made for how to 'latch' 'color', 'edgecolor', and 'hatchcolor' together

I fully endorse this request to have an interface to the hatch's color, linestyle, linewidth, alpha etc. The feature to control those parameters has been present in previous versions (at least color and linewidth) and it was a big mistake to change this.

Let me elaborate a bit on the necessity of controling the hatch.
Historically hatch has been mainly used to provide an option to distinguish shapes in black/white documents, which makes it somehow understandable to tie it to the backend and hardcode the properties.
However, today most information is shared online, viewable in color and expectations for visually appealing figures is ever growing. On the other hand, figures are still printed in black and white and there is a growing awareness of the problems of limited color vision, such that at least in some areas institutions are forced to provide their information in an accessible way. As I understand it, some of the features introduced with matplotlib 2.0 were highly motivated by those requirements.
How does this relate to hatching? Hatching makes it incredibly easy to produce visually appealing and yet accessible figures.

Another point is of course that different styled hatching allows to add a new dimension to a figure. What is linecolor, linestyle and linewidth in 1D is patchcolor and hatch in 2D. Controlling the hatch is therefore utterly important.

@ImportanceOfBeingErnest

@ImportanceOfBeingErnest Can you open a new issue with that? That sort of effect maybe beyond the scope of what the hatching can currently do (or maybe we only need linewidth?). I think a MEP (https://github.com/matplotlib/matplotlib/tree/master/doc/devel/MEP) proposing an API for what this would look like and how to implement it across all of the backends would be well received. Given that you have clearly thought about this, can you take a crack at that?

@Zaharid

I don't think we need a new issue here, it is exactly what @Zaharid is asking for:
Getting control over hatchcolor and hatchlinewidth.
I'm saying that extending this to include alpha and hatch density would be good.

The plot from above can actually be reproduced in matplotlib 2.0 in a very hacky way. The only thing really missing here is the ability to change hatchcolor, hatchlinewidth and density via the api.

import matplotlib.pyplot as plt
import matplotlib.hatch
plt.rcParams['hatch.color'] = '#aa007d'
plt.rcParams['hatch.linewidth'] = 25

class MyHatch(matplotlib.hatch.NorthEastHatch):
    def __init__(self, hatch, density):
        # chose some parameter that give desired result
        self.num_lines = 2 
        self.num_vertices = 6

matplotlib.hatch._hatch_types.append(MyHatch)

plt.bar([1], [3], color="#000177", edgecolor="#000177",  linewidth=10)
plt.bar([2], [2], color=None, edgecolor="#770058",  linewidth=10)
# using the fact, that if hatch is not recognized, last list item is chosen.
plt.bar([2], [2], color="#770058", hatch="unknown" )

plt.savefig(__file__+".png")
plt.show()

Fair, I had not grasped how useful the hatch.linewidth parameter was or hacking at the underyling hatch mechanism like that. I have been working on this project for ~4 years and I still am learning more details about how it works :-p

However, your comment about how hatch selection works is not quite correct, see https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/hatch.py#L194 . It creates hatch objects for all of the hatch types it know about and merges them. It happens that MyHatch does not correctly return a null-hatch when it should. Also, your example does not work for me

I would like to significantly narrow the scope of this issue to be fixing the no-multi-color-hatch regression in 2.0 and am asking for a new issue to work through all of the hatch-related knobs that will need to be added and which bits of configuration that are currently hard-coded or global parameters need to be pulled up to artist-level configuration.

Just to chime in, there is a nice example of setting per-patch hatching color here under the 'Colors' section, which worked in 1.5 but doesn't in 2.0 .

As others have pointed out, if you want a plot that works both in color and in black and white it's critical to be able to adjust the hatch color for each patch (and eventually all sorts of other things too like width and spacing etc)

@mangecoeur

@mangecoeur This should work again in 2.0.1 (fixed by #7976)

(that is when we release 2.0.1)

Hi. What is preferred way to achieve colored hatch currently? Currently my graph looks like:

From reading above, it looks like:

colored hatching used to work, in some versoin in the past, and
colored hatching works again, in not-yet-released 2.0.1

Should I better:

rollback to an older version of matplotlib (which version?), eg using pip? , or
somehow install matplotlib from github master branch?

(Ok, I rolled back to 1.5.3. That has its own issues, ie bar positioning, but fixes the hatch colors:

(edit: and a quick align='center' fixes the positioning change:

@hughperkins

@hughperkins Your understanding is correct and those are the two options. 2.0.1 should be very soon.

Hi,
I've been dabbling with the changes to hatching as of 2.0.2. While we now have much better control of the hatch colour (Thanks!). Currently, the only way I can find to remove the background colour and allow different coloured hatching is to set color = None, alpha = 0 in the fill_between args.

The problem with this is that the legends become useless. Is there something we can use instead to mimic this behaviour without using alpha?

   # Basic example
from matplotlib import pyplot as plt
import numpy as np

h1 = '//'
d1 = np.random.rand(130)
lab1 = 'Rand1'
h2 = '\\\\'
d2 = np.random.rand(200)
lab2 = 'Rand2'

fig, ax = plt.subplots(1)

def plt_hist(axis, data, hatch, label):
	counts, edges = np.histogram(data, bins=int(len(data)**.5))
	edges = np.repeat(edges, 2)
	hist = np.hstack((0, np.repeat(counts, 2), 0))

	outline, = ax.plot(edges,hist,linewidth=1.3)		
	axis.fill_between(edges,hist,0,
				edgecolor=outline.get_color(), hatch = hatch, label=label, 
				color = None, alpha = 0)  ## < removes facecolor

plt_hist(ax,d1,h1,lab1)
plt_hist(ax,d2,h2,lab2)
ax.legend()
ax.set_ylim(0,ax.get_ylim()[1])

Use facecolor='none'; don't mess with alpha.

Ugh.... Thanks. I had futilely tried facecolor = None

(Ok, I rolled back to 1.5.3. That has its own issues, ie bar positioning, but fixes the hatch colors:

(edit: and a quick align='center' fixes the positioning change:

Can you please help in this plot to generate the legends where edgecolor and hatch color are different in the unfilled bar plot?

@kamraaan1997

@kamraaan1997 Please open a new issue with a minimal example that shows a bug or if you are looking for support on how to use Matplotlib please post your question to https://discourse.matplotlib.org . Thank you!

Navigation Menu

Repository navigation

Allow multiple hatch colors in same figure #7901

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Allow multiple hatch colors in same figure #7901

Description

Activity

Zaharid commented on Jan 23, 2017

tacaswell commented on Jan 26, 2017

andim commented on Jan 28, 2017

tacaswell commented on Jan 28, 2017

ImportanceOfBeingErnest commented on Jan 30, 2017

tacaswell commented on Feb 1, 2017

ImportanceOfBeingErnest commented on Feb 1, 2017

tacaswell commented on Feb 2, 2017

mangecoeur commented on Feb 14, 2017

tacaswell commented on Feb 14, 2017

tacaswell commented on Feb 14, 2017

hughperkins commented on Apr 1, 2017

hughperkins commented on Apr 1, 2017

tacaswell commented on Apr 2, 2017

kasekun commented on May 20, 2017

QuLogic commented on May 20, 2017

kasekun commented on May 20, 2017

kamraaan1997 commented on Apr 11, 2022

tacaswell commented on Apr 11, 2022

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions