Completely custom legend in Matplotlib, Python

Question

I am using Matplotlib to basically draw a 'picture', not for plotting data.

In the 'picture' I use plt.annotate to label certain parts of the picture.

I now want to make a completely custom legend to indicate the meaning of the symbols.

Is there a way to define custom handles and labels, where the handles must be alphanumeric letters instead of the normal markers like '*' or 'o'.

Is this possible or must I construct the legend manually using plt.annotation?

Answer 1

There are a lot of ways to do it, but it's probably easiest to use a proxy artist in this case. You can use arbitrary text as a marker, so it's fairly easy to have fake Line2D's show labels instead of lines.

As an example (the bulk of this is the relatively "fancy" call to annotate):

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

def main():
    labels = ['A', 'B', 'C']
    positions = [(2, 5), (1, 1), (4, 8)]
    descriptions = ['Happy Cow', 'Sad Horse', 'Drooling Dog']

    # Plot the data, similar to what you described...
    fig, ax = plt.subplots()
    ax.imshow(np.random.random((10, 10)), interpolation='none')
    for label, xy in zip(labels, positions):
        ax.annotate(label, xy, xytext=(20, 20), size=15,
                    textcoords='offset points',
                    bbox={'facecolor':'white'},
                    arrowprops={'arrowstyle':'->'})

    # Create a legend with only labels
    proxies = [create_proxy(item) for item in labels]
    ax.legend(proxies, descriptions, numpoints=1, markerscale=2)

    plt.show()

def create_proxy(label):
    line = matplotlib.lines.Line2D([0], [0], linestyle='none', mfc='black',
                mec='none', marker=r'$\mathregular{{{}}}$'.format(label))
    return line

main()