Align x and y labels between axes #9652

jklymak · 2017-11-01T13:57:58Z

PR Summary

Aligns the x and y labels automatically at draw time.

Responds to windows resize.
works w/ tight_layout (and presumably [MRG] Constrained_layout (geometry manager) #9082)

Example:

import matplotlib
matplotlib.use('Qt5Agg')
import matplotlib.pyplot as plt
import numpy as np
import matplotlib.gridspec as gridspec

for after in [True, False]:
    fig = plt.figure(tight_layout=True)
    gs = gridspec.GridSpec(2, 2)

    ax = fig.add_subplot(gs[0,:])
    ax.plot(np.arange(0, 1e6, 1000))
    ax.set_ylabel('Test')
    for i in range(2):
        ax = fig.add_subplot(gs[1, i])
        ax.set_ylabel('Booooo')
        ax.set_xlabel('Hello')
        if i == 0:
            for tick in ax.get_xticklabels():
                tick.set_rotation(45)

    if after:
        # fig.align_ylabels()
        # fig.align_xlabels()
        fig.align_labels()  # do both x and y
        fig.savefig('After.png')
    else:
        fig.savefig('Before.png')

Before

After

PR Checklist

Has Pytest style unit tests
Code is PEP 8 compliant
New features are documented, with examples if plot related
Documentation is sphinx and numpydoc compliant
Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way
Check righthand/top labels
align_labels()
align titles: Done but Move title up if x-axis is on the top of the figure #9498 needs to be merged for it to work (or matter) ;-)

dstansby · 2017-11-01T16:44:01Z

Ooh, this looks nice. I also approve of having this as a function rather than another kwarg option.

anntzer · 2017-11-01T18:55:35Z

Very nice indeed, something I've always wanted to have.

One minor design point: I know that right now we tend to pass the renderer instance down the call stack, but AFAICT we really don't need to do that (we can just get it from artist.figure.canvas.renderer, and I don't think it makes sense to use another renderer anyways (but I may be wrong)). (I have somewhere a plan to remove all these renderers being explicitly shuffled through a bunch of functions, but that's a bit tricky to do due to backcompat :/)

jklymak · 2017-11-01T19:11:26Z

Well frankly it’s an attempt to head off calls for this to be in a geometry manager like #9082. You can do it but if you can do it more simply then why not.

@anntzer I’ll look at the renderer stuff. I was really just copying the _update_label_pos function that also passed the renderer so I assumed there were good reasons. I think there are cases where the figure canvas renderer is not defined so the wrapper in tight_layout is useful.

anntzer · 2017-11-01T19:22:56Z

As long as you keep renderer the last argument it remains possible to later deprecate it without too much pain so I don't mind keeping this discussion for later...

anntzer · 2017-11-01T22:06:29Z

examples/subplots_axes_and_figures/align_labels_demo.py

@@ -4,10 +4,10 @@
 ===============

 Aligning xlabel and ylabel using
-:func:`~matplotlib.figure.align_xlabels` and
-:func:`~matplotlib.figure.align_xlabels`
+:func:`matplotlib.figure.align_xlabels` and


for future reference, you don't need the :func: (or :meth: or :class:, etc.) here: the default role resolves any python object.

anntzer · 2017-11-01T22:09:53Z

doc/users/next_whats_new/2017-11-1_figure_align_labels.rst

+Subplot axes ``ylabels`` can not line up horizontally if the tick labels
+are very different widths.  The same can happen to ``xlabels`` if the
+ticklabels are rotated on one subplot (for instance).  The new methods
+on the ``Figure`` class: ``fig.align_xlabels()`` and ``fig.align_ylabels()``


rst syntax police here:

on the `Figure` class: `Figure.align_xlabels` and `Figure.align_ylabels`

or

on the `Figure` class: `~Figure.align_xlabels` and `~Figure.align_ylabels`

These are refering to things that have their own entries in the docs (the class and the method. The tilde is documented at http://www.sphinx-doc.org/en/stable/domains.html#role-py:obj and quite useful.

anntzer · 2017-11-01T22:10:34Z

examples/subplots_axes_and_figures/align_labels_demo.py

+===============
+
+Aligning xlabel and ylabel using
+:func:`matplotlib.figure.align_xlabels` and


see above re: syntax. the functions don't actually exist in the figure module.

anntzer · 2017-11-01T22:10:45Z

lib/matplotlib/axis.py

+    def _get_tick_boxes_siblings(self, renderer):
+        """
+        Get the bounding boxes for this axis and its sibblings
+        as set by `fig.align_xlabels` or  ``fig.align_ylables`.


anntzer · 2017-11-01T22:11:43Z

lib/matplotlib/figure.py

+        Choose the left-most label to align to.  Axes with the same subplot
+        column are aligned.
+
+        Parameters:


I'll channel the numpydoc police here.

Also seems normal to define align_xlabels first?

afvincent · 2017-11-02T17:42:02Z

examples/subplots_axes_and_figures/align_labels_demo.py

+    ax = fig.add_subplot(gs[1, i])
+    ax.plot(np.arange(1., 0., -0.1) * 2000., np.arange(1., 0., -0.1))
+    ax.set_ylabel('YLabel1 %d' % i)
+    ax.set_xlabel('XLaleb1 %d' % i)


"XLabel1" rather than "XLaleb1"?

PS: I am not an expert of this portion of the code but FWIW, I am very excited by such a functionality landing in Matplotlib :)

jklymak · 2017-11-02T19:03:26Z

squashed and rebased with master...

anntzer · 2017-11-02T19:14:24Z

@jklymak Can you please not include my name with the @ (@anntzer) in the commit message? everytime you force push the same commit message I get a notification :-)
Either just have anntzer, or push different commits and squash them before merging, whatever works for you.

jklymak · 2017-11-02T19:16:18Z

How annoying; my apologies!

jklymak · 2018-01-12T17:15:36Z

Rebase. Ping @anntzer, @dstansby, @afvincent who seemed interested; Pretty sure this is a non-breaking change. Happy to move to 3.0 if folks can't take a closer look next couple weeks.

anntzer · 2018-01-12T19:41:13Z

examples/subplots_axes_and_figures/align_labels_demo.py

+    if i == 0:
+        for tick in ax.get_xticklabels():
+            tick.set_rotation(55)
+fig.align_labels()  # same as fig.align_xlabels() and fig.align_ylabels()


replace and by (semicolon) (fig.align_xlabels() and fig.align_ylabels() is actually legal python but will only execute the first half :-))

anntzer · 2018-01-12T19:41:36Z

examples/subplots_axes_and_figures/align_labels_demo.py

+Aligning Labels
+===============
+
+Aligning xlabel and ylabel using


something wrong with the wrapping her

anntzer · 2018-01-12T19:42:38Z

lib/matplotlib/axis.py

+    def _get_tick_boxes_siblings(self, renderer):
+        """
+        Get the bounding boxes for this axis and its sibblings
+        as set by `Figure.align_xlabels` or  ``Figure.align_ylables`.


markup, typo "sibblings" above

anntzer · 2018-01-12T19:49:30Z

lib/matplotlib/figure.py

+                axx.yaxis._align_label_siblings += [ax.yaxis]
+
+    # place holder until #9498 is merged...
+    def align_titles(self, axs=None, renderer=None):


I think this is independent of #9498 (on the feature side; the implementations may be overlapping...), or did I miss somthing?

Oh, I remember now...

This code (align_titles) is not needed now, and hence the call to it is commented out below, because titles are always in the same position relative to the top of the axis (if placed automatically). So, going through this function is needlessly expensive (though for practical purposes its not that expensive).

If #9498 gets merged, the y-position of titles will be different, and this code will become useful.

~~No test, because if the title is placed manaully this shouldn't run....~~ EDIT: Ooops, no I don't offer that feature ;-)

anntzer · 2018-01-12T19:53:55Z

lib/matplotlib/figure.py

+                            same += [axc]
+
+            x0, y0 = ax.title.get_position()
+            for axx in same:


Looks easier (and cheaper) to get the max value of y in a first pass and then apply it throughout?

anntzer · 2018-01-12T19:57:16Z

lib/matplotlib/tests/test_figure.py

 import numpy as np
 import warnings
 import pytest


+@image_comparison(baseline_images=['figure_align_labels'])
+def test_align_labels():


also test for titles? not sure what's the status of that part

anntzer · 2018-01-12T19:59:22Z

One thing that's perhaps not obvious is that calling align_foo has a persistent effect (in other words it does not just do the alignment "for this single draw"). Which is very reasonable, but should be mentioned somewhere.

Otherwise, generally looks good (minor comments above).

anntzer · 2018-01-12T22:37:19Z

lib/matplotlib/axis.py

@@ -1675,8 +1675,8 @@ def set_ticks(self, ticks, minor=False):

    def _get_tick_boxes_siblings(self, renderer):
        """
-        Get the bounding boxes for this axis and its sibblings
-        as set by `Figure.align_xlabels` or  ``Figure.align_ylables`.
+        Get the bounding boxes for this `.axis` and it's siblings


anntzer · 2018-01-12T22:38:25Z

lib/matplotlib/figure.py

-            for axx in same:
-                _log.debug(' Same: %s', axx.xaxis.label)
-                axx.xaxis._align_label_siblings += [ax.xaxis]
+                    if (((labpo == 'bottom') and (rowc1 == row1)) or


not a big deal but that's a lot of extra parentheses (only the outermost are needed)

I guess somewhere in my head I knew that and precedes or, but I'm never sure, and I'm never sure from language to language, so...

jklymak · 2018-01-12T22:41:29Z

Thanks @anntzer

Removed title alignment for now. That depends on #9498, which is the only automatic way to have titles not be aligned; right now they are aligned by default. Of course if the user manually places one title, I assume they can align the other ones by hand because they know the position. I'll put the align_title code in #9498.

I did make a couple of small improvements (got rid of a loop), and tried to indicate what was going on in the comments (because even I was confused looking back at it).

Added the fact that this alignment is persistent to the docs. Also that it only applies to automatically applied label positions. Manually-placed labels don't get aligned. Though thats a bit subtle - I think the way it is now, if a manually aligned label is the outlier, others will get aligned to it. If its an inlier, then the others will not be aligned to it, and the manual label will also stay an inlier. I actually think thats desirable, but could change that behaviour if others disagree. I think its a bit of an edge case; again, if someone is lining things up by hand, they probaby shouldn't be calling this helper.

anntzer · 2018-01-13T00:18:58Z

Would it make sense to store the alignedness info on a Grouper (https://matplotlib.org/api/cbook_api.html#matplotlib.cbook.Grouper) stored on the Figure (one per axis), instead of in Axis._align_label_siblings? The main advantage would be that this way, if we ever figure out proper axes unsharing semantics (right now, shared axes cannot be unshared), hopefully the solution would also apply to unaligning (I don't think it's critical, just pointing out the possibility).

jklymak · 2018-01-13T00:33:53Z

The axes need not be shared if we align the labels (and indeed it would be strange for the labels to be misaligned in that case).

OTOH, (I think) I can see the advantage of a pair of Grouper in fig, simply to stop carying around MxN lists with repetitive info in them. i.e. Figure._aligned_xaxis_grp, Figure._aligned_yaxis_grp.

anntzer · 2018-01-13T00:38:25Z

The point is not to reuse the same grouper instance, just to use the same data structure.
I think this is very close to mergeable (even without the grouper thing).

anntzer · 2018-01-13T05:50:24Z

lib/matplotlib/figure.py

+
+        if axs is None:
+            axs = self.axes
+        axs = np.asarray(np.array(axs)).flatten().tolist()


You really really wanted to make sure that you got a list heh :-)
Currently this will also accept Axes instances (rather than Axes in a single-element list) and multi-dimensional arrays. Was that intended? Just making sure, I would as usual prefer a simpler interface but am only -0 on being more flexible.

Too much?

Ummm, I can't remember why I went so nuts on this.

I agree a singleton would be silly, though letting a list w/ one element would be fine.

It needs to be flattened, but before I can do that, it needs to be an np.array...

but, I have no idea why I have np.array and np.asarray in there.

Fixed?

axs = np.asarray(axs).flatten().tolist()

for ax in np.asarray(axs).flat: is enough (that'll give you an 1D iterator that directly refers to the original data).

Couldn't quite get that to work, and I want to save the list so I don't need the awkward construct twice. Using axs = np.asarray(axs).flatten().

TO be honest, not sure why flat doesn't work. It basically dropped one out of three axes on the figure in align_labels_demo.py so that third axis didn't get aligned. I don't really get why it doesn't work, but flatten() does, so...

Ah, I missed the fact that you had a nested loop also iterating over axs. In that case it is normal that the inner loop also advances the outer iterator and that won't work indeed.

As a side side note, using .ravel() instead of .flatten() is usually more efficient, because the latter always copies the data into a new array, whereas the former only does so when needed (i.e. when the data of the original array is not regularly spaced in memory). We're unlikely to ever have so many axes to run out of memory :-) but still a good habit IMO.

Ah, I see. Interesting. I never understood the difference between flatten() and ravel() and just use flatten because the word means something to me, whereas I'm not sure what ravel is supposed to mean.

"ravel" is apparently a synonym of "unravel" :) https://www.merriam-webster.com/dictionary/ravel?utm_campaign=sd&utm_medium=serp&utm_source=jsonld
[OT: reminds me of the day where my colleague explained to me that "being down for sthg" and "being up for sthg" meant essentially the same thing. English is sometimes weird :)]

Made even better by numpy's use of unravel to mean the opposite of ravel: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.unravel_index.html

anntzer

Very minor comments still being discussed but their resolution is not strictly required for this PR to go in.

dstansby added this to the v2.2 milestone Nov 1, 2017

anntzer reviewed Nov 1, 2017

View reviewed changes

afvincent reviewed Nov 2, 2017

View reviewed changes

jklymak changed the title ~~WIP: Align x and y labels between axes (Comments welcome)~~ Align x and y labels between axes Nov 2, 2017

jklymak force-pushed the alignxylabels branch 2 times, most recently from dd8e05e to 44b36df Compare November 2, 2017 19:02

jklymak force-pushed the alignxylabels branch from 44b36df to 86a6e32 Compare November 2, 2017 19:19

jklymak added the status: needs review label Dec 2, 2017

jklymak force-pushed the alignxylabels branch from 86a6e32 to 05e8fc2 Compare January 12, 2018 17:11

ENH: aligning x and y labels across axes

ce91095

jklymak force-pushed the alignxylabels branch from 05e8fc2 to ce91095 Compare January 12, 2018 18:26

anntzer reviewed Jan 12, 2018

View reviewed changes

jklymak force-pushed the alignxylabels branch from 5d225e9 to 1194941 Compare January 12, 2018 22:24

anntzer reviewed Jan 12, 2018

View reviewed changes

jklymak force-pushed the alignxylabels branch from 1194941 to 7928ecc Compare January 12, 2018 22:37

anntzer reviewed Jan 12, 2018

View reviewed changes

jklymak force-pushed the alignxylabels branch from 7928ecc to 1447ed9 Compare January 12, 2018 22:49

Refinements

a8f72f3

jklymak force-pushed the alignxylabels branch from 1447ed9 to a8f72f3 Compare January 12, 2018 23:38

anntzer added the Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. label Jan 13, 2018

Changed from lists to Grouper

17d1025

anntzer reviewed Jan 13, 2018

View reviewed changes

jklymak force-pushed the alignxylabels branch from 23273c3 to 64b270b Compare January 13, 2018 05:59

anntzer approved these changes Jan 13, 2018

View reviewed changes

jklymak force-pushed the alignxylabels branch from 64b270b to 9f00a79 Compare January 13, 2018 17:19

DOC: some more comments

2d47811

jklymak force-pushed the alignxylabels branch from 9f00a79 to 2d47811 Compare January 13, 2018 22:23

dstansby approved these changes Jan 24, 2018

View reviewed changes

dstansby merged commit 3d3472c into matplotlib:master Jan 24, 2018

QuLogic removed the status: needs review label Jan 24, 2018

jklymak deleted the alignxylabels branch January 24, 2018 23:57

QuLogic modified the milestones: needs sorting, v2.2.0 Feb 12, 2018

Align x and y labels between axes #9652

Align x and y labels between axes #9652

Conversation

jklymak commented Nov 1, 2017 • edited Loading

PR Summary

Example:

Before

After

PR Checklist

dstansby commented Nov 1, 2017

anntzer commented Nov 1, 2017

jklymak commented Nov 1, 2017

anntzer commented Nov 1, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jklymak commented Nov 2, 2017

anntzer commented Nov 2, 2017

jklymak commented Nov 2, 2017

jklymak commented Jan 12, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jklymak Jan 12, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

anntzer commented Jan 12, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jklymak commented Jan 12, 2018

anntzer commented Jan 13, 2018

jklymak commented Jan 13, 2018

anntzer commented Jan 13, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

anntzer left a comment

Choose a reason for hiding this comment

jklymak commented Nov 1, 2017 •

edited

Loading

jklymak Jan 12, 2018 •

edited

Loading

anntzer commented Jan 13, 2018 •

edited

Loading