Skip to content

Surprising alignment of (multiple) colorbars with constrained layout active #10627

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
afvincent opened this issue Feb 28, 2018 · 2 comments
Closed

Surprising alignment of (multiple) colorbars with constrained layout active #10627

afvincent opened this issue Feb 28, 2018 · 2 comments
Labels
topic: geometry manager LayoutEngine, Constrained layout, Tight layout
Milestone
v2.2.0

Comments

@afvincent
Copy link
Contributor

afvincent commented Feb 28, 2018 •
edited
Loading

Bug report

Bug summary

The new experimental layout manager seems to produce a rather surprising alignment of two colorbars when one is shared by several Axes instances, and another one is simply stealing space from a single Axes instance. It seems like @jklymak might have a fix in mind, but he also have a day-job ^^: milestoning for 2.2 just in case he manages to have a shot in time for it.

NB: originally from Gitter.

Code for reproduction

import matplotlib.pyplot as plt
import numpy as np

# Dummy data

arr_shape = (64, 256)

vmin, vmax = 80.0, 120.0
arr_A = np.random.uniform(low=vmin, high=vmax, size=arr_shape)
arr_B = np.random.uniform(low=vmin, high=vmax, size=arr_shape)

rel_error = 100 * (arr_A - arr_B) / arr_A

# Plotting time!

fig, axs = plt.subplots(3, 1, constrained_layout=True)

# Raw data arrays, with a shared colorbar
for ax, arr, lbl in ((axs[0], arr_A, "Case A"), (axs[1], arr_B, "Case B")):
    _im0 = ax.imshow(arr, vmin=vmin, vmax=vmax)
    ax.set_title(lbl)
fig.colorbar(_im0, ax=axs[:2], label="Raw Data (a.u.)", aspect=2*20, pad=-12)

# Relative difference at the bottom, with its own colorbar
ax = axs[2]
_im1 = ax.imshow(rel_error, cmap="coolwarm",
                vmin=min(rel_error.min(), -rel_error.max()),
                vmax=max(rel_error.max(), -rel_error.min()))
ax.set_title("(A - B) / A")
fig.colorbar(_im1, ax=axs[2], label="Difference (%)", aspect=20)

# Cosmeticks
for ax in axs.flat:
    ax.set_xticks([])
    ax.set_yticks([])

plt.show()

@jklymak I know, I know, it is not really “minimal”, but I was originally planning to show a use case rather than a MWE. And at least, it is a standalone example 😈.

Actual outcome

constrained_multi_cbars

Expected outcome

Something closer to
constrained_multi_cbars_workaround

Workaround
The figure in the Expected outcome section was produced by passing the supplementary parameter “pad=-12” in the relevant line:

fig.colorbar(_im0, ax=axs[:2], label="Raw Data (a.u.)", aspect=2*20, pad=-12)

Matplotlib version

  • Operating system: Linux (Fedora 27)
  • Matplotlib version: 2.2.0rc1 from conda
  • Python version: 3.6 from conda

Edit: English
@afvincent afvincent added this to the v2.2.0 milestone Feb 28, 2018
@jklymak
Copy link
Member

jklymak commented Feb 28, 2018 •
edited
Loading

Thanks @afvincent

That is not good. OTOH, there is a bit of an API problem in how to not make it happen.

What is happening here:

If you specify only one axes as the parent (your lowest axes), constrained_layout attaches the colorbar to the axes, inside the subplotspec.

If you specify a colorbar for multiple axes (your upper two), it attaches the colorbar to the gridspec, and shrinks all the subplotspec boxes of the gridspec to make room.

In your case, the bottom axes and colorbar are all inside the subplotbox for that axes, and the other colorbar is outside all the subplot boxes. So, constrained_layout is doing what I told it to!

And we need this for the case where you have (for instance) two columns of axes, and attach a colorbar to an axes in the left column, then you don't want that colorbar to the right of all the subplots in the gridspec, you want it with the axes.

I'd propose as a solution that we pass a gridspec-aware colorbar as an itterable so:

fig.colorbar(_im0, ax=[axs[2]])

instead of

fig.colorbar(_im0, ax=axs[2])

would get you what you want. Right now all do is check that there is one element in the argument.

Thats a bit obscure, but at least it'd stop playing with padding...

@jklymak jklymak mentioned this issue Feb 28, 2018
Merged
3 tasks
@jklymak jklymak added the topic: geometry manager LayoutEngine, Constrained layout, Tight layout label Feb 28, 2018
@efiring
Copy link
Member

efiring commented Mar 3, 2018

Closed by #10629.

@efiring efiring closed this as completed Mar 3, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
topic: geometry manager LayoutEngine, Constrained layout, Tight layout
Projects
None yet
Milestone
v2.2.0
Development

No branches or pull requests
3 participants
@efiring @jklymak @afvincent