Extreme spacing with large numbers of subplots

[tylerharvey]

Bug report

Bug summary

Subplots decrease in size with increasing numbers of subplots, but with non-square grids of subplots, the padding along the short axis is not correspondingly decreased. As a result, the padding between subplots grows extremely large with large numbers of subplots. This can be manually fixed with wspace and hspace in subplots_adjust(), but only for a given size of subplot, so a user needs to model the padding growth in order to adequately correct it. Tight_layout() only makes the spacing equally way too large along both axes.

Code for reproduction

import matplotlib.pyplot as plt
from numpy import random

n = 8
fig = plt.figure()
for i in range(2*n):
  ax = fig.add_subplot(2,n,i+1)
  ax.imshow(random.random((10,10)))
plt.savefig('{:0}_col_test.png'.format(n))
fig = plt.figure()
for i in range(2*n):
  ax = fig.add_subplot(n,2,i+1)
  ax.imshow(random.random((10,10)))
plt.savefig('{:0}_row_test.png'.format(n))

Actual outcome

With tight_layout():

Expected outcome

I would have expected constant figure padding with number of subplots.

Matplotlib version

• Operating system: Debian Stretch
• Matplotlib version: 3.0.2 (from pip)
• Matplotlib backend (print(matplotlib.get_backend())): TkAgg
• Python version: 3.5.3
• Jupyter version (if applicable):
• Other libraries:

[jklymak]

This is expected if you use imshow because the aspect ratio of the axes are 1:1. You can either make the size of the figure such that you don't have as much white space, or use axes_grid1 (https://matplotlib.org/gallery/axes_grid1/demo_axes_grid.html) which is does the layout by packing axes together as close as possible and leaving the white space on the margins.

The reason matplotlib doesn't do this by default is because usually folks want to make axes as big as possible and let the aspect ratio of the plot be whatever it needs to be to make things big. But that paradigm doesn't apply as well if axes have fixed data aspect ratios, hence the axes_grid1 toolkit was developed.

[tylerharvey]

Is the default behavior of tight_layout() here really expected? In most other cases, tight_layout() gets rid of unnecessary whitespace. In the case of a single plot with fixed aspect ratio, tight_layout() takes only the closer-fitting of the height or the width of the figure. In this case it just adds more whitespace.

You're definitely right that axes_grid1 solves this problem. Thanks. Maybe it'd help to add a link to it in the 'See also' box for subplot() and add_subplot().

[WeatherGod]

The issue isn't with tight_layout(). The issue is that imshow() imposes a default aspect ratio of 1:1. tight_layout() does not change the size of the figure, and does not change the aspect ratio of the axes objects, only the spacing between the axes objects. So, with that constraint, it can only maximize the axes objects while maintaining that aspect ratio. Now, it is possible to change the default aspect ratio, which is basically what axes_grid1 does (along with some other stuff). Then, tight_layout() will be free to maximize the axes objects without the aspect ratio constraint.

…

On Tue, Dec 18, 2018 at 6:31 PM tylerharvey ***@***.***> wrote: Is the default behavior of tight_layout() here really expected? In most other cases, tight_layout() gets rid of unnecessary whitespace. In the case of a single plot with fixed aspect ratio, tight_layout() takes only the closer-fitting of the height or the width of the figure. In this case it just adds more whitespace. You're definitely right that axes_grid1 solves this problem. Thanks. Maybe it'd help to add a link to it in the 'See also' box for subplot() and add_subplot(). — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#13018 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARy-L6lOMhP6V24s-rDKcdEmvSQS552ks5u6Xq9gaJpZM4ZZI9f> .

[jklymak]

fig.tight_layout(pad=0) will give you tighter spacing.

https://matplotlib.org/api/tight_layout_api.html