How to make heatmap square in Seaborn FacetGrid

Question

I'm using Seaborn's FacetGrid class to plot a set of matrices using the heatmap function, also from Seaborn. However, I cannot adjust the aspect ratio of these subplots. Here is an example:

import seaborn as sns
import numpy as np
import pandas as pd

# Generate a set of sample data
np.random.seed(0)
indices = pd.MultiIndex.from_product((range(5), range(5), range(5)), names=('label0', 'label1', 'label2'))
data = pd.DataFrame(np.random.uniform(0, 100, size=len(indices)), index=indices, columns=('value',)).reset_index()

def draw_heatmap(*args, **kwargs):
    data = kwargs.pop('data')
    d = data.pivot(index=args[1], columns=args[0], values=args[2])
    sns.heatmap(d, **kwargs)

fg = sns.FacetGrid(data, col='label0')
fg.map_dataframe(draw_heatmap, 'label1', 'label2', 'value', cbar=False)

That produces this image:

Notice how the elements of the matrix plots are rectangular, rather than square. If I were making one plot (without the FacetGrid), I could pass the option square=True to sns.heatmap to make the plot square. However, if I put that in the arguments to map_dataframe, I get this result:

fg = sns.FacetGrid(data, col='label0')
fg.map_dataframe(draw_heatmap, 'label1', 'label2', 'value', cbar=False, square=True)

It looks like the axes are too big for the data, even though the default aspect ratio for the FacetGrid is 1. (And explicitly setting aspect=1 when constructing the FacetGrid doesn't help.)

Is there any way to fix this?

Answer 1

When you set aspect argument of FacetGrid not only axis limits but the sizes and positions of axis labels and titles are also took into consideration to calculate proper axis geometry. Therefore heatmaps are rectangles.

To adjust quadratic heatmap and axis you may set corrected axis bounding box position manually with function set_bbox.

Another way is a little bit tricky. Function ax.set_aspect with the second argument box_forced would try to adjust heatmap and axis. However to completely remove remain gray borders you may hide it by setting axis background color according to figure face color like here:

import seaborn as sns
import numpy as np
import pandas as pd
import matplotlib.pylab as plt

# Generate a set of sample data
np.random.seed(0)
indices = pd.MultiIndex.from_product((range(5), range(5), range(5)),
 names=('label0', 'label1', 'label2'))
data = pd.DataFrame(np.random.uniform(0, 100, size=len(indices)),
 index=indices, columns=('value',)).reset_index()

def draw_heatmap(*args, **kwargs):
    data = kwargs.pop('data')
    d = data.pivot(index=args[1], columns=args[0], values=args[2])
    sns.heatmap(d, **kwargs)

fg = sns.FacetGrid(data, col='label0')
fg.map_dataframe(draw_heatmap, 'label1', 'label2', 'value', cbar=False, square = True)
# get figure background color
facecolor=plt.gcf().get_facecolor()
for ax in fg.axes.flat:
    # set aspect of all axis
    ax.set_aspect('equal','box-forced')
    # set background color of axis instance
    ax.set_axis_bgcolor(facecolor)
plt.show()

Result: