DiscreteNorm does not support set_under/set_over colormap extremes

[lacstorm]

[Description of the bug or feature.]
Error when using boudary normalization for colormap.

Steps to reproduce

A "Minimal, Complete and Verifiable Example" will make it much easier for maintainers to help you.

import matplotlib.pyplot as plt
import proplot as pplt
import numpy as np
import cmaps
import matplotlib.colors as colors
import matplotlib.ticker as ticker


state = np.random.RandomState(51423)
data = (state.rand(20, 20))*50
data[5:15,5:15] = 0.002
data[10:20,10:20] = 0.6
print(data.max(),data.min())

levels = [0.1,0.5,1.,2.,3.,4.,5.,6.,8.,10.,20.,40]
norm = colors.BoundaryNorm(boundaries=np.array(levels), ncolors=len(levels)-1)

# figure with proplot
fig, ax = pplt.subplots()
m0 = ax.contourf(data, levels=levels, extend='both', cmap=cmaps.cmorph,
colorbar="r",colorbar_kw={'ticks':levels})
ax.format(suptitle='plot with proplot')
fig.savefig('color_proplot.png')

# figure with matplotlib
fig, ax = plt.subplots()
m1 = ax.contourf(data, levels=levels, norm=norm, extend='both',cmap=cmaps.cmorph[1:])
m1.cmap.set_under(cmaps.cmorph.colors[0])
cb = plt.colorbar(m1, ax=ax, ticks=levels, extendfrac='auto', 
                     aspect=18, shrink=.95)

cb.ax.minorticks_off()

ax.set_title("plot with matplotlib")
fig.savefig('color_matplotlib.png')

# figure with proplot and boudary normalization
fig, ax = pplt.subplots()
m0 = ax.contourf(data, levels=levels, norm=norm, extend='both', cmap=cmaps.cmorph[1:],
colorbar="r",colorbar_kw={'ticks':levels})
m0.cmap.set_under(cmaps.cmorph.colors[0])

ax.format(suptitle='plot with proplot1')
fig.savefig('color_proplot1.png')

Expected behavior: [What you expected to happen]

Actual behavior: [What actually happened]

When using default segmented normalizaton in proplot:

And when using boudary normailiztion of matplotlib:

Equivalent steps in matplotlib

See the code above.

Proplot version

0.6.1
Paste the result of import proplot; print(proplot.version) here.

[zxdawn]

@lacstorm It looks a little confused.

I would suggest you rewrite the example and state the problem in detail.
The example code should be a simple and minimal one.

1. The variables from wrfout* files can be replaced by numpy array
2. If the projection isn't necessary for this issue, you can just use Cartesian axis.
3. You can just attach the image by dragging that to the edit box

[lacstorm]

@zxdawn
Hi xin.
Thanks, I'll test and submit a simple case later!

[zxdawn]

@lacstorm You're welcome. PS: you can edit the original one by clicking the dot button on the right corner ;)

[lacstorm]

@zxdawn
Hi xin, I submit a very simple example.
Note that in the second figure, the colors for 20-40 and above 40 are the same.
And for the second plot, the colormap is the same with the first plot, but the contour is not right.

[lukelbd]

Hi @lacstorm,

I think this might have something to do with BoundaryNorm. Could you try without passing norm=norm to contourf? By default proplot automatically applies a "boundary norm" when the values in levels are unevenly spaced, so creating your own boundary norm is usually not necessary (see this section from the docs).

I also noticed you used set_under but did not user set_over. Adding m1.cmap.set_over(cmaps.cmorph.colors[-1]) might give you the desired result.

Also I can't reproduce the example because I think cmaps is a local module on your system. You could share it by uploading it to https://file.io or posting a google drive link.

[lukelbd]

Nevermind re: that first suggestion -- I see you already tried that.

Adding set_over(cmap.cmorph.colors[-1]) still might fix your problem. And I think this will also require using cmaps.cmorph[1:-1] instead of cmaps.cmorph[1:]. For example:

cmap = plot.Colormap(cmaps.cmorph[1:-1])
cmap.set_under(cmaps.cmorph[0])
cmap.set_over(cmaps.cmorph[-1])
ax.contourf(..., cmap=cmap)

[lukelbd]

To be clear, what is cmaps.cmorph? Is that a list of colors?

[lacstorm]

@lukelbd
Hi luke. I tried to do this like this:
(I tried to use only norm=norm or levels=levels, still ain't right.)

# figure with proplot and boudary normalization
fig1, ax1 = pplt.subplots()
m2 = ax1.contourf(data, levels=levels, norm=norm, extend='both', cmap=cmaps.cmorph[1:-1],
colorbar="r",colorbar_kw={'ticks':levels})
m2.cmap.set_under(cmaps.cmorph.colors[0])
m2.cmap.set_over(cmaps.cmorph.colors[-1])

ax1.format(suptitle='plot with proplot2')
fig1.savefig('color_proplot1.png')

And I still got the same result with "plot with proplot1":

Please note that the cmap.set_under is not working, values below 0.1 is contoured by cmaps.cmorph.colors[1].
cmaps is a package to transfer ncl or self_defined colormap to python, and could be found here.
The cmorph is a self-defined colormap, which is a list of 13 colors:

ncolors= 13
#  r   g   b
 255 255 255
   0 225 225
   0 165 185
   0 185   0
   0 255  45
 255 255   0
 225 205   0
 160 120   0
 160  40   0
 185   0   0
 255   0 155
 185   0 255
 125   0 255

You can test with the code below, so the cmorph colormap wont bother you.

import matplotlib.pyplot as plt
import proplot as pplt
import numpy as np
import matplotlib.colors as colors
import matplotlib.ticker as ticker


state = np.random.RandomState(51423)
data = (state.rand(20, 20))*50
data[5:15,5:15] = 0.002
data[10:20,10:20] = 0.6
print(data.max(),data.min())

# define colormap
cmorph = [[255.,255,255],
   [0,225,225],
   [0,165,185],
   [0,185, 0],
   [0 ,255 , 45],
 [255, 255 ,  0],
 [225 ,205 ,  0],
 [160, 120 ,  0],
 [160 , 40 ,  0],
 [185 ,  0 ,  0],
 [255  , 0 ,155],
 [185  , 0 ,255],
 [125 ,  0, 255]]

cmorph = np.array(cmorph)
cmorph = cmorph/255.
cmap0 = colors.ListedColormap(cmorph)

levels = [0.1,0.5,1.,2.,3.,4.,5.,6.,8.,10.,20.,40]
norm = colors.BoundaryNorm(boundaries=np.array(levels), ncolors=len(levels)-1)

# figure with proplot
fig, ax = pplt.subplots()
m0 = ax.contourf(data, levels=levels, extend='both', cmap=cmap0,
colorbar="r",colorbar_kw={'ticks':levels})
ax.format(suptitle='plot with proplot')
fig.savefig('color_proplot.png')

# figure with matplotlib
fig, ax = plt.subplots()
m1 = ax.contourf(data, levels=levels, norm=norm, extend='both',cmap=colors.ListedColormap(cmap0.colors[1:]))
m1.cmap.set_under(cmap0.colors[0])
cb = plt.colorbar(m1, ax=ax, ticks=levels, extendfrac='auto', 
                     aspect=18, shrink=.95)

cb.ax.minorticks_off()

ax.set_title("plot with matplotlib")
fig.savefig('color_matplotlib.png')

# figure with proplot and boudary normalization
fig1, ax1 = pplt.subplots()
m2 = ax1.contourf(data, levels=levels, norm=norm, extend='both', cmap=colors.ListedColormap(cmap0.colors[1:]),
colorbar="r",colorbar_kw={'ticks':levels})
m2.cmap.set_under(cmap0.colors[0])
m2.cmap.set_over(cmap0.colors[-1])

ax1.format(suptitle='plot with proplot2')
fig1.savefig('color_proplot1.png')

[lukelbd]

Sorry I couldn't get to this sooner, but I finally figured it out (see f23ff70). This was actually two separate bugs:

• plot.Colormap() automatically translates native matplotlib colormaps to proplot's special colormap subclasses plot.colors.LinearSegmentedColormap and plot.colors.ListedColormap. However when doing this, proplot failed to copy the "extreme" color values.
• plot.DiscreteNorm() failed to account for the possibility that there is a discrete transition between the minimum/maximum colors "within" the colormap range (colormap coordinates 0 and 1) and the out of bounds colors (coordianates 0 - eps and 1 + eps). Now it takes this into account.

Here's your example on master (can be installed with pip install git+https://github.com/lukelbd/proplot):

import matplotlib.pyplot as plt
import proplot as pplt
import numpy as np
import matplotlib.colors as colors
import matplotlib.ticker as ticker


state = np.random.RandomState(51423)
data = (state.rand(20, 20))*50
data[5:15,5:15] = 0.002
data[10:20,10:20] = 0.6
#print(data.max(),data.min())

# define colormap
cmorph = [[255.,255,255],
   [0,225,225],
   [0,165,185],
   [0,185, 0],
   [0 ,255 , 45],
 [255, 255 ,  0],
 [225 ,205 ,  0],
 [160, 120 ,  0],
 [160 , 40 ,  0],
 [185 ,  0 ,  0],
 [255  , 0 ,155],
 [185  , 0 ,255],
 [125 ,  0, 255]]

cmorph = np.array(cmorph) / 255.0
cmap0 = colors.ListedColormap(cmorph[1:-1])
cmap0.set_under(cmorph[0])
cmap0.set_over(cmorph[-1])
levels = [0.1,0.5,1.,2.,3.,4.,5.,6.,8.,10.,20.,40]

# Figure with proplot
fig, ax = pplt.subplots()
m0 = ax.contourf(
  data, levels=levels, extend='both', cmap=cmap0,
  colorbar='b', colorbar_kw={'ticks':levels}
)

ax.format(suptitle='plot with proplot')

[lacstorm]

@lukelbd
Hi, luke. It's ok now.
Thanks for developing such an effective package!