Unexpected behavior when explicitly providing colors and unevenly spaced levels

[AWSisco]

Description

When supplying my own colors and unevenly spaced levels explicitly, I observe an unexpected behavior where some colors are omitted and others are repeated.

Steps to reproduce

import proplot as pplt
import numpy as np

N=20
state=np.random.RandomState(51423)
data=np.cumsum(state.rand(N, N), axis=1)

colors=['white','indigo1','indigo3','indigo5','indigo7','indigo9','yellow1','yellow3','yellow5','yellow7','yellow9','violet1','violet3']

fig,axs=pplt.subplots(nrows=2, ncols=1, refheight=3.5)

axs[0].pcolormesh(data, levels=[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12],
                  colors=colors, colorbar='r', extend='both', colorbar_kw={'label':'Expected color distribution'})

axs[1].pcolormesh(data,levels=[1., 1.25, 1.5, 2., 2.5, 3., 3.75, 4.5, 6., 7.5, 9., 12.],
                  colors=colors, colorbar='r', extend='both', colorbar_kw={'label':'Unexpected color distribution'})

Expected behavior: I expected that all 13 named colors would still show when mapped to the unevenly spaced levels on axs[1].

Actual behavior: In this example on axs[1], some colors are omitted and replaced by others. I also observe this issue when providing my own RGB tuples.

Equivalent steps in matplotlib

Explicitly providing named colors and unevenly spaced levels in Matplotlib gives the desired behavior.

Proplot version

ProPlot 0.9.5
Matplotlib 3.4.3
Python 3.9.7

[lukelbd]

Huh -- I think this has something to do with SegmentedNorm, which is applied by default whenever you pass an unevenly-spaced levels list. Does passing norm='linear' to pcolormesh fix the problem? Or alternatively... perhaps SegmentedNorm is not getting applied when it should be, so you could try norm='segmented'.

[lukelbd]

Here's a demo of the effect of Normalize (the default linear one) vs. SegmentedNorm. This kind of seems like SegmentedNorm is somehow not getting applied when it should be... resulting in uneven steps across your color list. You could also try manually passing norm=pplt.SegmentedNorm(levels) if norm='segmented' does not work.

[AWSisco]

Explicitly specifying the norm doesn't seem to resolve the issue.

import proplot as pplt
import numpy as np

N=20
state=np.random.RandomState(51423)
data=np.cumsum(state.rand(N, N), axis=1)

colors=['white','indigo1','indigo3','indigo5','indigo7','indigo9','yellow1','yellow3','yellow5','yellow7','yellow9','violet1','violet3']
levels_even = pplt.arange(1,12,1)
levels_uneven=[1., 1.25, 1.5, 2., 2.5, 3., 3.75, 4.5, 6., 7.5, 9., 12.]

fig,axs=pplt.subplots(nrows=2, ncols=2, refwidth=3.)

axs[0].pcolormesh(data, levels=levels_even,   colors=colors, colorbar='r', extend='both')
axs[1].pcolormesh(data, levels=levels_uneven, colors=colors, colorbar='r', extend='both', norm='linear')
axs[2].pcolormesh(data, levels=levels_uneven, colors=colors, colorbar='r', extend='both', norm='segmented')
axs[3].pcolormesh(data,                       colors=colors, colorbar='r', extend='both', norm=pplt.SegmentedNorm(levels_uneven))

t = ['Even levels','norm=\'linear\'','norm=\'segmented\'','pplt.SegmentedNorm(levels_uneven)']
for i, ax in enumerate(axs):
    ax.format(title=t[i], titleloc='l')

[AWSisco]

After some additional experimenting, I managed to create the desired behavior by manually constructing my cmap and norm using matplotlib.colors.ListedColormap and Matplotlib.colors.BoundaryNorm, respectively. I borrowed some of this from #190, so perhaps my issue is related?

The values plotted on axs[1] are the actual values of the underlying data. I'm not fluent in what's happening under the hood in ProPlot, but maybe this will give some hints to those who are.

import proplot as pplt
import numpy as np
import matplotlib.colors as mcolors

#Some data to plot
N=20
state=np.random.RandomState(51423)
data=np.cumsum(state.rand(N, N), axis=1)

#Colors and levels I want
colors=['white','indigo1','indigo3','indigo5','indigo7','indigo9','yellow1','yellow3','yellow5','yellow7','yellow9','violet1','violet3']
levels_uneven=[1., 1.25, 1.5, 2., 2.5, 3., 3.75, 4.5, 6., 7.5, 9., 12.]

#Manually create a colormap from the named colors
colors_list = []
for c in colors:
    colors_list.append(pplt.to_rgb(c))
manual_cmap = mcolors.ListedColormap(colors_list)

norm = mcolors.BoundaryNorm(boundaries=np.array(levels_uneven), ncolors=len(levels_uneven)-1)

fig,axs=pplt.subplots(nrows=1, ncols=2, refwidth=5)

axs[0].pcolormesh(data, levels=levels_uneven, colors=colors, colorbar='r', extend='both', labels=True, precision=2, labels_kw={'size':6})
axs[1].pcolormesh(data, cmap=manual_cmap, norm=norm, colorbar='r', extend='both', labels=True, precision=2, labels_kw={'size':6})

axs[0].format(title='Unexpected behavior', titleloc='l')
axs[1].format(title='Desired behavior with a cmap and norm work-around ', titleloc='l')

[lukelbd]

Thanks for the helpful examples! This turned out to be a very quick fix -- it's on the master branch now (2ea4665). You can install with pip install git+https://github.com/lukelbd/proplot.git.

[lukelbd]

Here's your example after the update. Note some ticks on the right "linear" colorbar are missing by default (to prevent overcrowding) while ticks on the left SegmentedNorm colorbar are all shown (to avoid hiding the locations of non-linear thresholds in the scaling).

import proplot as pplt
import numpy as np
N=20
state=np.random.RandomState(51423)
data=np.cumsum(state.rand(N, N), axis=1)
colors=['white','indigo1','indigo3','indigo5','indigo7','indigo9','yellow1','yellow3','yellow5','yellow7','yellow9','violet1','violet3']
levels_even = pplt.arange(1,12,1)
levels_uneven=[1., 1.25, 1.5, 2., 2.5, 3., 3.75, 4.5, 6., 7.5, 9., 12.]
fig, axs = pplt.subplots(ncols=2, refwidth=3.)
axs[0].pcolor(np.random.rand(10, 10) * 12, levels=levels_uneven, colors=colors, colorbar='r', extend='both')
axs[1].pcolor(np.random.rand(10, 10) * 12, levels=levels_even, colors=colors, colorbar='r', extend='both')

[AWSisco]

Thanks for such a quick turnaround on getting this resolved. Much appreciated!