Seaborn FacetGrid for stratified countplots?

Question

Note: A full reproduction notebook for this question can be found on GitHub.

I have a data set with a distribution of HTTP response codes that I would like to group by class. Sample data can be generated like so:

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

%matplotlib inline

mock_http_response_data = pd.DataFrame({
    'response_code':np.repeat([200, 201, 202, 204, 302, 304, 400, 404, 500, 502], 250 ),
})

I have added a column to the data, based on 'response status', called 'response class'. The response class contains the label corresponding to the class of the particular response:

• 2xx: success
• 3xx: warning
• 4xx: client error
• 4xx: server error

The function to determine the response class is:

def determine_response_class(row):    
    response_code = row['response_code']

    if response_code >= 200 and response_code < 300:
        return 'success'
    elif response_code >= 300 and response_code < 400:
        return 'warning'
    elif response_code >= 400 and response_code < 500:
        return 'client_error'
    elif response_code >= 500 and response_code < 600:
        return 'server_error'
    else:
        return 'unknown'

And the column is added like so:

# Add 'Response class' column to API Logs, where response class is determined by HTTP status code
mock_http_response_data['response_class'] = mock_http_response_data.apply(determine_response_class, axis='columns')

The 'response status' (HTTP status code) data plots correctly with a basic countplot:

sns.countplot(
    x='_source.response_status',
    data=results_df,
    color='teal',
    saturation=0.7)

When I try to create a FacetGrid of countplots, the charts seem to work, but the labels are incorrect:

grid = sns.FacetGrid(mock_http_response_data, col='response_class')

grid.map(sns.countplot, 'response_code')

I would expect that the FacetGrid of countplots would have the following x-axis labels:

• 200
• 201
• 202
• 302
• 304
• 400
• 404
• 500
• 502

How can I create a FacetGrid of countplots so that the labels are correct and the faceted data are sorted from high to low (e.g. the 'success' class column)?

Answer 1

The problem of wrong labels appears because by default, the x axes of the subplots are shared, hence all plots will have the same x-axis as the last plot.

You can use the sharex=False argument in order to prevent sharing of the axes:

grid = sns.FacetGrid(df, col='class', sharex=False)

import pandas as pd
import numpy as np; np.random.seed(42)
import matplotlib.pyplot as plt
import seaborn as sns

codes = [200, 201, 202, 204, 302, 304, 400, 404, 500, 502]
p = np.random.rand(len(codes))
p = p/p.sum()
df = pd.DataFrame({ 'code': np.random.choice(codes, size=300, p=p) })

def determine_response_class(row):    
response_code = row['code']

if response_code >= 200 and response_code < 300:
    return 'success'
elif response_code >= 300 and response_code < 400:
    return 'warning'
elif response_code >= 400 and response_code < 500:
    return 'client_error'
elif response_code >= 500 and response_code < 600:
    return 'server_error'
else:
    return 'unknown'

df['class'] = df.apply(determine_response_class, axis='columns')

grid = sns.FacetGrid(df, col='class', sharex=False)

grid.map(sns.countplot, 'code')

plt.show()

The problem of sorting is now a chicken-or-egg problem. In order to set the order of the columns you need to know the counts for each, which are determined as part of the plotting. At this point it is probably wise to stick to a clear separation between data generation, analysis and visualization. The following would show a sorted graph, without the use of FacetGrid, by first counting an sorting the values in the dataframe.

import pandas as pd
import numpy as np; np.random.seed(42)
import matplotlib.pyplot as plt
import seaborn as sns

codes = [200, 201, 202, 204, 302, 304, 400, 404, 500, 502]
p = np.random.rand(len(codes))
p = p/p.sum()
df = pd.DataFrame({ 'code': np.random.choice(codes, size=300, p=p) })

def determine_response_class(row):    
    response_code = row['code']

    if response_code >= 200 and response_code < 300:
        return 'success'
    elif response_code >= 300 and response_code < 400:
        return 'warning'
    elif response_code >= 400 and response_code < 500:
        return 'client_error'
    elif response_code >= 500 and response_code < 600:
        return 'server_error'
    else:
        return 'unknown'

df['class'] = df.apply(determine_response_class, axis='columns')

df2 = df.groupby(["code","class"]).size().reset_index(name="count") \
        .sort_values(by="count", ascending=0).reset_index(drop=True)

fig, axes = plt.subplots(ncols=4, sharey=True, figsize=(8,3))
for ax,(n, group) in zip(axes, df2.groupby("class")):
    sns.barplot(x="code",y="count", data=group, ax=ax, color="C0", order=group["code"])
    ax.set_title(n)

plt.tight_layout()
plt.show()