How to change order and grouping of entries in a matplotlib legend?-CodePudding

This is what my legend looks like right now:

I want it to look like this:

My code:

ax = cars.plot.barh(stacked=True)
ax.invert_yaxis()
plt.legend(loc="center", ncols=4)

I tried adding the line plt.plot(0, np.zeros([1, 3]), '.', ms=0, label='fake') and ax.plot(0, np.zeros([1, 3]), '.', ms=0, label='fake') but the result is this:

The entries labelled "fake" should be added from the back. I can then change the label from "fake" to "" to create empty entries.

CodePudding user response：

I don't think that is explicitly supported. You can work around it by adding additional, fake entries to the legend (in your case 3). Consider:

import numpy as np
import matplotlib.pyplot as plt

plt.figure()

# Actual data
for i in range(13):
    plt.plot(np.random.random(), np.random.random(), '.', label=chr(ord('A')   i))

# Fake data, for the legend
plt.plot(0, np.zeros([1, 3]), '.', ms=0, label=' ')

plt.legend(ncol=4)
plt.show()  # or plt.savefig('figname.png')

Here I've used a marker size (ms) of 0, ensuring that the plotted fake data points will not appear on either the plot nor the legend.

CodePudding user response：

You could add some dummy legend handles to fill up the empty space:

from matplotlib import pyplot as plt

labels = 'abcdefghijklm'
for i, label in enumerate(labels):
    plt.bar([i], (i   1) ** 2, color=plt.cm.turbo_r(i / len(labels)), label=label)
handles, labels = plt.gca().get_legend_handles_labels()
dummy_handle = plt.Rectangle((0, 0), 0, 0, color='none', label='')
plt.legend(handles=handles   3*[dummy_handle], ncol=4)
plt.show()