My solution to https://twitter.com/zh1zh4n6/status/1759823358340825149

A legend is a collection of OffsetBox instances. You may collect multiple legends and rearange them to your need.

import matplotlib.pyplot as plt

Example of multiple legends

We will start with a modified version of the multiple legend example from multiple-legends-on-the-same-axes

Note that multiple calls of ax.legend will replace an exising legend. To work around this, we manually add the legend artist to the axes using ax.add_artist.

#! echo: false

def plot(ax):
    line1, = ax.plot([1, 2, 3], label="Line 1", linestyle='--')
    line2, = ax.plot([3, 2, 1], label="Line 2", linewidth=4)

    sct1 = ax.scatter([0, 1, 2], [1.5, 4, 2.], label="Sct 1", marker="s")
    sct2 = ax.scatter([0, 1, 2], [4, 3, 0.5], label="Sct 2", marker="^")

    return [line1, line2], [sct1, sct2]

title_props = dict(alignment="left",
                   title_fontproperties=dict(weight="bold"))

fig, ax = plt.subplots(num=1, clear=True)

[line1, line2], [sct1, sct2] = plot(ax)

# Create a legend for the first set.
first_legend = ax.legend(handles=[line1, sct1], loc='upper right',
                         title="Data 1", **title_props)

# Add the legend manually to the Axes.
ax.add_artist(first_legend)

# Create another legend for the second.
ax.legend(handles=[line2, sct2], loc='lower right',
          title="Data 2", **title_props)

<matplotlib.legend.Legend at 0x7fe8cdde6170>

Assembling multiple legends

In the modified version below, We will create multiple legend boxes, which will be assemebed into a single offset box. Note that we don’t want individual legends be displayed.

fig, ax = plt.subplots(num=2, clear=True)

[line1, line2], [sct1, sct2] = plot(ax)

# Legend is a collection of offset_box instances, wrapped around with
# AnchoredOffsetbox. We will create legends and collect the offset_box
# instances, but without showing them on the screen for now.
offsetboxes = []
for title, handles in [("Data 1", [line1, sct1]),
                       ("Data 2", [line2, sct2])]:
    leg = ax.legend(handles=handles,
                        title=title, **title_props)
    offsetboxes.append(leg._legend_box)

ax.legend_.remove() # We don't want the legend to be displayed.

We now have a list of offset-boxes. One can vertically pack boxes into a single box and place it on the corner of the axes, similar to a legend.

# We will add the collected offsetbox. They first need to be packed in a box.
# We use `VPacker` which will pack its childrent vertically.
from matplotlib.offsetbox import VPacker, AnchoredOffsetbox
vp = VPacker(children=offsetboxes, sep=15)

# Then this will be added to the axes using `AnchoredOffsetbox`, which can be
# placed just like legend.
ob = AnchoredOffsetbox(loc="upper right", bbox_to_anchor=ax.bbox, child=vp)
ax.add_artist(ob)

Of course it is possible to place the box outside the axes. There are several ways to do this. I will use axes_grid1 toolkit, which I am most familiar with.

We will start with a sample example.

fig, ax = plt.subplots(num=3, clear=True)

[line1, line2], [sct1, sct2] = plot(ax)

# Legend is a collection of offset_box instances, wrapped around with
# AnchoredOffsetbox. We will create legends and collect the offset_box
# instances, but without showing them on the screen for now.
offsetboxes = []
for title, handles in [("Data 1", [line1, sct1]),
                       ("Data 2", [line2, sct2])]:
    leg = ax.legend(handles=handles,
                        title=title, **title_props)
    offsetboxes.append(leg._legend_box)

ax.legend_.remove() # We don't want the legend to be displayed.

Placing the box outside of the main axes

We will create a new axes on the right side of the main axes. In addtion, we want the width of this axes set to the width of the offset box.


from mpl_toolkits.axes_grid1 import make_axes_locatable
from mpl_toolkits.axes_grid1.axes_size import MaxWidth

vp = VPacker(children=offsetboxes, sep=15)

divider = make_axes_locatable(ax)
artist_list = [] # we start with an empty artist list.
ax_right = divider.append_axes("right", size=MaxWidth(artist_list), pad=0.1)

ob = AnchoredOffsetbox(loc="upper left", bbox_to_anchor=ax_right.bbox, child=vp,
                       pad=0, borderpad=0)

ax_right.add_artist(ob)

# We add `ob` to the artist_list, so that the width of `ax_right` is adjusted.
artist_list.append(ob)

ax_right.set_axis_off()