Saving images corrupts x and y tick labels

[mattharrison]

I'm trying to create some Seaborn plots for an upcoming book and the x/y tick values are getting corrupted when saving it as a PNG.

Using version 0.12.1.

# bug in seaborn objects matplotlib saving
import matplotlib.pyplot as plt
import pandas as pd
import seaborn.objects as so
url = 'https://raw.githubusercontent.com/vincentarelbundock/Rdatasets/master/csv/DAAG/possum.csv'

df = pd.read_csv(url, index_col=[0])
fig, ax = plt.subplots(figsize=(6,4))
(so
 .Plot(df, x='age', y='skullw', color='sex')
 .add(so.Dots())
 .on(fig)
 .plot()
)
fig.savefig('/tmp/so.png', dpi=600, bbox_inches='tight')

The result looks like this:

However, if I remove the saving logic....

(so
 .Plot(df, x='age', y='skullw', color='sex')
 .add(so.Dots())
)

The plot looks like this:

[mwaskom]

Bear in mind the documentation for Plot.on:

Matplotlib object to use. Passing matplotlib.axes.Axes will add artists without otherwise modifying the figure. Otherwise, subplots will be created within the space of the given matplotlib.figure.Figure or matplotlib.figure.SubFigure.

What you're doing is creating a figure anad axes, then passing the figure to Plot, so it creates a new set of axes overlapping the one you already created.

The solution is either (a) initialize only the figure with pyplot (with plt.figure()) or pass ax to Plot.on. Note that your choice will also the themetheme, since rc params are cached on the Axes at its time of creation.

PS if your only goal here is to save a png and you aren't doing any other post-processing, why bother with pyplot when you could use Plot.save?

[mattharrison]

I didn't know that .save existed.

My goal is to export the image and this is the process that I've always used for matplotlib derived tooling (previous Seaborn, Pandas, Yellowbrick, scikit-plot, shap, scikit-learn, etc).

I wouldn't expect saving the figure to mess anything up if there was an existing axes. I will use plt.figure instead of plt.subplots. Thanks!

[mwaskom]

Right, saving the figure isn't messing anything up. Remove the save bit and you will see that it's all due to which object you give Plot.on.

the process that I've always used for matplotlib derived tooling

Why make it more complicated than necessary though?