Matplotlib customize the legend to show squares instead of rectangles

Question

Here is my attempt to change the legend of a barplot from rectangle to square:

import matplotlib.patches as patches

rect1 = patches.Rectangle((0,0),1,1,facecolor='#FF605E')
rect2 = patches.Rectangle((0,0),1,1,facecolor='#64B2DF')
plt.legend((rect1, rect2), ('2016', '2015'))

But when I plot this, I still see rectangles instead of squares:

Any suggestions on how can I do this?

---

I tried both solutions provided by @ImportanceOfBeingErnest and @furas, here are the results:

@ImportanceOfBeingErnest's solution is the easiest to do:

plt.rcParams['legend.handlelength'] = 1
plt.rcParams['legend.handleheight'] = 1.125

Here is the result:

My final code looks like this:

plt.legend((df.columns[1], df.columns[0]), handlelength=1, handleheight=1) # the df.columns = the legend text

@furas's solution produces this, I don't know why the texts are further away from the rectangles, but I am sure the gap can be changed somehow:

Answer 1

Matplotlib provides the rcParams

legend.handlelength  : 2.    # the length of the legend lines in fraction of fontsize
legend.handleheight  : 0.7   # the height of the legend handle in fraction of fontsize

You can set those within the call to plt.legend()

plt.legend(handlelength=1, handleheight=1)

or using the rcParams at the beginning of your script

import matplotlib
matplotlib.rcParams['legend.handlelength'] = 1
matplotlib.rcParams['legend.handleheight'] = 1

Unfortunately providing equal handlelength=1, handleheight=1 will not give a perfect rectange. It seems handlelength=1, handleheight=1.125 will do the job, but this may depend on the font being used.

---

An alternative, if you want to use proxy artists may be to use the square markers from the plot/scatter methods.

bar1 = plt.plot([], marker="s", markersize=15, linestyle="", label="2015")

and supply it to the legend, legend(handles=[bar1]). Using this approach needs to have set matplotlib.rcParams['legend.numpoints'] = 1, otherwise two markers would appear in the legend.

---

Here is a full example of both methods

import matplotlib.pyplot as plt
plt.rcParams['legend.handlelength'] = 1
plt.rcParams['legend.handleheight'] = 1.125
plt.rcParams['legend.numpoints'] = 1


fig, ax = plt.subplots(ncols=2, figsize=(5,2.5))

# Method 1: Set the handlesizes already in the rcParams
ax[0].set_title("Setting handlesize")
ax[0].bar([0,2], [6,3], width=0.7, color="#a30e73", label="2015", align="center")
ax[0].bar([1,3], [3,2], width=0.7, color="#0943a8", label="2016", align="center" )
ax[0].legend()

# Method 2: use proxy markers. (Needs legend.numpoints to be 1)
ax[1].set_title("Proxy markers")
ax[1].bar([0,2], [6,3], width=0.7, color="#a30e73", align="center" )
ax[1].bar([1,3], [3,2], width=0.7, color="#0943a8", align="center" )
b1, =ax[1].plot([], marker="s", markersize=15, linestyle="", color="#a30e73",  label="2015")
b2, =ax[1].plot([], marker="s", markersize=15, linestyle="", color="#0943a8",  label="2016")
ax[1].legend(handles=[b1, b2])

[a.set_xticks([0,1,2,3]) for a in ax]
plt.show()

producing

Answer 2

It seems they change it long time ago - and now some elements can't be used directly in legend.

Now it needs handler: Implementing a custom legend handler

import matplotlib.pyplot as plt
import matplotlib.patches as patches
from matplotlib.legend_handler import HandlerPatch

# --- handlers ---

class HandlerRect(HandlerPatch):

    def create_artists(self, legend, orig_handle,
                       xdescent, ydescent, width, height,
                       fontsize, trans):

        x = width//2
        y = 0
        w = h = 10

        # create
        p = patches.Rectangle(xy=(x, y), width=w, height=h)

        # update with data from oryginal object
        self.update_prop(p, orig_handle, legend)

        # move xy to legend
        p.set_transform(trans)

        return [p]


class HandlerCircle(HandlerPatch):

    def create_artists(self, legend, orig_handle,
                       xdescent, ydescent, width, height,
                       fontsize, trans):

        r = 5
        x = r + width//2
        y = height//2

        # create 
        p = patches.Circle(xy=(x, y), radius=r)

        # update with data from oryginal object
        self.update_prop(p, orig_handle, legend)

        # move xy to legend
        p.set_transform(trans)

        return [p]

# --- main ---

rect = patches.Rectangle((0,0), 1, 1, facecolor='#FF605E')
circ = patches.Circle((0,0), 1, facecolor='#64B2DF')

plt.legend((rect, circ), ('2016', '2015'),
            handler_map={
               patches.Rectangle: HandlerRect(),
               patches.Circle: HandlerCircle(),
            })

plt.show()

Legend reserves place for rectangle and this method doesn't change it so there is so many empty space.