Matplotlib external axes ruin subplot layout

Question

I would like to plot a figure with 3 subplots. The middle one has 3 different x-axes, one of which is detached and placed below the subplot. When I use Gridspec for the layout the plot areas are spaced equidistantly, but the padding between the axes labels of different subplots is hugely different:

Here is the code to reproduce the figure:

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec

fig = plt.figure(figsize=(3.375, 6.5))
gs0 = gridspec.GridSpec(3, 1, figure=fig)

ax0 = fig.add_subplot(gs0[0])
ax0.set_xlabel('x label 0')

ax1 = fig.add_subplot(gs0[1])
ax1.set_xlabel('x label 1a')
secax1 = ax1.twiny()
secax1.xaxis.set_ticks_position('bottom')
secax1.xaxis.set_label_position('bottom')
secax1.spines['bottom'].set_position(('outward', 40))
secax1.set_xlabel('x label 1b')
thax1 = ax1.twiny()
thax1.set_xlabel('x label 1c')

ax2 = fig.add_subplot(gs0[2])
ax2.set_xlabel('x label 2a')
ax2.set_ylabel('y label 2')
secax2 = ax2.twiny()
secax2.set_xlabel('x label 2b')

plt.tight_layout()
plt.savefig('3 subplots same size.png', dpi=300)
plt.show()

I'm looking for a way to either make the spacing between the complete subfigures equal, with everything like the additional axes and their labels. Or a way to manually shift the subplots within the grid. The subplots don't need to maintain the same size.

I tried changing the height_ratios as

gs0 = gridspec.GridSpec(3, 1, figure=fig, height_ratios=[1, 1.5, 1])

but it doesn't affect the spaces between the plots.

Not sure it works, but did you try constrained_layout instead of tight_layout? — Jody Klymak, Mar 17 '21 at 14:44
@JodyKlymak Nice, this takes care of this issue automatically. I hope it will soon be out of the [experimental phase](https://matplotlib.org/stable/tutorials/intermediate/constrainedlayout_guide.html?highlight=constrained_layout#codecell1:~:text=Warning,ax.set_position()%20in%20your%20code%20with%20constrained_layout%3DFalse.). — A. P., Mar 17 '21 at 14:58

DDaly · Answer 1 · 2021-03-17T14:11:51.123

You could use plt.subplots and put invisiable "gap plots" in between your real plots then you can adjust the gaps by changing the height_ratios of the gap plots

f, (ax0, gap1, ax1, gap2, ax2) = plt.subplots(5, 1,figsize=(3.375,8), gridspec_kw={'height_ratios': [1,1,1,1.75,1]})


gap1.axis('off')# Make the gap plots invisable
gap2.axis('off')# Make the gap plots invisable

#ax0 = fig.add_subplot(gs0[0])
ax0.set_xlabel('x label 0')

#ax1 = fig.add_subplot(gs0[2])
ax1.set_xlabel('x label 1a')
secax1 = ax1.twiny()
secax1.xaxis.set_ticks_position('bottom')
secax1.xaxis.set_label_position('bottom')
secax1.spines['bottom'].set_position(('outward', 40))
secax1.set_xlabel('x label 1b')
thax1 = ax1.twiny()
thax1.set_xlabel('x label 1c')

#ax2 = fig.add_subplot(gs0[5])
ax2.set_xlabel('x label 2a')
ax2.set_ylabel('y label 2')
secax2 = ax2.twiny()
secax2.set_xlabel('x label 2b')

Output

Awesome, this works. With the `height_ratio` one can manually shift the figures around. If the y-axis label should not be clipped one needs to enable the tight_layout again, which let's you change the vertical distance between the subplots. I get a good layout if I add `plt.tight_layout(h_pad=-8.0)` after the rest. But admittedly, the solution with `constrained_layout` is more convenient. — A. P., Mar 17 '21 at 14:57

score 1 · Accepted Answer · answered Mar 17 '21 at 15:05

This kind of thing is where constrained_layout does better than tight_layout. tight_layout only allows one margin size, so makes lots of room between the rows of subplots. constrained_layout keeps one upper and lower margin per row of the gridspec.

Yes, constrained_layout is marked experimental. That is so its behaviour can be changed without warning. But the API is not likely to change.

import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec

fig, (ax0, ax1, ax2)  = plt.subplots(3,1, figsize=(3.375, 6.5), 
                                   constrained_layout=True)

ax0.set_xlabel('x label 0')

ax1.set_xlabel('x label 1a')
secax1 = ax1.twiny()
secax1.xaxis.set_ticks_position('bottom')
secax1.xaxis.set_label_position('bottom')
secax1.spines['bottom'].set_position(('outward', 40))
secax1.set_xlabel('x label 1b')
thax1 = ax1.twiny()
thax1.set_xlabel('x label 1c')

ax2.set_xlabel('x label 2a')
ax2.set_ylabel('y label 2')
secax2 = ax2.twiny()
secax2.set_xlabel('x label 2b')

Matplotlib external axes ruin subplot layout

2 Answers2