user308827 user308827 - 3 months ago 137
Python Question

Discrete legend in seaborn heatmap plot

I am using the data present here to construct this heat map using seaborn and pandas.

The input csv file is here: https://www.dropbox.com/s/5jc1vr6u8j7058v/LUH2_trans_matrix.csv?dl=0

Code:

import pandas
import seaborn.apionly as sns

# Read in csv file
df_trans = pandas.read_csv('LUH2_trans_matrix.csv')

sns.set(font_scale=0.8)
cmap = sns.cubehelix_palette(start=2.8, rot=.1, light=0.9, as_cmap=True)
cmap.set_under('gray') # 0 values in activity matrix are shown in gray (inactive transitions)
df_trans = df_trans.set_index(['Unnamed: 0'])
ax = sns.heatmap(df_trans, cmap=cmap, linewidths=.5, linecolor='lightgray')

# X - Y axis labels
ax.set_ylabel('FROM')
ax.set_xlabel('TO')

# Rotate tick labels
locs, labels = plt.xticks()
plt.setp(labels, rotation=0)
locs, labels = plt.yticks()
plt.setp(labels, rotation=0)

# revert matplotlib params
sns.reset_orig()


As you can see from csv file, it contains 3 discrete values: 0, -1 and 1. I want a discrete legend instead of the colorbar. Labeling 0 as A, -1 as B and 1 as C. How can I do that?

Answer

Well, there's definitely more than one way to accomplish this. In this case, with only three colors needed, I would pick the colors myself by creating a LinearSegmentedColormap instead of generating them with cubehelix_palette. If there were enough colors to warrant using cubehelix_palette, I would define the segments on colormap using the boundaries option of the cbar_kws parameter. Either way, the ticks can be manually specified using set_ticks and set_ticklabels.

The following code sample demonstrates the manual creation of LinearSegmentedColormap, and includes comments on how to specify boundaries if using a cubehelix_palette instead.

import matplotlib.pyplot as plt
import pandas
import seaborn.apionly as sns
from matplotlib.colors import LinearSegmentedColormap

sns.set(font_scale=0.8)
dataFrame = pandas.read_csv('LUH2_trans_matrix.csv').set_index(['Unnamed: 0'])

# For only three colors, it's easier to choose them yourself.
# If you still really want to generate a colormap with cubehelix_palette instead,
# add a cbar_kws={"boundaries": linspace(-1, 1, 4)} to the heatmap invocation
# to have it generate a discrete colorbar instead of a continous one.
myColors = ((0.8, 0.0, 0.0, 1.0), (0.0, 0.8, 0.0, 1.0), (0.0, 0.0, 0.8, 1.0))
cmap = LinearSegmentedColormap.from_list('Custom', myColors, len(myColors))

ax = sns.heatmap(dataFrame, cmap=cmap, linewidths=.5, linecolor='lightgray')

# Manually specify colorbar labelling after it's been generated
colorbar = ax.collections[0].colorbar
colorbar.set_ticks([-0.667, 0, 0.667])
colorbar.set_ticklabels(['B', 'A', 'C'])

# X - Y axis labels
ax.set_ylabel('FROM')
ax.set_xlabel('TO')

# Only y-axis labels need their rotation set, x-axis labels already have a rotation of 0
_, labels = plt.yticks()
plt.setp(labels, rotation=0)

plt.show()

Heatmap using red, green, and blue as colors with a discrete colorbar