I have a dataframe which contains two columns, id and value. Every few rows have the same id, but all have different values. I want to make a boxen plot which shows distribution of values for all ids along with the distribution of accumulated values from first id to the current one. This is a simple dataframe:

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

df = pd.DataFrame(columns=['id', 'value'])

for id in range(5):
    df = pd.concat([df, pd.DataFrame({'id': [id]*10, 'value':np.random.rand(10)})]).reset_index(drop=True)

This:

plt.figure(figsize=(10, 5))

sns.boxenplot(x='id', y='value', data=df, width=0.4)

plt.show()

plots only the distribution of values for each id. I want to have the distribution of the accumulated ids close to each id value. For example:

for id=0, both are the same, at 0 and 0.5 I want to have the same boxes

for id=1, at x=1 I want to have a box for values with id==1 and at 1.5 a box for values with id==0 or id==1

for id=2, at x=2 I want to have a box for values with id==2 and at 2.5 a box for values with id==0 or id==1 or id==2

and so on.

I could not find a way to put box plots of dataframes with different sizes in one plot.

CodePudding user response：

Using pandas GroupBy() to compute the sum

df = df.groupby('id').sum()

plt.figure(figsize=(10, 5))

sns.boxenplot(x=df.index, y=df.value, data=df, width=0.4)

plt.show()

CodePudding user response：

There are a couple ways to do this-

1. Reshape your data to match up with what seaborn expects as input (copying data, and inserting it into a single DataFrame

2. Tap into some lower matplotlib functionality and create the chart with more manual control.

Seaborn - Data Manipulation

I'm essentially storing repeated data with the same id under a different label 'cumulative' and 'non-cumulative'. Once I do that, I can use seaborns interface to generate the expected plot.

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

plt.rc('axes.spines', left=False, top=False, right=False)

rng = np.random.default_rng(0)

df = pd.DataFrame({
    'id': np.arange(5).repeat(10),
    'value': rng.random(5 * 10)
})

data = {}
cumul_groups = []
for id_, grp in df.groupby('id'):
    cumul_groups.append(id_)
    
    data[(id_, 'cumulative')] = df.loc[df['id'].isin(cumul_groups), 'value']
    data[(id_, 'non-cumulative')] = grp['value']

ax = sns.boxenplot(x='id', y='value', hue='type', data=pd.concat(data, names=['id', 'type']).reset_index())
ax.legend(loc='lower left', bbox_to_anchor=(1, .5))