Setting a fixed size for points in legend

You can make a Line2D object that resembles your chosen markers, except with a different marker size of your choosing, and use that to construct the legend. This is nice because it doesn't require placing an object in your axes (potentially triggering a resize event), and it doesn't require use of any hidden attributes. The only real downside is that you have to construct the legend explicitly from lists of objects and labels, but this is a well-documented matplotlib feature so it feels pretty safe to use.

from matplotlib.lines import Line2D
import matplotlib.pyplot as plt
import numpy as np

def rand_data():
    return np.random.uniform(low=0., high=1., size=(100,))

# Generate data.
x1, y1 = [rand_data() for i in range(2)]
x2, y2 = [rand_data() for i in range(2)]

plt.figure()
plt.scatter(x1, y1, marker='o', label='first', s=20., c='b')
plt.scatter(x2, y2, marker='o', label='second', s=35., c='r')

# Create dummy Line2D objects for legend
h1 = Line2D([0], [0], marker='o', markersize=np.sqrt(20), color='b', linestyle='None')
h2 = Line2D([0], [0], marker='o', markersize=np.sqrt(20), color='r', linestyle='None')

# Set axes limits
plt.gca().set_xlim(-0.2, 1.2)
plt.gca().set_ylim(-0.2, 1.2)

# Plot legend.
plt.legend([h1, h2], ['first', 'second'], loc="lower left", markerscale=2,
           scatterpoints=1, fontsize=10)
plt.show()

I had a look into the source code of matplotlib. Bad news is that there does not seem to be any simple way of setting equal sizes of points in the legend. It is especially difficult with scatter plots (wrong: see the update below). There are essentially two alternatives:

1. Change the maplotlib code
2. Add a transform into the PathCollection objects representing the dots in the image. The transform (scaling) has to take the original size into account.

Neither of these is very much fun, though #1 seems to be easier. The scatter plots are especially challenging in this respect.

However, I have a hack which does probably what you want:

import matplotlib.pyplot as plt
import numpy as np

def rand_data():
    return np.random.uniform(low=0., high=1., size=(100,))

# Generate data.
x1, y1 = [rand_data() for i in range(2)]
x2, y2 = [rand_data() for i in range(2)]

plt.figure()
plt.plot(x1, y1, 'o', label='first', markersize=np.sqrt(20.), c='b')
plt.plot(x2, y2, 'o', label='second', markersize=np.sqrt(35.), c='r')
# Plot legend.
lgnd = plt.legend(loc="lower left", numpoints=1, fontsize=10)

#change the marker size manually for both lines
lgnd.legendHandles[0]._legmarker.set_markersize(6)
lgnd.legendHandles[1]._legmarker.set_markersize(6)
plt.show()

This gives:

Which seems to be what you wanted.

The changes:

• scatter changed into a plot, which changes the marker scaling (hence the sqrt) and makes it impossible to use changing marker size (if that was intended)
• the marker size changed manually to be 6 points for both markers in the legend

As you can see, this utilizes hidden underscore properties (_legmarker) and is bug-ugly. It may break down at any update in matplotlib.

Update

Haa, I found it. A better hack:

import matplotlib.pyplot as plt
import numpy as np

def rand_data():
    return np.random.uniform(low=0., high=1., size=(100,))

# Generate data.
x1, y1 = [rand_data() for i in range(2)]
x2, y2 = [rand_data() for i in range(2)]

plt.figure()
plt.scatter(x1, y1, marker='o', label='first', s=20., c='b')
plt.scatter(x2, y2, marker='o', label='second', s=35., c='r')
# Plot legend.
lgnd = plt.legend(loc="lower left", scatterpoints=1, fontsize=10)
lgnd.legendHandles[0]._sizes = [30]
lgnd.legendHandles[1]._sizes = [30]
plt.show()

Now the _sizes (another underscore property) does the trick. No need to touch the source, even though this is quite a hack. But now you can use everything scatter offers.

Similarly to the answer, assuming you want all the markers with the same size:

lgnd = plt.legend(loc="lower left", scatterpoints=1, fontsize=10)
for handle in lgnd.legendHandles:
    handle.set_sizes([6.0])

With MatPlotlib 2.0.0