blocks.Pcolormesh seems to have different behavior than plt.pcolormesh

[Xin-yang-Liu]

I am not sure if it is a feature or a bug.
When I was playing with the example code:

import numpy as np
import matplotlib.pyplot as plt
import animatplot as amp


x = np.linspace(-2, 2, 5)
y = np.linspace(-2, 2, 5)
t = np.linspace(0, 2*np.pi, 40)

X, Y, T = np.meshgrid(x, y, t)

Z = np.sin(X*X+Y*Y-T)

block = amp.blocks.Pcolormesh(X[:,:,0], Y[:,:,0], Z, t_axis=2, cmap='RdBu')
plt.colorbar(block.quad)
plt.gca().set_aspect('equal')

anim = amp.Animation([block], amp.Timeline(t))

anim.controls()
plt.show()

I used a very coarse grid 55 instead of the original 5050, so that it is more clear.
The data array Z is of size 55 at a certain time step. But the generated animation shows a grid of shape 44.
It seems blocks.Pcolormesh use the value in Z as the "node" value in animation, which is different in plt.pcolormesh whereZ value is treated as the "mesh cell" value.

[johnomotani]

What versions of animatplot and matplotlib are you using? For me with animatplot-0.4.2 and matplotlib-3.3.3, from the following the pcolormesh plot is identical (apart from tick labels) to the first frame of the animation.

import numpy as np
import matplotlib.pyplot as plt
import animatplot as amp


x = np.linspace(-2, 2, 5)
y = np.linspace(-2, 2, 5)
t = np.linspace(0, 2*np.pi, 40)

X, Y, T = np.meshgrid(x, y, t)

Z = np.sin(X*X+Y*Y-T)

plt.pcolormesh(X[:,:,0], Y[:,:,0], Z[:,:,0], cmap='RdBu')
plt.colorbar()


plt.figure()

block = amp.blocks.Pcolormesh(X[:,:,0], Y[:,:,0], Z, t_axis=2, cmap='RdBu')
plt.colorbar(block.quad)
plt.gca().set_aspect('equal')

anim = amp.Animation([block], amp.Timeline(t))

anim.controls()
plt.show()

We did change Pcolormesh recently though: #53

[Xin-yang-Liu]

I am using the exact same version of animatplot and matplotlib as you.


the first figure is generated using plt.pcolormesh at time 0. The grid is 5 by 5.
In the animation, the grid is 4*4

[johnomotani]

Have you changed the default value of matplotlib.rcParams["pcolor.shading"]? Your output is what I get with plt.pcolormesh(X[:,:,0], Y[:,:,0], Z[:,:,0], shading="nearest"). shading="auto" (which should be equivalent to shading="nearest" for this case) will become the matplotlib default at some point, but for the moment the default is still shading="flat". We should pick up the default from matplotlib though, here

animatplot/animatplot/blocks/image_like.py

Line 56 in 26b3c79

self. shading = kwargs.get("shading", plt.rcParams.get("pcolor.shading", "flat"))

so I'm confused.

animatplot is behaving as I'd expect, it's your matplotlib plot that seems odd. Maybe you could try creating a clean virtual environment (or conda environment) and make the plt.pcolormesh plot again?

[Xin-yang-Liu]

I did not change any default value.
I also tried in a clean conda environment (python 3.7.8, matplotlib 3.3.3, numpy 1.19.4) and got the same figure.

[Xin-yang-Liu]

Actually I was using plt.pcolormesh(Z[:,:,0]) and got the figure with shading="nearest"(5 * 5 grid) by default
Besides, I tried plt.pcolormesh(X[:,:,0],Y[:,:,0],Z[:,:,0]). It seems the default value becomes shading="flat" (4 * 4 grid)

[johnomotani]

animatplot at the moment doesn't support calling like Pcolormesh(Z) without X and Y arrays. I think a PR adding that functionality would be welcome.

The current behaviour is consistent with matplotlib's pcolormesh when passing X, Y and Z, so this isn't a bug. The matplotlib behaviour without X and Y arguments is like calling (if Z is an array of shape (n,m))

plt.pcolormesh(np.arange(m+1), np.arange(n+1), Z)