How to subplot seaborn catplot (kind='count') on-top of catplot (kind='violin') with sharex=True

Question

So far I have tried the following code:

# Import to handle plotting
import seaborn as sns

# Import pyplot, figures inline, set style, plot pairplot
import matplotlib.pyplot as plt

# Make the figure space
fig = plt.figure(figsize=(2,4))
gs = fig.add_gridspec(2, 4)
ax1 = fig.add_subplot(gs[0, :])
ax2 = fig.add_subplot(gs[1, :])

# Load the example car crash dataset
tips = sns.load_dataset("tips")

# Plot the frequency counts grouped by time
sns.catplot(x='sex', hue='smoker',
                                   kind='count',
                                   col='time',
                                   data=tips,
                                   ax=ax1)

# View the data
sns.catplot(x='sex', y='total_bill', hue='smoker',
                                                   kind='violin',
                                                   col='time',
                                                   split='True', 
                                                   cut=0, 
                                                   bw=0.25, 
                                                   scale='area',
                                                   scale_hue=False,
                                                   inner='quartile',
                                                   data=tips,
                                                   ax=ax2)

plt.close(2)
plt.close(3)
plt.show()

This seems to stack the categorial plots, of each kind respectively, on top of eachother.

What I want are the resulting plots of the following code in a single figure with the countplot in row one and the violin plot in row two.

# Import to handle plotting
import seaborn as sns

# Import pyplot, figures inline, set style, plot pairplot
import matplotlib.pyplot as plt

# Load the example car crash dataset
tips = sns.load_dataset("tips")

# Plot the frequency counts grouped by time
sns.catplot(x='sex', hue='smoker',
                                   kind='count',
                                   col='time',
                                   data=tips)

# View the data
sns.catplot(x='sex', y='total_bill', hue='smoker',
                                                   kind='violin',
                                                   col='time',
                                                   split='True', 
                                                   cut=0, 
                                                   bw=0.25, 
                                                   scale='area',
                                                   scale_hue=False,
                                                   inner='quartile',
                                                   data=tips)

The actual categorical countplot that I would like to span row one of a figure that also contains a categorical violin plot (Ref. Image 3):

The actual categorical violin plot that I would like to span row two of a figure that also contains a categorical countplot (Ref. Image 2):

I tried the following code which forced the plots to be in the same figure. The downside is that the children of the figure/axes did not transfer, i.e. axis-labels, legend, and grid lines. I feel pretty close with this hack but need another push or source for inspiration. Also, I'm no longer able to close the old/unwanted figures.

# Import to handle plotting
import seaborn as sns

# Import pyplot, figures inline, set style, plot pairplot
import matplotlib.pyplot as plt

# Set some style
sns.set_style("whitegrid")

# Load the example car crash dataset
tips = sns.load_dataset("tips")

# Plot the frequency counts grouped by time
a = sns.catplot(x='sex', hue='smoker',
                                       kind='count',
                                       col='time',
                                       data=tips)

numSubs_A = len(a.col_names)

for i in range(numSubs_A):
    for p in a.facet_axis(0,i).patches:
        a.facet_axis(0,i).annotate(str(p.get_height()), (p.get_x()+0.15, p.get_height()+0.1))

# View the data
b = sns.catplot(x='sex', y='total_bill', hue='smoker',
                                                       kind='violin',
                                                       col='time',
                                                       split='True', 
                                                       cut=0, 
                                                       bw=0.25, 
                                                       scale='area',
                                                       scale_hue=False,
                                                       inner='quartile',
                                                       data=tips)

numSubs_B = len(b.col_names)

# Subplots migration
f = plt.figure()
for i in range(numSubs_A):
    f._axstack.add(f._make_key(a.facet_axis(0,i)), a.facet_axis(0,i))
for i in range(numSubs_B):
    f._axstack.add(f._make_key(b.facet_axis(0,i)), b.facet_axis(0,i))

# Subplots size adjustment
f.axes[0].set_position([0,1,1,1])
f.axes[1].set_position([1,1,1,1])
f.axes[2].set_position([0,0,1,1])
f.axes[3].set_position([1,0,1,1])

Answer 1

It is in general not possible to combine the output of several seaborn figure-level functions into a single figure. See (this question, also this issue). I once wrote a hack to externally combine such figures, but it has several drawbacks. Feel free to use it if it works for you.

But in general, consider creating the plot you desired manually. In this case it could look like this:

import seaborn as sns
import matplotlib.pyplot as plt
sns.set()

fig, axes = plt.subplots(2,2, figsize=(8,6), sharey="row", sharex="col")

tips = sns.load_dataset("tips")
order = tips["sex"].unique()
hue_order = tips["smoker"].unique()


for i, (n, grp) in enumerate(tips.groupby("time")):
    sns.countplot(x="sex", hue="smoker", data=grp, 
                  order=order, hue_order=hue_order, ax=axes[0,i])
    sns.violinplot(x='sex', y='total_bill', hue='smoker', data=grp,
                   order=order, hue_order=hue_order,
                   split='True', cut=0, bw=0.25, 
                   scale='area', scale_hue=False,  inner='quartile', 
                   ax=axes[1,i])
    axes[0,i].set_title(f"time = {n}")

axes[0,0].get_legend().remove()
axes[1,0].get_legend().remove()
axes[1,1].get_legend().remove()
plt.show()

Answer 2

seaborn.catplot does not accept an "ax" argument, hence the problem with your first code.

It appears that some hacking is needed to accomplish the x-sharing you aim for:

How to plot multiple Seaborn Jointplot in Subplot

As such, you could save the time and effort, and just manually stack the two figures from your second code.