Seaborn Heatmap correlation won't fit annotation digits

Question

I'm trying to plot a triangular correlation matrix using seaborn heatmap but the cells won't fit the annotation digits.

Any idea how I make them fit nicely inside their respective heatmap cell?

I already tried changing the figsize and that did not help. Also tried using square=False.

I'm using seaborn==0.11.2 and matplotlib==3.4.3

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

# Generate a dummy df
df = pd.DataFrame(np.random.rand(44,44))

label_lens = [16, 16, 16, 16, 16, 16, 16, 16, 20, 11,
              9, 10, 10, 16, 16, 16, 16, 12, 45, 10, 10,
             10, 10, 10, 10, 10, 10, 12, 12, 50, 50, 50,
             50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50]

col_labels = []
for label_len in label_lens:
    col_labels.append('X'*label_len)

df.columns = col_labels

# Build correlation matrix df
correlation_matrix = df.corr()

# Get Diagonal Mask. Square matrix is not relevant.
mask = np.triu(np.ones_like(correlation_matrix, dtype=bool))

# Set up the matplotlib figure
f, ax = plt.subplots(figsize=(30, 15))

# Draw the heatmap with the mask and correct aspect ratio
sns_plot = sns.heatmap(correlation_matrix,
                       mask=mask,
                       annot=True,
                       fmt='.2f',
                       square=True)
f.set_tight_layout(True)
f.savefig("my_corr_matrix.pdf")

I replaced my labels here with placeholders of the same sizes as the actual labels.

Increase the figure size `f, ax = plt.subplots(figsize=(30, 20))` and remove `square=True` works with the sample data, but your tick labels are longer so you'll probably need larger than 30 — Trenton McKinney, Nov 10 '21 at 14:40
I tried up to figsize=(90, 60). No luck. The figure gets massive but the annotations still don't fit. — Matheus Torquato, Nov 10 '21 at 15:05
[Plot Image](https://i.stack.imgur.com/8GI3h.png) I can reproduce your issue. — Trenton McKinney, Nov 10 '21 at 15:28
@TrentonMcKinney, please, see the updated code. Now you can reproduce with the dummy labels. — Matheus Torquato, Nov 10 '21 at 15:43
I said remove `square=True` which you didn't do. `f, ax = plt.subplots(figsize=(30, 15))` works fine then; needed a little more space with the long ticklabels — Trenton McKinney, Nov 10 '21 at 15:45
[plot sample](https://i.stack.imgur.com/rCivW.png) looks like it does, and I'm using the same package versions. — Trenton McKinney, Nov 10 '21 at 15:56
@TrentonMcKinney Using the exact same code above I'm getting [this](https://postimg.cc/zyYXD5H8) in a macOS — Matheus Torquato, Nov 10 '21 at 16:07
If you're using a Jupyter Notebook, restart the kernel and clear the outputs. Otherwise I don't know, since I can't reproduce the issue. — Trenton McKinney, Nov 10 '21 at 16:18

score 2 · Accepted Answer · answered Nov 11 '21 at 12:28

2

As pointed in the comments, using square=False with figsize=(30, 15) fixed the problem.

answered Nov 11 '21 at 12:28

Matheus Torquato

1,293
18
25

mrbTT · Answer 2 · 2021-11-12T18:09:39.813

The issue is that your labels are taking too much space of your jupyter max width. You probably have 2 options, the first one is editing jupyter width according to this answer:

https://stackoverflow.com/a/34058270/2970272

The second one would be lowering how much you're using the plot space, by trimming your labels, lowering the font, removing the colorbar and such. From the code bellow check all the "## HERE" comments to easily find what I've changed.

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

# Generate a dummy df
df = pd.DataFrame(np.random.rand(44,44))

label_lens = [16, 16, 16, 16, 16, 16, 16, 16, 20, 11,
              9, 10, 10, 16, 16, 16, 16, 12, 45, 10, 10,
             10, 10, 10, 10, 10, 10, 12, 12, 50, 50, 50,
             50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50]
#label_lens= [5]*len(label_lens)

col_labels = []
for label_len in label_lens:
    col_labels.append(f"{'X'*label_len}"[:10]) ## HERE -- max char limit

df.columns = col_labels

# Build correlation matrix df
correlation_matrix = df.corr()

# Get Diagonal Mask. Square matrix is not relevant.
mask = np.triu(np.ones_like(correlation_matrix, dtype=bool))

# Set up the matplotlib figure
f, ax = plt.subplots(figsize=(30, 20)) ## here increased figsize

## HERE - max tick font size
ax.tick_params(axis='both', labelsize=8)

# Draw the heatmap with the mask and correct aspect ratio
sns_plot = sns.heatmap(correlation_matrix,
                       mask=mask,
                       annot=True,
                       fmt='.2f',
                       square=True,
                       cbar=False) ## HERE removed colorbar
f.set_tight_layout(True)
## HERE - rotating ticks to give more space
plt.setp(ax.yaxis.get_majorticklabels(), rotation=-45)
plt.setp(ax.xaxis.get_majorticklabels(), rotation=-45)
plt.show()
#f.savefig("my_corr_matrix.pdf")

Tried that. No luck. The labels went all over the place, including below the cbar. — Matheus Torquato, Nov 10 '21 at 13:29
@MatheusTorquato, I edited my answer for a different approach than the square=False that would not cramp the data. — mrbTT, Nov 12 '21 at 18:12

Seaborn Heatmap correlation won't fit annotation digits

2 Answers2