How to iteratively plot different data as boxplots in seaborn (without them overlapping)?

Question

Is there a way to iteratively plot data using seaborn's sns.boxplot() without having the boxplots overlap? (without combining datasets into a single pd.DataFrame())

Background

Sometimes when comparing different (e.g. size/shape) datasets, a mutual comparison is often useful and can be made by binning the datasets by a different shared variable (via pd.cut() and df.groupby(), as shown below).

Previously, I have iteratively plotted these "binned" data as boxplots on the same axis by looping separate DataFrames using matplotlib's ax.boxplot() (by providing y axis location values as a position argument to to ensure boxplots don't overlap).

Example

Below is an simplified example that shows the overlapping plots in when using sns.boxplot():

import seaborn as sns
import random
import pandas as pd
import matplotlib.pyplot as plt

# Get the tips dataset and select a subset as an example
tips = sns.load_dataset("tips")
variable_to_bin_by = 'tip'
binned_variable = 'total_bill'
df = tips[[binned_variable,  variable_to_bin_by] ]

# Create a second dataframe with different values and shape
df2 = pd.concat( [ df.copy() ] *5 )    
# Use psuedo random numbers to convey that df2 is different to df
scale = [ random.uniform(0,2) for i in range(len(df2[binned_variable])) ]  
df2[ binned_variable ] = df2[binned_variable].values * scale * 5
dfs = [ df, df2 ]

# Group the data by a list of bins
bins = [0, 1, 2, 3, 4]
for n, df in enumerate( dfs ):
    gdf = df.groupby( pd.cut(df[variable_to_bin_by].values, bins ) )
    data = [ i[1][binned_variable].values  for i in gdf]
    dfs[n] = pd.DataFrame( data, index = bins[:-1])

# Create an axis for both DataFrames to be plotted on
fig, ax = plt.subplots()

# Loop the DataFrames and plot
colors = ['red', 'black']    
for n in range(2):
    ax = sns.boxplot( data=dfs[n].T, ax=ax, width=0.2, orient='h', 
                      color=colors[n] )
plt.ylabel( variable_to_bin_by )
plt.xlabel( binned_variable )
plt.show()

More detail

I realise the simplified example above could resolved by combining the DataFrames and providing the hue argument to sns.boxplot().

Updating the index of the DataFrames provide also doesn't help, as y values from the last DataFrame provided is then used.

Providing the kwargs argument (e.g. kwargs={'positions': dfs[n].T.index}) won't work as this raises a TypeError.

TypeError: boxplot() got multiple values for keyword argument 'positions'

The setting sns.boxplot()'s dodge argument to True doesn't solve this.

you need to create plt settings in the loop, after before and after `sns.boxplot` — Mohsen_Fatemi, Jan 09 '18 at 21:35
The usual strategy would indeed be to combine the data into a single dataframe. Why is this not an option here? — ImportanceOfBeingErnest, Jan 09 '18 at 21:37
@Mohsen_Fatemi, please could give the settings you recommend updating? — tsherwen, Jan 10 '18 at 14:32
@ImportanceOfBeingErnest Indeed. I have updated the example following your comment to try and highlight that the DataFrames in this example differ in shape and content. — tsherwen, Jan 10 '18 at 14:34

score 1 · Accepted Answer · answered Jan 10 '18 at 11:34

Funnily enough, the "hack" that I proposed earlier today in this answer could be applied here.

It complicates the code a bit because seaborn expects a long-form dataframe instead of a wide-form to use hue-nesting.

# Get the tips dataset and select a subset as an example
tips = sns.load_dataset("tips")
df = tips[['total_bill',   'tip'] ]

# Group the data by 
bins = [0, 1, 2, 3, 4]
gdf = df.groupby( pd.cut(df['tip'].values, bins ) )
data = [ i[1]['total_bill'].values  for i in gdf]
df = pd.DataFrame( data , index = bins[:-1]).T
dfm = df.melt() # create a long-form database
dfm.loc[:,'dummy'] = 'dummy'

# Create a second, slightly different, DataFrame
dfm2 = dfm.copy()
dfm2.value = dfm.value*2
dfs = [ dfm, dfm2 ]
colors = ['red', 'black']
hue_orders = [['dummy','other'], ['other','dummy']]

# Create an axis for both DataFrames to be plotted on
fig, ax = plt.subplots()

# Loop the DataFrames and plot
for n in range(2):
    ax = sns.boxplot( data=dfs[n], x='value', y='variable', hue='dummy', hue_order=hue_orders[n], ax=ax, width=0.2, orient='h', 
                      color=colors[n] )
ax.legend_.remove()
plt.show()

Thanks Diziet! Your example and then a [adding legend manually](https://matplotlib.org/users/legend_guide.html#creating-artists-specifically-for-adding-to-the-legend-aka-proxy-artists) gives me exactly what I was after. — tsherwen, Jan 11 '18 at 11:31

How to iteratively plot different data as boxplots in seaborn (without them overlapping)?

1 Answers1