3d plots with aspect='equal'

aspect='equal' argument is completely misused by axes3D.
To verify:

import matplotlib.pyplot as plt
from mpl_toolkits import mplot3d
fig = plt.figure()
ax = fig.add_subplot((111), aspect='equal', projection='3d')
ax.scatter((1, 2), (1, 1.2), (1, 2))

The figure is rendered with all axes scaled to the data scale, thus ignoring the aspect setting.
If one tries to rotate the figure, using GUI, it is rescaled to fix into a very uneven rectangle, presumably calculated from x-scale vs y-scale, and hence with aspect ratio 1:5.

an "aspect" kwarg doesn't make a lot of sense for 3d plots, which is why even matlab doesn't do it. I have an old branch that provides some very basic functionality in this direction, but I never finished. See here:

http://matplotlib.1069221.n5.nabble.com/mplot3d-and-daspect-tp11521p11523.html

PRs against this branch would be welcomed!

Why doesn't it make sense? If I want to e.g. mark coordinates in a building, or any system where three axes have the same dimension, I do want that the same axes is chosen for each of the axes (that's where I actually encountered the bug). And of course regardless of that, the current handling is bugged.

I would also be interested by this feature.
By the way, did you found a way to tackle the problem?

@koepferl

We also came across this issue - if I run the following code:

from mpl_toolkits.mplot3d import Axes3D
import matplotlib.pyplot as plt
import numpy as np
from itertools import product, combinations

fig = plt.figure()
ax = fig.add_subplot(1,1,1,projection='3d')

#draw cube
r = [-1, 1]
for s, e in combinations(np.array(list(product(r,r,r))), 2):
    if np.sum(np.abs(s-e)) == r[1]-r[0]:
        ax.plot3D(*zip(s,e), color="b")

I get:

and adding

ax.set_aspect('equal')

then gives

which is simply wrong, not just sub-optimal. The cube lines should always be parallel to the axes. In my opinion, it should look more like

But at the very least, if set_aspect('equal') is going to do the wrong thing, it would be better if it raised a NotImplemented error.

(this issue was originally pointed out to me by @koepferl)

My research entails lots of 3D plotting on unit spheres. Needless to say, a unit sphere is a perfect sphere and to display it properly all axes need to have the same scale. This is another use case that requires this functionality.

I would love to help out, but don't have the time to dig into the code right now. I really hope this feature gets some priority for future releases.

Just in case somebody encounters this bug, and needs a workaround: the way I handle this is by manually calling ax.auto_scale_xyz before rendering.

Actually there are several ways of showing a 3D model on a 2D screen, mainly including axonometric projection and perspective projection. There are also many kinds of axonometric projections. So even the unit length of three axises are the same, they look different on the screen on many circumstances. A cubic box may look not that cubic if drawn in different ways. So if the 3D plot class is to be improved, projection method should be determined beforehand.

@Hanlin-Dong

@Hanlin-Dong I agree, but no matter how the projection is done, the projection for the data must match the projection for the axes. In the example above it doesn't, hence the bug report. Also the bug report isn't about the projection as much as the impossibility to correctly set the aspect ratios.

@akhmerov

I also think that this feature is quite important for 3d plots.

True, it might not be that important if you visualize data without
an underlying real-world geometry. But I have to visualize the
heat distribution on motor parts in 3D and this looks quite strange if the
aspect ratio is incorrect. I guess the same it true with the buildings that
@akhmerov mentioned.

@WeatherGod mentioned above that Matlab does not provide this feature, but
I am quite sure it does.
This is the manpage, it also shows 3D examples for "normal" and "equal".
http://de.mathworks.com/help/matlab/creating_plots/aspect-ratio-for-2-d-axes.html
More information on how it is implemented as axis property:
http://de.mathworks.com/help/matlab/ref/axes-properties.html#prop_DataAspectRatio

I guess most of it is already implemented and working in matplotlib but I came across
a small issue, see my post and example code in #1104.

I just got this plot after calling ax.set_aspect('equal'). Definitely not equal based on the ticks!

Related and possibly a first step for fixing this: #8896

Closing due to #13474 and the fact that the non-optimal behaviour is discussed in previous issues and PRs references above...

@jklymak

@jklymak what do you mean by the non-optimal behavior? Does closing this issue mean that 3D plots with equal aspect ratio aren't on the roadmap?

I hope he is closing because of (still open) #8896 just to clean the issue tracker, as #13474 is only a workaround on how to deal with the behaviour until it is implemented.

Correct. OTOH I don't know of anyone actively developing mplot3d, so if someone wants this feature, it would require a volunteer..

The original link is https://stackoverflow.com/questions/13685386/matplotlib-equal-unit-length-with-equal-aspect-ratio-z-axis-is-not-equal-to and has already been mentioned above.

Removed dependency on specific matplotlib version

Fixed the matplotlib 3D set_aspect('equal') issue

In order to get a future notification of aspect='equal' working, which open issue (or unmerged PR) should I subscribe to?

Please see the note at https://matplotlib.org/stable/users/prev_whats_new/whats_new_3.3.0.html#axes3d-no-longer-distorts-the-3d-plot-to-match-the-2d-aspect-ratio (from #17515) for how to get this effect.

If setting this with a string is important please open a new issue with what you would expect "equal" to do under various conditions / adjustable settings.

The code that would need to be changed to implement this is

matplotlib/lib/mpl_toolkits/mplot3d/axes3d.py

Lines 275 to 347 in a3fbf18

    
               def set_aspect(self, aspect, adjustable=None, anchor=None, share=False): 
        
                   """ 
        
                   Set the aspect ratios. 
        
                   Axes 3D does not current support any aspect but 'auto' which fills 
        
                   the axes with the data limits. 
        
                   To simulate having equal aspect in data space, set the ratio 
        
                   of your data limits to match the value of `.get_box_aspect`. 
        
                   To control box aspect ratios use `~.Axes3D.set_box_aspect`. 
        
                   Parameters 
        
                   ---------- 
        
                   aspect : {'auto'} 
        
                       Possible values: 
        
                       =========   ================================================== 
        
                       value       description 
        
                       =========   ================================================== 
        
                       'auto'      automatic; fill the position rectangle with data. 
        
                       =========   ================================================== 
        
                   adjustable : None 
        
                       Currently ignored by Axes3D 
        
                       If not *None*, this defines which parameter will be adjusted to 
        
                       meet the required aspect. See `.set_adjustable` for further 
        
                       details. 
        
                   anchor : None or str or 2-tuple of float, optional 
        
                       If not *None*, this defines where the Axes will be drawn if there 
        
                       is extra space due to aspect constraints. The most common way to 
        
                       to specify the anchor are abbreviations of cardinal directions: 
        
                       =====   ===================== 
        
                       value   description 
        
                       =====   ===================== 
        
                       'C'     centered 
        
                       'SW'    lower left corner 
        
                       'S'     middle of bottom edge 
        
                       'SE'    lower right corner 
        
                       etc. 
        
                       =====   ===================== 
        
                       See `.set_anchor` for further details. 
        
                   share : bool, default: False 
        
                       If ``True``, apply the settings to all shared Axes. 
        
                   See Also 
        
                   -------- 
        
                   mpl_toolkits.mplot3d.axes3d.Axes3D.set_box_aspect 
        
                   """ 
        
                   if aspect != 'auto': 
        
                       raise NotImplementedError( 
        
                           "Axes3D currently only supports the aspect argument " 
        
                           f"'auto'. You passed in {aspect!r}." 
        
                       ) 
        
                   if share: 
        
                       axes = {*self._shared_x_axes.get_siblings(self), 
        
                               *self._shared_y_axes.get_siblings(self), 
        
                               *self._shared_z_axes.get_siblings(self), 
        
                               } 
        
                   else: 
        
                       axes = {self} 
        
                   for ax in axes: 
        
                       ax._aspect = aspect 
        
                       ax.stale = True 
        
                   if anchor is not None: 
        
                       self.set_anchor(anchor, share=share)

and

matplotlib/lib/mpl_toolkits/mplot3d/axes3d.py

Lines 407 to 422 in a3fbf18

    
           def apply_aspect(self, position=None): 
        
               if position is None: 
        
                   position = self.get_position(original=True) 
        
               # in the superclass, we would go through and actually deal with axis 
        
               # scales and box/datalim. Those are all irrelevant - all we need to do 
        
               # is make sure our coordinate system is square. 
        
               trans = self.get_figure().transSubfigure 
        
               bb = mtransforms.Bbox.from_bounds(0, 0, 1, 1).transformed(trans) 
        
               # this is the physical aspect of the panel (or figure): 
        
               fig_aspect = bb.height / bb.width 
        
               box_aspect = 1 
        
               pb = position.frozen() 
        
               pb1 = pb.shrunk_to_aspect(box_aspect, pb, fig_aspect) 
        
               self._set_position(pb1.anchored(self.get_anchor(), pb), 'active')

and likely

matplotlib/lib/mpl_toolkits/mplot3d/axes3d.py

Lines 365 to 405 in a3fbf18

    
               def set_box_aspect(self, aspect, *, zoom=1): 
        
                   """ 
        
                   Set the axes box aspect. 
        
                   The box aspect is the ratio of height to width in display 
        
                   units for each face of the box when viewed perpendicular to 
        
                   that face.  This is not to be confused with the data aspect 
        
                   (which for Axes3D is always 'auto').  The default ratios are 
        
                   4:4:3 (x:y:z). 
        
                   To simulate having equal aspect in data space, set the box 
        
                   aspect to match your data range in each dimension. 
        
                   *zoom* controls the overall size of the Axes3D in the figure. 
        
                   Parameters 
        
                   ---------- 
        
                   aspect : 3-tuple of floats or None 
        
                       Changes the physical dimensions of the Axes3D, such that the ratio 
        
                       of the axis lengths in display units is x:y:z. 
        
                       If None, defaults to 4:4:3 
        
                   zoom : float 
        
                       Control overall size of the Axes3D in the figure. 
        
                   """ 
        
                   if aspect is None: 
        
                       aspect = np.asarray((4, 4, 3), dtype=float) 
        
                   else: 
        
                       orig_aspect = aspect 
        
                       aspect = np.asarray(aspect, dtype=float) 
        
                       if aspect.shape != (3,): 
        
                           raise ValueError( 
        
                               "You must pass a 3-tuple that can be cast to floats. " 
        
                               f"You passed {orig_aspect!r}" 
        
                           ) 
        
                   # default scale tuned to match the mpl32 appearance. 
        
                   aspect *= 1.8294640721620434 * zoom / np.linalg.norm(aspect) 
        
                   self._box_aspect = aspect 
        
                   self.stale = True

It is not obvious to me what the correct heuristics are to adjust the data limits / box aspect are under the different adjustable regimes (or if the adjustable values should be the same as in 2D?).

I suspect that this is 10-30 (apply the standard factor of pi, but would also be very happy to be over estimating) hours of work (most of it reading the mplot3d code) to implement for someone who knows Python well.

@SirSharpest Can you please explain why the process in https://matplotlib.org/stable/users/prev_whats_new/whats_new_3.3.0.html#axes3d-no-longer-distorts-the-3d-plot-to-match-the-2d-aspect-ratio is insufficient?

all what axis('equal') should do is this:

xlim = ax.get_xlim3d()
ylim = ax.get_ylim3d()
zlim = ax.get_zlim3d()
ax.set_box_aspect((xlim[1]-xlim[0], ylim[1]-ylim[0], zlim[1]-zlim[0]))

There is a PR that hopefully will be included in 3.6. #23409

Navigation Menu

Repository navigation

3d plots with aspect='equal' #1077

11 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	def set_aspect(self, aspect, adjustable=None, anchor=None, share=False):
	"""
	Set the aspect ratios.

	Axes 3D does not current support any aspect but 'auto' which fills
	the axes with the data limits.

	To simulate having equal aspect in data space, set the ratio
	of your data limits to match the value of `.get_box_aspect`.
	To control box aspect ratios use `~.Axes3D.set_box_aspect`.

	Parameters
	----------
	aspect : {'auto'}
	Possible values:

	========= ==================================================
	value description
	========= ==================================================
	'auto' automatic; fill the position rectangle with data.
	========= ==================================================

	adjustable : None
	Currently ignored by Axes3D

	If not None, this defines which parameter will be adjusted to
	meet the required aspect. See `.set_adjustable` for further
	details.

	anchor : None or str or 2-tuple of float, optional
	If not None, this defines where the Axes will be drawn if there
	is extra space due to aspect constraints. The most common way to
	to specify the anchor are abbreviations of cardinal directions:

	===== =====================
	value description
	===== =====================
	'C' centered
	'SW' lower left corner
	'S' middle of bottom edge
	'SE' lower right corner
	etc.
	===== =====================

	See `.set_anchor` for further details.

	share : bool, default: False
	If ``True``, apply the settings to all shared Axes.

	See Also
	--------
	mpl_toolkits.mplot3d.axes3d.Axes3D.set_box_aspect
	"""
	if aspect != 'auto':
	raise NotImplementedError(
	"Axes3D currently only supports the aspect argument "
	f"'auto'. You passed in {aspect!r}."
	)

	if share:
	axes = {*self._shared_x_axes.get_siblings(self),
	*self._shared_y_axes.get_siblings(self),
	*self._shared_z_axes.get_siblings(self),
	}
	else:
	axes = {self}

	for ax in axes:
	ax._aspect = aspect
	ax.stale = True

	if anchor is not None:
	self.set_anchor(anchor, share=share)

	def apply_aspect(self, position=None):
	if position is None:
	position = self.get_position(original=True)

	# in the superclass, we would go through and actually deal with axis
	# scales and box/datalim. Those are all irrelevant - all we need to do
	# is make sure our coordinate system is square.
	trans = self.get_figure().transSubfigure
	bb = mtransforms.Bbox.from_bounds(0, 0, 1, 1).transformed(trans)
	# this is the physical aspect of the panel (or figure):
	fig_aspect = bb.height / bb.width

	box_aspect = 1
	pb = position.frozen()
	pb1 = pb.shrunk_to_aspect(box_aspect, pb, fig_aspect)
	self._set_position(pb1.anchored(self.get_anchor(), pb), 'active')

	def set_box_aspect(self, aspect, *, zoom=1):
	"""
	Set the axes box aspect.

	The box aspect is the ratio of height to width in display
	units for each face of the box when viewed perpendicular to
	that face. This is not to be confused with the data aspect
	(which for Axes3D is always 'auto'). The default ratios are
	4:4:3 (x:y:z).

	To simulate having equal aspect in data space, set the box
	aspect to match your data range in each dimension.

	zoom controls the overall size of the Axes3D in the figure.

	Parameters
	----------
	aspect : 3-tuple of floats or None
	Changes the physical dimensions of the Axes3D, such that the ratio
	of the axis lengths in display units is x:y:z.

	If None, defaults to 4:4:3

	zoom : float
	Control overall size of the Axes3D in the figure.
	"""
	if aspect is None:
	aspect = np.asarray((4, 4, 3), dtype=float)
	else:
	orig_aspect = aspect
	aspect = np.asarray(aspect, dtype=float)
	if aspect.shape != (3,):
	raise ValueError(
	"You must pass a 3-tuple that can be cast to floats. "
	f"You passed {orig_aspect!r}"
	)
	# default scale tuned to match the mpl32 appearance.
	aspect = 1.8294640721620434 zoom / np.linalg.norm(aspect)

	self._box_aspect = aspect
	self.stale = True

3d plots with aspect='equal' #1077

Description

Activity

WeatherGod commented on Aug 20, 2012

akhmerov commented on Aug 20, 2012

vallettea commented on Jan 23, 2013

astrofrog commented on Feb 24, 2013

tehingo commented on Mar 25, 2013

akhmerov commented on Mar 25, 2013

Hanlin-Dong commented on Jan 6, 2016

akhmerov commented on Jan 6, 2016

solarjoe commented on Jan 18, 2016

samuela commented on May 6, 2018

timhoffm commented on May 6, 2018

11 remaining items

jklymak commented on Mar 3, 2019

akhmerov commented on Mar 3, 2019

solarjoe commented on Mar 3, 2019

jklymak commented on Mar 3, 2019

zh-plus commented on Aug 14, 2019

solarjoe commented on Aug 14, 2019

jondo commented on Jun 2, 2021

tacaswell commented on Jun 3, 2021

tacaswell commented on Jun 21, 2021

logari81 commented on Aug 2, 2022

oscargus commented on Aug 2, 2022

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions