从使用`pd.Categorical`过滤的数据帧绘制Python seaborn

Question

I'm trying to plot some data from a subset of my dataframe, but it is plotting empty ticks for data that should have been filtered out. I know the issue is that I used pd.Categorical() , but I need to. How do I plot only the filtered data (ie just a1 and a2 ) and no extra ticks? Example:

import numpy as np
import pandas as pd
data = {'A':['a2', 'a2', 'a2', 'a1', 'a1', 'a1', 'a3', 'a3', 'a3'],
        'B': np.random.normal(0, 1, 9)}

df = pd.DataFrame(data)

df :

df
Out[1]: 
    A         B
0  a2 -1.076173
1  a2 -2.574480
2  a2  0.863081
3  a1  1.411732
4  a1 -0.937692
5  a1  0.929105
6  a3 -1.071276
7  a3  0.901292
8  a3  0.740417


# Sort A using pd.categorical
df['A'] = pd.Categorical(df['A'], ['a1', 'a2', 'a3'])
df = df.sort_values(by='A')

plotdf = df.loc[df['A']!='a3']

plotdf should now be a subset of df ... which it is:

plotdf
Out[2]: 
    A         B
3  a1  1.411732
4  a1 -0.937692
5  a1  0.929105
0  a2 -1.076173
1  a2 -2.574480
2  a2  0.863081

But when we plot it has retained the filtered-out tick position:

import matplotlib.pyplot as plt
import seaborn as sns

fig, ax = plt.subplots()
sns.barplot(x='A', y='B', data=plotdf)
plt.show()

Do I need to re-specify the categories before I plot? Seems a bit odd...

Answer 1

This seems to be an effect of the categorical type that maintains all of its possible values even if they are not always present (see print(plotdf['A'].dtype) ).

for example, running plotdf.groupby('A').size() returns

A
a1    3
a2    3
a3    0

with category a3 showing up despite not being present in the dataframe.

In any case, if you don't want to see this empty space on the plot, you can tell seaborn which categories to plot using the order= parameter:

sns.barplot(x='A', y='B', data=plotdf, order=['a1', 'a2'])

Note that if you want to be generic, you could do order=plotdf['A'].unique()