Matplotlib plot over imshow while keeping axis size

Question

Whenever I use imshow() to plot an image, plotting 1D data over it in a twinned bottom x-axis changes the size and aspect ratio of the initial x-axis created with for imshow(). How do I avoid this behavior? Here is how to reproduce the issue:

import numpy as np
import matplotlib
matplotlib.use('macosx')
import matplotlib.pyplot as plt
im = np.random.rand(2856, 4290)
light_curve = im[1000, :]
fig = plt.figure(1, figsize=(10,10))
ax1 = plt.subplot(2,1,1)
ax1.imshow(im, cmap='gray', origin='lower')
ax2 = plt.subplot(2,1,2)
ax2.imshow(im, cmap='gray', origin='lower')
# Setting aspect ratio to equal does not help
ax2.set_aspect('equal')
ax21 = ax2.twinx()
ax21.plot(light_curve, alpha=0.7)
# Setting axis limits does not help
ax1.axis([0, im.shape[1], 0, im.shape[0]])
ax21.set_xlim([0, im.shape[1]])

And here is what it looks like with my graphical backend (macosx, if that is of any relevance)

Top: image with imshow(). Bottom: plot overlaid, image axis aspect and size changed

Isn't it the purpose of twinx() used above to help with this in the first place? So how may I keep the initial imshow() x-axis fixed and have the subsequent axis of the 1D plot simply fit, without resizing or messing with the aspect ratio, without going completely manual with building my axes?

Answer 1

It is indeed a bit unfortunate that the aspect does not propagate to the twin axes in the sense that it would have the same box around it.

I think the only way to overcome this is to calculate the aspect manually and set it for the twin axes as well.

import numpy as np
import matplotlib.pyplot as plt
im = np.random.rand(285, 429)
light_curve = im[100, :]
fig = plt.figure(1, figsize=(8,8))
ax1 = plt.subplot(2,1,1)
ax1.imshow(im, cmap='gray', origin='lower')
ax2 = plt.subplot(2,1,2)
ax2.imshow(im, cmap='gray', origin='lower')
ax2.set_aspect("equal", "box-forced")
ax21 = ax2.twinx()
ax21.plot(light_curve, alpha=0.7)
# Setting axis limits does not help
ax21.set_xlim(ax1.get_xlim())
a = np.diff(ax21.get_ylim())[0]/np.diff(ax1.get_xlim())*im.shape[1]/im.shape[0]
ax21.set_aspect(1./a, "box-forced")
plt.show()

enter image description here

• Wall-E: Thanks, it works nicely. I'll play a bit more with it to see how stable things are if I play with either axis.