Repository navigation
Logit scale doesn't position x/ylabel correctly first draw #11386
Description
Bug report
Bug summary
Using constrained_layout or tight_layout in conjunction with a logit scale fails or crashes.
Constrained layout:
import matplotlib.pyplot as plt
fig, ax = plt.subplots(constrained_layout=True)
ax.plot([.2,.4,.6,.8],[.2, .4, .5, .9])
ax.set_xscale('logit')
plt.show()
results in a crash of python in Windows. I.e. it shows a pop up saying python needs to be closed because it crashed.
Tight layout
import matplotlib.pyplot as plt
fig, ax = plt.subplots()
ax.plot([.2,.4,.6,.8],[.2, .4, .5, .9])
ax.set_xscale('logit')
fig.tight_layout()
plt.show()
produces a long traceback for a ValueError: cannot convert float NaN to integer.
D:\...\envs\mplgit\lib\site-packages\numpy\core\_methods.py:29: RuntimeWarning: invalid value encountered in reduce
return umr_minimum(a, axis, None, out, keepdims)
D:\...\envs\mplgit\lib\site-packages\numpy\core\_methods.py:26: RuntimeWarning: invalid value encountered in reduce
return umr_maximum(a, axis, None, out, keepdims)
posx and posy should be finite values
posx and posy should be finite values
posx and posy should be finite values
posx and posy should be finite values
posx and posy should be finite values
posx and posy should be finite values
posx and posy should be finite values
Traceback (most recent call last):
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\backends\backend_qt5.py", line 513, in _draw_idle
self.draw()
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\backends\backend_agg.py", line 421, in draw
self.figure.draw(self.renderer)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\artist.py", line 50, in draw_wrapper
return draw(artist, renderer, *args, **kwargs)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\figure.py", line 1478, in draw
renderer, self, artists, self.suppressComposite)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\image.py", line 138, in _draw_list_compositing_images
a.draw(renderer)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\artist.py", line 50, in draw_wrapper
return draw(artist, renderer, *args, **kwargs)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\axes\_base.py", line 2583, in draw
mimage._draw_list_compositing_images(renderer, self, artists)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\image.py", line 138, in _draw_list_compositing_images
a.draw(renderer)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\artist.py", line 50, in draw_wrapper
return draw(artist, renderer, *args, **kwargs)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\axis.py", line 1177, in draw
ticks_to_draw = self._update_ticks(renderer)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\axis.py", line 1017, in _update_ticks
tick_tups = list(self.iter_ticks()) # iter_ticks calls the locator
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\axis.py", line 961, in iter_ticks
majorLocs = self.major.locator()
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\ticker.py", line 1922, in __call__
return self.tick_values(vmin, vmax)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\ticker.py", line 1930, in tick_values
locs = self._raw_ticks(vmin, vmax)
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\ticker.py", line 1876, in _raw_ticks
nbins = np.clip(self.axis.get_tick_space(),
File "d:\...\python\matplotlib\source\matplotlibgit\matplotlib\lib\matplotlib\axis.py", line 2503, in get_tick_space
return int(np.floor(length / size))
ValueError: cannot convert float NaN to integer
Expected outcome
Using either of above without layout manager gives the correct figure.
Also, using either of above without the line ax.set_xscale('logit') produces the expected output.
This may be related to #5456.
Matplotlib version
- Operating system: Windows 8.1
- Matplotlib version: 2.2.2.post1271+g0a70730fa
- Matplotlib backend (print(matplotlib.get_backend())): Qt5Agg, TkAgg
- Python version: 3.6
Activity
jklymak commented on Jun 7, 2018
I don’t have a computer to see what the issue could be. Off the top of my head I can’t see why either layout manager would care what the scale is so there must be something funky going on here...
fredrik-1 commented on Jun 8, 2018
I tried to look at it and my guess is that the problem is that the following spine consist of infs in the first call in axis._update_label_position. But I don't even know what a spine is. I just tried to find out when something started to look wrong.
subplotpars consist of np.nans after tight_layout. This seems to be due to axes. get_tightbbox and the above calculation in axis._update_label_position.
afvincent commented on Jun 8, 2018
@fredrik-1 Thanks :)! It may indeed be consistent with my quick look at the issue, which suggests that something is going wrong with the positions of some ticks (=> the attached text label(s) possibly receiving a non finite position, which could lead an “infinite” length somewhere).
FYI, spines are the “borders” of the axes. See for example this plot (bottom right)
jklymak commented on Jun 10, 2018
Logit is putting some text at -inf; I assume there is a 1/0 error in there somewhere.
jklymak commented on Jun 11, 2018
So, just to clarify this bug:
The error is in _update_label_position. The following returns a bbox with Bbox(x0=inf, y0=inf, x1=-inf, y1=-inf):
The reason is that for the first call to spine.get_path() it returns nonesense:
Path(array([[ 0., 13.], [ 0., 13.]]), None)and a logit scale doesn't deal well with the values of 13, which are greater than 1. (a/(1-a) == inf)
Interesting other scales return the same path; Of course the ylabel position shouldn't need to know anything about the y-extent of the path, so the bad value in the path doesn't normally affect the transform. But because it goes to inf, it screws up the matrix algebra of the transform and makes a mess for the Logit scale.
Note that on master, the logit scale does not display the ylabel on the first draw.
Subsequent draws, the Path is correct, and the ylabel appears.
The fudnamental issue is Spines.linear_spine sets the Path to 0, 13 or 13, 0. That choice doesn't matter, and is obviously just a placeholder, but I think should be less than 1.