如何在 Matplotlib 中生成嵌套图例

Question

I have to plot in Matplotlib a quantity which is the sum of various contributions.

I would like to highlight this fact in the legend of the plot by listing the various contribution as sub-elements of the main legend entry.

A sketch of the result I would like to obtain can be found in the picture below. Note that I do not need to necessarily achieve exactly the legend that is depicted, but just something similar.

Answer 1

You can try creating two separate legends to your figure. Sure, it's a trick rather than a direct feature of the legend object, as there seems to be no implementation of what you need in matplotlib. But playing with the numbers in bbox and the fontsize you can customize it pretty nicely.

import matplotlib.pyplot as plt 
import numpy as np

x = np.arange(0.0, 1, 0.01)
x1 = np.sin(2*np.pi*x)
x2 = np.sin(2*np.pi*x+1)
x3 = np.sin(2*np.pi*x+2)


fig, ax = plt.subplots()
f1, = ax.plot(x1, 'r', lw=4)
f2, = ax.plot(x2, 'k', lw=2)
f3, = ax.plot(x3, 'b', lw=2)

legend1 = plt.legend([f1], ["Main legend"], fontsize=12, loc=3, bbox_to_anchor=(0,0.1,0,0), frameon=False)
legend2 = plt.legend((f2, f3), ('sublegend 1', 'sublegend 2'), fontsize=9,
                 loc=3, bbox_to_anchor=(0.05,0,0,0), frameon=False)
plt.gca().add_artist(legend1)
plt.show()

EDIT:

Well, if we insert 2 legends, why not just inserting a completely new figure as inset inside the bigger figure, dedicated for a legend, inside which you can draw and write whatever you like? Admittedly it's a hard work, you have to design each and every line inside including the precise location coordinates. But that's the way I could think of for doing what you wanted:

import matplotlib.pyplot as plt
import numpy as np

x = np.arange(0.0, 1, 0.01)
x1 = np.sin(2*np.pi*x)
x2 = np.sin(2*np.pi*x+1)
x3 = np.sin(2*np.pi*x+2)


fig, ax = plt.subplots()
f1, = ax.plot(x1, 'r', lw=4)
f2, = ax.plot(x2, 'k', lw=2)
f3, = ax.plot(x3, 'b', lw=2)

## set all lines for inner figure
yline1 = np.array([-0.15, -0.15]) 
line1 = np.array([2, 10])
yline2 = np.array([3, 0])
line2 = np.array([4, 4])
yline3 = np.array([1.5, 1.5])
line3 = np.array([4, 6])
yline4 = np.array([1.5, 1.5])
line4 = np.array([7, 10])
yline5 = np.array([3, 3])
line5 = np.array([4, 6])
yline6 = np.array([3, 3])
line6 = np.array([7, 10])

## inset figure
axin1 = ax.inset_axes([2.5, -1, 30, 0.5], transform=ax.transData)  # 

## plot all lines
axin1.plot(line1, yline1, linewidth=4, c='r')
axin1.plot(line2, yline2, 'k', lw=1)
axin1.plot(line3, yline3, 'k', lw=1)
axin1.plot(line4, yline4, 'b', lw=3)
axin1.plot(line5, yline5, 'k', lw=1)
axin1.plot(line6, yline6, 'k', lw=3)

## text
axin1.text(12, 0, 'MAIN', fontsize=12)
axin1.text(12, 1.7, 'Subtext 1', fontsize=10)
axin1.text(12, 3.2, 'Subtext 2', fontsize=10)

## adjust
axin1.set_ylim([4, -1])
axin1.set_xlim([0, 27])
axin1.set_xticklabels('')
axin1.set_yticklabels('')

Answer 2

I looked for a custom example in the legend and could not see any indication of lowering the level. You can just line up the objects in the legend. I've created a hierarchy of the presented images in the form of colors and markers. The official reference has been customized. This has the effect of eliminating the need to annotate only the legend in a special way.

import matplotlib.pyplot as plt
from matplotlib.legend_handler import HandlerLineCollection, HandlerTuple

fig, ax1 = plt.subplots(1, 1, constrained_layout=True)

params = {'legend.fontsize': 16,
          'legend.handlelength': 3}

plt.rcParams.update(params)

x = np.linspace(0, np.pi, 25)
xx = np.linspace(0, 2*np.pi, 25)
xxx = np.linspace(0, 3*np.pi, 25)

p1, = ax1.plot(x, np.sin(x), lw=5, c='r')
p2, = ax1.plot(x, np.sin(xx), 'm-d', c='g')
p3, = ax1.plot(x, np.sin(xxx), 'm-s', c='b')

# Assign two of the handles to the same legend entry by putting them in a tuple
# and using a generic handler map (which would be used for any additional
# tuples of handles like (p1, p2)).
l = ax1.legend([p1, (p1, p2), (p1, p3)], ['Legend entry', 'Contribution 1', 'Contribution 2'], scatterpoints=1,
               numpoints=1, markerscale=1.3, handler_map={tuple: HandlerTuple(ndivide=None, pad=1.0)})

plt.show()