Combining two matplotlib colormaps

Question

I would like to merge two colormaps into one, such that I can use one cmap for negative values and the other one for positive values.

At the moment I do it with masked arrays and plotting one image with one cmap and the other image with the other, resulting in:

with the following data

dat = np.random.rand(10,10) * 2 - 1
pos = np.ma.masked_array(dat, dat<0)
neg = np.ma.masked_array(dat, dat>=0)

I plotted pos with gist_heat_r and neg with binary.

I would like to have a single colorbar with the combined cmap's, so this is not the correct approach for me.

So, how do I take two existing cmaps's and merge them into one?

EDIT: I admit, this is a duplicate, but the answer that's given is much more clear here. Also the example images make it more clear.

Scintillator

Interesting question. But for your case, you would normally use a diverging colormap (two hues, one for positive numbers another for negative ones). your combination of reds and greys is available in plt.cm.RdGy. This yields a continuous map around zero. If I'm not mistaken you have a discontinuity at 0. Is that what you want? – Temperature

Thank you for your answer, but that is not what I want. The positive and negative values mean something else. And to stay coherent in the paper I'm writing, I would like to use the same cmap in all pictures, which it the one for positive values. – Scintillator

possible duplicate of stacking colormaps – Boysenberry

Answer 1

Colormaps are basically just interpolation functions which you can call. They map values from the interval [0,1] to colors. So you can just sample colors from both maps and then combine them:

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors

data = np.random.rand(10,10) * 2 - 1

# sample the colormaps that you want to use. Use 128 from each so we get 256
# colors in total
colors1 = plt.cm.binary(np.linspace(0., 1, 128))
colors2 = plt.cm.gist_heat_r(np.linspace(0, 1, 128))

# combine them and build a new colormap
colors = np.vstack((colors1, colors2))
mymap = mcolors.LinearSegmentedColormap.from_list('my_colormap', colors)

plt.pcolor(data, cmap=mymap)
plt.colorbar()
plt.show()

Result:

NOTE: I understand that you might have specific needs for this, but in my opinion this is not a good approach: How will you distinguish -0.1 from 0.9? -0.9 from 0.1?

One way to prevent this is to sample the maps only from ~0.2 to ~0.8 (e.g.: colors1 = plt.cm.binary(np.linspace(0.2, 0.8, 128))) so they wont go all the way up to black:

Temperature

This is exactly what I wanted, thanks! Clean solution, perfect! – Scintillator

Any idea how to do it in a function where the two colors maps names are given as parameters? plt.cm[color_map_name] doesn't seem to to work – Willful

Regarding the worry on same different values mapping to same color. Some times this is desired, such as if your data is an angle. Then 0=2*pi for instance. In that case to choose colors as colors1 = plt.cm.gray(np.linspace(0., 1, 128)) and colors2 = plt.cm.hot_r(np.linspace(0., 1, 128)) makes a lot of sense. – Cheer

@NoamPeled use getattr. Like such (adding the predefined matplotlib colormap to your list of colormaps to stack): cmaplist.append(getattr(plt.cm,'coolwarm')(np.linspace(0,1,256,endpoint=False)+0.5/256)). Note for my purposes I needed some specific endpoint customization to avoid overlaps. ( wanted to have multiple 1-D heatmaps with different colormaps on a single pcolormesh. Here for details: #43761930 – Curiosa

This is great. I am trying this approach for a scatter plot, but the colorbar has a strange effect with the split going above 0 (picture). Any ideas? – Tatia

Answer 2

I agree with the accepted answer but in order not to sample just between ~0.2 to ~0.8, you can for example reverse the binary colormap. Here is how to do it.

colors1 = plt.cm.binary_r(np.linspace(0., 1, 128))
colors2 = plt.cm.gist_heat_r(np.linspace(0, 1, 128))

colors = np.vstack((colors1, colors2))
mymap = mcolors.LinearSegmentedColormap.from_list('my_colormap', colors)

plt.pcolor(data, cmap=mymap)
plt.colorbar()
plt.show() 

The main change is replacing plt.cm.binary with plt.cm.binary_r.

Stanger