ENH: box aspect for axes #14917

ImportanceOfBeingErnest · 2019-07-30T01:10:45Z

PR Summary

Matplotlib axes' aspect refers to the data, i.e. it defines the ratio of vertical vs. horizontal data units. However, it seems people often (mis)use it in order to set the aspect of the axes box, which is of course possible by knowing the limits of the data and using adjustable="box". But it inevitably leads to problems, e.g.

It doesn't work when there is a twinned axes in the game (as in Improve handling of shared axes with specified aspect ratio #10033 (comment), or Define aspect ratio when using twinx in new version of matplotlib)
It breaks down when limits change, i.e. in most interactive cases (zoom/pan...)
It may lead to overly complicated solutions (Create equal aspect (square) plot with multiple axes when data limits are different?), not even the axes_grid1 toolkit is able to solve natively.

In short, sometimes people just want to make a square plot.

So it seems useful to introduce a box_aspect parameter, which sets the aspect (i.e. the ratio between height and width) of the axes box, independent of the data.

Some usecases:

A. A square axes, independent of data

fig1, ax = plt.subplots()

ax.set_xlim(300,400)
ax.set_box_aspect(1)

B. Shared square axes

fig2, (ax, ax2) = plt.subplots(ncols=2, sharey=True)

ax.plot([0,10])
ax2.plot([10,20])

ax.set_box_aspect(1)
ax2.set_box_aspect(1)

C. Square twin axes

fig3, ax = plt.subplots()

ax2 = ax.twinx()

ax.plot([0,10])
ax2.plot([12,10])

ax.set_box_aspect(1)

D. Normal plot next to image (works with constrained_layout)

fig4, (ax, ax2) = plt.subplots(ncols=2, constrained_layout=True)

im = np.random.rand(16,27)
ax.imshow(im)

ax2.plot([23,45])
ax2.set_box_aspect(im.shape[0]/im.shape[1])

E. Square joint/marginal plot

fig5, axs = plt.subplots(2,2, sharex="col", sharey="row", 
                        gridspec_kw = dict(height_ratios=[1,3], 
                                           width_ratios=[3,1]))
axs[0,1].set_visible(False)
axs[0,0].set_box_aspect(1/3)
axs[1,0].set_box_aspect(1)
axs[1,1].set_box_aspect(3/1)

x,y = np.random.randn(2,400) * np.array([[.5],[180]])
axs[1,0].scatter(x,y)
axs[0,0].hist(x)
axs[1,1].hist(y, orientation="horizontal")

F. Equal data aspect, unequal box aspect

fig6, ax = plt.subplots()

ax.add_patch(plt.Circle((5,3), 1))
ax.set_aspect("equal", adjustable="datalim")
ax.set_box_aspect(0.5)
ax.autoscale()

Issues:

~~As @jklymak already pointed out, there might be a problem with layout managers, and indeed at least case E. from above fails with contrained_layout.~~ (fixed by FIX constrained_layout w/ hidden axes #14919)
Obviously a fixed box_aspect only makes sense with adjustable="datalim". Since the order in which box_aspect and adjustable are set is arbitrary, this still needs to define in which cases to error out, or silently fall back.

PR Checklist

Has Pytest style unit tests
Code is finally Flake 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)

jklymak · 2019-07-30T01:46:29Z

Seems cool. But I think we need to sit down and think about all the axes sizing and sharing options a bit more holistically.

I think if you change the non-original position then you can keep constrained layout working.

ImportanceOfBeingErnest · 2019-07-30T02:39:35Z

I think if you change the non-original position then you can keep constrained layout working.

I would have thought that is what I'm doing in line 1527. Will need to check this at some later point.

[..] we need to sit down and think about all the axes sizing and sharing options a bit more holistically.

Sure, that's why I put it out in its current state, to see what is possible and to define some working use cases. In general what's still missing is a sharing of geometries (as opposed to sharing of data); or maybe call it a twinning across subplots.

jklymak · 2019-07-30T03:05:57Z

Great, then I'm surprised there are issues with constrained_layout since its no different than a data-aspect constrained axes. constrained_layout only cares about the "original" size.

jklymak · 2019-07-30T03:26:38Z

Thanks - that was a bug: #14918, and solution: #14919

anntzer · 2019-07-30T08:38:49Z

I think it's a great idea :)
I believe this is mutually exclusive with setting the old data aspect? If so, perhaps this should be exposed in the same API, e.g. set_aspect(aspect, "data"/"axes") (the second argument defaulting to "data" which maintains backcompat) and internally stored in the same attribute? The point being that after

ax.set_aspect(a, "data")
ax.set_aspect(b, "axes")

it's not too surprising if the second call overwrites the first one, whereas in

ax.set_aspect(a)
ax.set_box_aspect(b)

... do we want the second to overwrite the first one? do we want to add a check and error out? or whatnot.

ImportanceOfBeingErnest · 2019-07-30T12:12:02Z

No, the usual (data-)aspect and the box aspect are not mutually exclusive. In fact using both may make perfect sense, as shown in case F.. (It's just that adjustable cannot be "box" in case of using a box aspect.)

anntzer · 2019-07-30T12:31:26Z

Ah, thanks for the clarification, I missed that; ignore the nonsense I wrote above.

jklymak · 2019-07-30T15:31:49Z

Some things to consider: What happens if the user manually sets the position and the box_aspect?

ax.set_position([0.1, 0.1, 0.9, 0.2])
ax.set_box_aspect(1.)

For constrained_layout, the set_position method takes the axes out of the automatic layout because we assume the user wants that position set. But I indicate that just by setting the layout boxes associated with the axes to None in ax.set_position, which you probably don't want to depend on here.

I'm not sure which behaviour you want here, but you'll need to decide which call trumps which here. Even more tricky is

ax.set_box_aspect(1.)
ax.set_position([0.1, 0.1, 0.9, 0.2])

ImportanceOfBeingErnest · 2019-07-31T15:44:17Z

The order of setting position and box_aspect isn't relevant. But I added a test to make sure that's the case.

andrzejnovak · 2019-08-08T15:32:28Z

Just so it doesn't get lost from #15010 , I would really like if it was possible to choose this behaviour as a default setting in rcParams.

ImportanceOfBeingErnest · 2019-08-09T03:06:25Z

I mentionned my problem with an rc parameter for this in #15001 (comment):

if it's to be set at init time of any axes, it would sure cause problems e.g. with colorbar axes [also: Widget axes, insets etc.] and also the interplay between the adjustable parameter would need to be sorted before the axes is drawn.

I could imagine though to let this be taken as argument by the axes' init, to allow for something like

fig, axs = plt.subplots(2,2, subplot_kw=dict(box_aspect=1))

andrzejnovak · 2019-08-09T07:04:24Z

I think that's a good compromise. It could be made to work by only applying it to the first axes. append_axes, add_subplot and add_axes would check if there already is one in the figure and go back to auto, but I could see why that would not be very transparent.

tacaswell · 2019-08-12T21:58:35Z

Power cycled to try to re-run against master and then belated realized this won't help with circle...

Half of the warnings were from the required target not being hit, pushed a commit making sure they are generated (we do not auto-doc Axes).

tacaswell · 2019-08-12T22:52:09Z

ok, I understand the other half of the errors.

we auto-generate and insert

    .. table::
       :class: property-table

       =======================================================================================   =====================================================================================================
       Property                                                                                  Description                                                                                          
       =======================================================================================   =====================================================================================================
       :meth:`adjustable <matplotlib.axes._base._AxesBase.set_adjustable>`                       {'box', 'datalim'}                                                                                   
       :meth:`agg_filter <matplotlib.artist.Artist.set_agg_filter>`                              a filter function, which takes a (m, n, 3) float array and a dpi value, and returns a (m, n, 3) array
       :meth:`alpha <matplotlib.artist.Artist.set_alpha>`                                        float or None                                                                                        
       :meth:`anchor <matplotlib.axes._base._AxesBase.set_anchor>`                               2-tuple of floats or {'C', 'SW', 'S', 'SE', ...}                                                     
       :meth:`animated <matplotlib.artist.Artist.set_animated>`                                  bool                                                                                                 
       :meth:`aspect <matplotlib.axes._base._AxesBase.set_aspect>`                               {'auto', 'equal'} or num                                                                             
       :meth:`autoscale_on <matplotlib.axes._base._AxesBase.set_autoscale_on>`                   bool                                                                                                 
       :meth:`autoscalex_on <matplotlib.axes._base._AxesBase.set_autoscalex_on>`                 bool                                                                                                 
       :meth:`autoscaley_on <matplotlib.axes._base._AxesBase.set_autoscaley_on>`                 bool                                                                                                 
       :meth:`axes_locator <matplotlib.axes._base._AxesBase.set_axes_locator>`                   Callable[[Axes, Renderer], Bbox]                                                                     
       :meth:`axisbelow <matplotlib.axes._base._AxesBase.set_axisbelow>`                         bool or 'line'                                                                                       
       :meth:`box_aspect <matplotlib.axes._base._AxesBase.set_box_aspect>`                       None, or a number                                                                                    
       :meth:`clip_box <matplotlib.artist.Artist.set_clip_box>`                                  `.Bbox`                                                                                              
       :meth:`clip_on <matplotlib.artist.Artist.set_clip_on>`                                    bool                                                                                                 
       :meth:`clip_path <matplotlib.artist.Artist.set_clip_path>`                                [(`~matplotlib.path.Path`, `.Transform`) | `.Patch` | None]                                          
       :meth:`contains <matplotlib.artist.Artist.set_contains>`                                  callable                                                                                             
       :meth:`facecolor <matplotlib.axes._base._AxesBase.set_facecolor>`                         color                                                                                                
       :meth:`fc <matplotlib.axes._base._AxesBase.set_facecolor>`                                color                                                                                                
       :meth:`figure <matplotlib.axes._base._AxesBase.set_figure>`                               `.Figure`                                                                                            
       :meth:`frame_on <matplotlib.axes._base._AxesBase.set_frame_on>`                           bool                                                                                                 
       :meth:`gid <matplotlib.artist.Artist.set_gid>`                                            str                                                                                                  
       :meth:`in_layout <matplotlib.artist.Artist.set_in_layout>`                                bool                                                                                                 
       :meth:`label <matplotlib.artist.Artist.set_label>`                                        object                                                                                               
       :meth:`navigate <matplotlib.axes._base._AxesBase.set_navigate>`                           bool                                                                                                 
       :meth:`navigate_mode <matplotlib.axes._base._AxesBase.set_navigate_mode>`                 unknown                                                                                              
       :meth:`path_effects <matplotlib.artist.Artist.set_path_effects>`                          `.AbstractPathEffect`                                                                                
       :meth:`picker <matplotlib.artist.Artist.set_picker>`                                      None or bool or float or callable                                                                    
       :meth:`position <matplotlib.axes._base._AxesBase.set_position>`                           [left, bottom, width, height] or `~matplotlib.transforms.Bbox`                                       
       :meth:`prop_cycle <matplotlib.axes._base._AxesBase.set_prop_cycle>`                       unknown                                                                                              
       :meth:`rasterization_zorder <matplotlib.axes._base._AxesBase.set_rasterization_zorder>`   float or None                                                                                        
       :meth:`rasterized <matplotlib.artist.Artist.set_rasterized>`                              bool or None                                                                                         
       :meth:`sketch_params <matplotlib.artist.Artist.set_sketch_params>`                        (scale: float, length: float, randomness: float)                                                     
       :meth:`snap <matplotlib.artist.Artist.set_snap>`                                          bool or None                                                                                         
       :meth:`title <matplotlib.axes._axes.Axes.set_title>`                                      str                                                                                                  
       :meth:`transform <matplotlib.artist.Artist.set_transform>`                                `.Transform`                                                                                         
       :meth:`url <matplotlib.artist.Artist.set_url>`                                            str                                                                                                  
       :meth:`visible <matplotlib.artist.Artist.set_visible>`                                    bool                                                                                                 
       :meth:`xbound <matplotlib.axes._base._AxesBase.set_xbound>`                               unknown                                                                                              
       :meth:`xlabel <matplotlib.axes._axes.Axes.set_xlabel>`                                    str                                                                                                  
       :meth:`xlim <matplotlib.axes._base._AxesBase.set_xlim>`                                   (left: float, right: float)                                                                          
       :meth:`xmargin <matplotlib.axes._base._AxesBase.set_xmargin>`                             float greater than -0.5                                                                              
       :meth:`xscale <matplotlib.axes._base._AxesBase.set_xscale>`                               {"linear", "log", "symlog", "logit", ...}                                                            
       :meth:`xticklabels <matplotlib.axes._base._AxesBase.set_xticklabels>`                     List[str]                                                                                            
       :meth:`xticks <matplotlib.axes._base._AxesBase.set_xticks>`                               unknown                                                                                              
       :meth:`ybound <matplotlib.axes._base._AxesBase.set_ybound>`                               unknown                                                                                              
       :meth:`ylabel <matplotlib.axes._axes.Axes.set_ylabel>`                                    str                                                                                                  
       :meth:`ylim <matplotlib.axes._base._AxesBase.set_ylim>`                                   (bottom: float, top: float)                                                                          
       :meth:`ymargin <matplotlib.axes._base._AxesBase.set_ymargin>`                             float greater than -0.5                                                                              
       :meth:`yscale <matplotlib.axes._base._AxesBase.set_yscale>`                               {"linear", "log", "symlog", "logit", ...}                                                            
       :meth:`yticklabels <matplotlib.axes._base._AxesBase.set_yticklabels>`                     List[str]                                                                                            
       :meth:`yticks <matplotlib.axes._base._AxesBase.set_yticks>`                               unknown                                                                                              
       :meth:`zorder <matplotlib.artist.Artist.set_zorder>`                                      float                                                                                                
       =======================================================================================   =====================================================================================================

into a bunch of docstrings which points to the base class (because it asks the methods what class they belong to). Will push a commit 'fixing' this soon...

anntzer · 2019-09-11T15:13:20Z

I think this is basically good to go.
Minor nit: perhaps name this "axes_aspect"? consistently with transAxes, with the places that accept "axes"/"data" as transform specifiers, etc. Obviously it's a bit(...) late to rename the classic "aspect" to "data_aspect", though...

ImportanceOfBeingErnest · 2019-09-11T15:45:06Z

I would agree that if we had .set_data_aspect, naming this .set_axes_aspect would be ideal. But given that the data aspect is by now commonly known as "the aspect", it is important to distinguish clearly.

In general we have the semantics of object.set_property(...) throughout the code, which reads as "set the object's property to ...". In the case of the aspect, axes.set_aspect, needs to read "set the axes' aspect to ...". Following this, axes.set_axes_aspect would read "set the axes' axes aspect to" in which case there is little indication for that to be different from "set the axes' aspect to"; and that would be the data aspect.

I therefore find .set_box_aspect much clearer: The axes has an aspect (which is related to data), and it has a box-aspect (which is related to its box).

jklymak · 2019-09-11T18:18:08Z

OK, I'll check later, but what if you set two axes to have different aspect ratio, and use a layout manager? I'm just a little concerned there are all sorts of edge cases here that will cause problems, but I may be completely incorrect.

jklymak

Actually because this works on the active position of the axes, it works fine w/ the layout managers. Or at least I couldn't easily break it.

jklymak · 2019-09-11T19:24:01Z

I'll merge tomorrow, if no one else does, or @ImportanceOfBeingErnest you can self-merge if we give this a day or so to make sure there are no further objections.

jklymak · 2019-10-01T16:45:15Z

Thanks @ImportanceOfBeingErnest I think this will be a pretty valuable tool in the axes shaping toolbox!

The way that bbox_inches='tight' is implemented we need to ensure that we do not try to adjust the aspect during the draw (because we have temporarily de-coupled the reported figure size from the transforms which results in the being distorted). Previously we did not have a way to fix the aspect ratio in screen space of the Axes (only the aspect ratio in dataspace) however in 3.3 we gained this ability for both Axes (matplotlib#14917) and Axes3D (matplotlib#8896 / matplotlib#16472). Rather than add an aspect value to `set_aspect` to handle this case, in the tight_bbox code we monkey-patch the `apply_aspect` method with a no-op function and then restore it when we are done. Previously we would set the aspect to "auto" and restore it in the same places. closes matplotlib#16463.

ImportanceOfBeingErnest added topic: geometry manager LayoutEngine, Constrained layout, Tight layout status: work in progress topic: ticks axis labels status: needs comment/discussion needs consensus on next step labels Jul 30, 2019

jklymak mentioned this pull request Jul 30, 2019

FIX constrained_layout w/ hidden axes #14919

Merged

6 tasks

ImportanceOfBeingErnest force-pushed the box-aspect branch from 25da964 to cdd2843 Compare July 31, 2019 15:43

ImportanceOfBeingErnest force-pushed the box-aspect branch 3 times, most recently from 3b19944 to 0e2fdb9 Compare August 1, 2019 01:46

ImportanceOfBeingErnest removed the status: work in progress label Aug 1, 2019

ImportanceOfBeingErnest added this to the v3.2.0 milestone Aug 1, 2019

ImportanceOfBeingErnest mentioned this pull request Aug 7, 2019

Add axes.aspect as an option to rcParams #15001

Closed

6 tasks

jklymak mentioned this pull request Aug 8, 2019

Adding colorbar without affecting the plotting area size. #15010

Closed

ImportanceOfBeingErnest force-pushed the box-aspect branch 3 times, most recently from 68af0c2 to f4cd76e Compare August 10, 2019 23:35

ImportanceOfBeingErnest closed this Aug 12, 2019

ImportanceOfBeingErnest reopened this Aug 12, 2019

ImportanceOfBeingErnest force-pushed the box-aspect branch 3 times, most recently from 458ff92 to f40a5b7 Compare August 12, 2019 14:27

ImportanceOfBeingErnest mentioned this pull request Aug 12, 2019

How to handle sphinx nitpicky mode #15042

Closed

tacaswell closed this Aug 12, 2019

tacaswell reopened this Aug 12, 2019

tacaswell mentioned this pull request Aug 12, 2019

DOC: more nitpick follow up #15048

Merged

6 tasks

ImportanceOfBeingErnest force-pushed the box-aspect branch from fb8ddd7 to db45c2f Compare August 15, 2019 01:45

box aspect for axes

20134e2

ImportanceOfBeingErnest force-pushed the box-aspect branch from db45c2f to 20134e2 Compare August 15, 2019 02:53

tacaswell modified the milestones: v3.2.0, v3.3.0 Aug 24, 2019

ImportanceOfBeingErnest added status: needs review and removed status: needs comment/discussion needs consensus on next step labels Aug 28, 2019

anntzer approved these changes Sep 11, 2019

View reviewed changes

jklymak approved these changes Sep 11, 2019

View reviewed changes

jklymak merged commit 805cb72 into matplotlib:master Oct 1, 2019

ImportanceOfBeingErnest deleted the box-aspect branch October 3, 2019 01:25

QuLogic removed the status: needs review label Oct 3, 2019

jklymak mentioned this pull request May 28, 2020

FIX: add set_box_aspect, improve tight bounding box for Axes3D + fix bbox_inches support with fixed box_aspect #17515

Merged

6 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ENH: box aspect for axes #14917

ENH: box aspect for axes #14917

ImportanceOfBeingErnest commented Jul 30, 2019 •

edited

Loading

jklymak commented Jul 30, 2019

ImportanceOfBeingErnest commented Jul 30, 2019

jklymak commented Jul 30, 2019

jklymak commented Jul 30, 2019

anntzer commented Jul 30, 2019

ImportanceOfBeingErnest commented Jul 30, 2019

anntzer commented Jul 30, 2019 •

edited

Loading

jklymak commented Jul 30, 2019

ImportanceOfBeingErnest commented Jul 31, 2019

andrzejnovak commented Aug 8, 2019

ImportanceOfBeingErnest commented Aug 9, 2019 •

edited

Loading

andrzejnovak commented Aug 9, 2019

tacaswell commented Aug 12, 2019

tacaswell commented Aug 12, 2019

anntzer commented Sep 11, 2019

ImportanceOfBeingErnest commented Sep 11, 2019

jklymak commented Sep 11, 2019

jklymak left a comment

jklymak commented Sep 11, 2019

jklymak commented Oct 1, 2019

ENH: box aspect for axes #14917

ENH: box aspect for axes #14917

Conversation

ImportanceOfBeingErnest commented Jul 30, 2019 • edited Loading

PR Summary

Some usecases:

Issues:

PR Checklist

jklymak commented Jul 30, 2019

ImportanceOfBeingErnest commented Jul 30, 2019

jklymak commented Jul 30, 2019

jklymak commented Jul 30, 2019

anntzer commented Jul 30, 2019

ImportanceOfBeingErnest commented Jul 30, 2019

anntzer commented Jul 30, 2019 • edited Loading

jklymak commented Jul 30, 2019

ImportanceOfBeingErnest commented Jul 31, 2019

andrzejnovak commented Aug 8, 2019

ImportanceOfBeingErnest commented Aug 9, 2019 • edited Loading

andrzejnovak commented Aug 9, 2019

tacaswell commented Aug 12, 2019

tacaswell commented Aug 12, 2019

anntzer commented Sep 11, 2019

ImportanceOfBeingErnest commented Sep 11, 2019

jklymak commented Sep 11, 2019

jklymak left a comment

Choose a reason for hiding this comment

jklymak commented Sep 11, 2019

jklymak commented Oct 1, 2019

ImportanceOfBeingErnest commented Jul 30, 2019 •

edited

Loading

anntzer commented Jul 30, 2019 •

edited

Loading

ImportanceOfBeingErnest commented Aug 9, 2019 •

edited

Loading