Matplotlib Subplots Are Too Narrow With Tight Layout

Question

I am currently trying to plot many subplots in Matplotlib (Python 3.6, Matplotlib 2.0.0) using GridSpec. Here is the minimal working example:

import matplotlib.pyplot as plt
from matplotlib.gridspec import *

# Color vector for scatter plot points
preds = np.random.randint(2, size=100000)

# Setup the scatter plots
fig = plt.figure(figsize=(8,8))
grid = GridSpec(9, 9)

# Create the scatter plots
for ii in np.arange(0, 9):
    for jj in np.arange(0, 9):
        if (ii > jj):
            ax = fig.add_subplot(grid[ii, jj])
            x = np.random.rand(100000)*2000
            y = np.random.rand(100000)*2000
            ax.scatter(x, y, c=preds)

This is the result without any modifications:

Of course the spacing between subplots is unsatisfactory so I did what I usually do and used tight_layout(). But as can be seen in the figure below, tight_layout() squeezes the width of the plots unacceptably:

Instead of using tight_layout(), I figured I should just adjust the subplots manually using subplots_adjust(). Below is the figure with subplots_adjust(hspace=1.0, wspace=1.0).

The result is almost correct, and with a little more tweaking the space between subplots would be perfect. However the subplots appear too small to adequately convey information.

Is there a better way to get proper spacing between subplots while still maintaining aspect ratio and a large enough subplot size? The only possible solution I could come up with was to use subplots_adjust() with a larger figsize, but this results in a very large space between the edges of the figure and the subplots.

Any solutions are appreciated.

Answer 1

As all your axes have the same x and y ranges, I would choose to show the tick labels only on the outer Axes. For a grid of equally-sized subplots, this is easily automated with the sharex and sharey keywords of plt.subplots(). Of course, if you set up a grid of 9x9 subplots, that gives you more plots than you want, but you can either make the redundant plots invisible (for instance with Axes.set_visible or remove them entirely. In the example below I go with the latter.

from matplotlib import pyplot as plt
import numpy as np

fig, axes = plt.subplots(
    nrows=9, ncols=9, sharex=True, sharey=True, figsize = (8,8)
)

# Color vector for scatter plot points
preds = np.random.randint(2, size=1000)

# Create the scatter plots
for ii in np.arange(0, 9):
    for jj in np.arange(0, 9):
        if (ii > jj):
            ax = axes[ii,jj]
            x = np.random.rand(1000)*100
            y = np.random.rand(1000)*2000
            ax.scatter(x, y, c=preds)
        else:
            axes[ii,jj].remove() ##remove Axes from fig
            axes[ii,jj] = None   ##make sure that there are no 'dangling' references.    

plt.show()

The resulting figure looks like this:

This can be of course adjusted further with something like subplots_adjust(). Hope this helps.