Python: Shifted logarithmic colorbar, white color offset to center

Question

I am struggling to plot my data using pyplot.imshow. I use the 'RdBu_r' colormap, and I need the white color to be at value 1 on a logarithmic scale which is not centered at 1. But how can I do it? I tried 'center=1' which works for seaborn, but there is no such attribute in matplotlib. I also tried this:

import matplotlib.pyplot as plt
im=plt.imshow(proportion, cmap="RdBu_r", norm=LogNorm(), vmin=0.01, vmax=10)
axs=plt.gca()
cb = plt.colorbar(im, ax=axs,extend="both")

where proportion is my data array, ranging from 0.01 to 10. However there seems to be no way to specify that the white should be at 1 on this scale. Is there a way to do that?

Note again that I need to make use a gradient of colors here and a logarithmic normalization.

Do you want to center the white color on `1` on a logarithmic scale? Is the `LogNorm()` part of the problem or an attempted solution? There are some posts about defining a midpoint on a colorscale, [this one](https://stackoverflow.com/questions/20144529/shifted-colorbar-matplotlib), which is also now part of the [matplotlib documentation](https://matplotlib.org/users/colormapnorms.html#custom-normalization-two-linear-ranges), also [this one](https://stackoverflow.com/questions/7404116/defining-the-midpoint-of-a-colormap-in-matplotlib) which shows some alternative attempts. — ImportanceOfBeingErnest, Feb 05 '18 at 15:20

ImportanceOfBeingErnest · Accepted Answer · 2018-02-06T00:56:28.097

There are some questions and answers about defining a midpoint on a colorscale. Especially this one, which is also now part of the matplotlib documentation.

The idea is to subclass matplotlib.colors.Normalize and let it take a further argument midpoint. This can then be used to linearly interpolate the two ranges on either side of the midpoint to the ranges [0,0.5] and [0.5,1].

To have a midpoint on a logarithmic scale, we can in principle do the same thing, just that we subclass matplotlib.colors.LogNorm and take the logarithm of all values, then interpolate this logarithm on the ranges [0,0.5] and [0.5,1].

In the following example we have data between 0.001 and 10. Using the usual LogNorm this results in the middle of the colormap (white in the case of the RdBu colormap) to be at 0.1. If we want to have white at 1, we specify 1 as the midpoint in the MidPointLogNorm.

import numpy as np
import matplotlib.pyplot as plt
from  matplotlib.colors import LogNorm

x,y = np.meshgrid(np.linspace(-3,0,19), np.arange(10))
f = lambda x,y : 10**x*(1+y)
z = f(x,y)

fig, (ax,ax2) = plt.subplots(ncols=2, figsize=(12,4.8))

im = ax.pcolormesh(x,y,z, cmap="RdBu_r", norm=LogNorm(vmin=z.min(), vmax=z.max()))
fig.colorbar(im, ax=ax)
ax.set_title("LogNorm")

class MidPointLogNorm(LogNorm):
    def __init__(self, vmin=None, vmax=None, midpoint=None, clip=False):
        LogNorm.__init__(self,vmin=vmin, vmax=vmax, clip=clip)
        self.midpoint=midpoint
    def __call__(self, value, clip=None):
        # I'm ignoring masked values and all kinds of edge cases to make a
        # simple example...
        x, y = [np.log(self.vmin), np.log(self.midpoint), np.log(self.vmax)], [0, 0.5, 1]
        return np.ma.masked_array(np.interp(np.log(value), x, y))


im2 = ax2.pcolormesh(x,y,z, cmap="RdBu_r", 
                            norm=MidPointLogNorm(vmin=z.min(), vmax=z.max(), midpoint=1))
fig.colorbar(im2, ax=ax2)
ax2.set_title("MidPointLogNorm")
plt.show()

Updated solution which works for nan values: You need to replace the nan values by some value (best one outside the range of values from the array) then mask the array by those numbers. Inside the MidPointLogNorm we need to take care of nan values, as shown in this question.

import numpy as np
import matplotlib.pyplot as plt
from  matplotlib.colors import LogNorm

x,y = np.meshgrid(np.linspace(-3,0,19), np.arange(10))
f = lambda x,y : 10**x*(1+y)
z = f(x,y)
z[1:3,1:3] = np.NaN

#since nan values cannot be used on a log scale, we need to change them to 
# something other than nan, 
replace = np.nanmax(z)+900
z = np.where(np.isnan(z), replace, z)
# now we can mask the array
z = np.ma.masked_where(z == replace, z)

fig, (ax,ax2) = plt.subplots(ncols=2, figsize=(12,4.8))

im = ax.pcolormesh(x,y,z, cmap="RdBu_r", norm=LogNorm(vmin=z.min(), vmax=z.max()))
fig.colorbar(im, ax=ax)
ax.set_title("LogNorm")

class MidPointLogNorm(LogNorm):
    def __init__(self, vmin=None, vmax=None, midpoint=None, clip=False):
        LogNorm.__init__(self,vmin=vmin, vmax=vmax, clip=clip)
        self.midpoint=midpoint
    def __call__(self, value, clip=None):
        result, is_scalar = self.process_value(value)
        x, y = [np.log(self.vmin), np.log(self.midpoint), np.log(self.vmax)], [0, 0.5, 1]
        return np.ma.array(np.interp(np.log(value), x, y), mask=result.mask, copy=False)


im2 = ax2.pcolormesh(x,y,z, cmap="RdBu_r", 
                            norm=MidPointLogNorm(vmin=z.min(), vmax=z.max(), midpoint=1))
fig.colorbar(im2, ax=ax2)
ax2.set_title("MidPointLogNorm")
plt.show()

That is very helpful! Could you also comment on what should I do to avoid NaN values in my data array? — Tom Smith, Feb 06 '18 at 00:29
if I have several columns of nans which I need for my plot (otherwise I will have difficulties with axes and ticklabels), it says that all values must be positive... I can upload a picture if it makes it clearer... — Tom Smith, Feb 06 '18 at 00:33
You need to replace the nan values by some value number outside the range of values in the array, such that the logarithm can be taken of those values. I updated the answer. — ImportanceOfBeingErnest, Feb 06 '18 at 01:01

Python: Shifted logarithmic colorbar, white color offset to center

1 Answers1

Linked