Colorbars on Projections may or may not work

[natgeo-wong]

Hi, I tried to plot colorbars on projections. However, I'm getting different results on different computers.

On my work computer, I get a nicely-proportioned colorbar

But on my home computer, I get a weirdly proportioned colorbar that indicates to me that the colorbar was plotted on the initial square (default aspect = 1) and then shrunk down.

I'm not too sure what is wrong here though. I have proplot installed the same way on both my computers.

[lukelbd]

Very strange. Thanks for the report. It seems proplot is failing to properly scale the colorbar triangle "extensions" and failing to restrict the tick locations, because it thinks the subplots are taller than they really are (in "physical" units, not pixels). You could be right about the colorbars getting "squashed" -- or it could just be getting the height wrong for some other reason.

For now, as a workaround, you can pass e.g. extendfrac=0.05 to colorbar() to uniformly scale the extensions (by default, proplot uses extendsize, which tries to specify extensions in physical units [e.g. font size-widths] rather than fractional units -- in your example, extendsize is failing). You can also manually specify the ticks with e.g. ticks=1 and ticks=5.

Do you mind sharing the proplot version, matplotlib backend, and figure resolutions you're using on each workstation? You can copy the following code:

import proplot as pplt
print('Version', pplt.__version__)
print('Backend', pplt.rc.backend)
print('Figure dpi', pplt.rc.figuredpi)
print('Save dpi', ppt.rc.savefigdpi)

Also, do you mind printing the size in "inches" of the figures and axes in your example? You can do so as follows:

import proplot as pplt
fig, axs = pplt.subplots()  # for example
# ...
# your plotting code here
# ...
print('Figure', fig.get_size_inches())
for ax in axs:
    print('Axes', ax.number, ax._get_size_inches())

[natgeo-wong]

Hi! This might be some issue with how Julia using Python.

I did a raw test on the actual python installation and it looks okay. But it looks like somewhere in the pipeline where Julia calls Python (I use Julia for most of my work) we run into this problem.

So this is technically an issue outside the scope here.

[lukelbd]

Thanks for the update. Always cool to hear from people using proplot in Julia -- never imagined that was possible! In principle I think this is a legit issue though, since overcrowding / undercrowding of "discrete" colorbar ticks will always happen if the figure size changes after the colorbar is drawn.

I decided to fix this by introducing DiscreteLocator. This is sort of like FixedLocator and AutoLocator at once -- it selects from a subset of locations at the level boundaries depending on the axis size (see below examples). Previously, this only worked for static figure sizes, but now you can interactively resize a window and the ticks should auto-adjust.

Re-applying extendsize every time the figure size changes would be more complicated... and I suspect expensive. I'd have to redraw the entire colorbar solids with Colorbar.draw_all() every time. So I think I'll leave that alone for now and close this issue. Overlapping tick labels was definitely the more glaring problem in your example anyway.

import proplot as pplt
levels = pplt.arange(0, 2, 0.1)
data = np.random.rand(5, 5) * 2
for width in (0.8, 1.2, 2):
    fig, axs = pplt.subplots(share=False, ncols=2, nrows=2, refwidth=width)
    for i, ax in enumerate(axs):
        cmd = ax.contourf if i // 2 == 0 else ax.pcolormesh
        m = cmd(data, levels=levels, extend='both')
        ax.colorbar(m, loc='t' if i // 2 == 0 else 'b')
        ax.colorbar(m, loc='l' if i % 2 == 0 else 'r')