frequency and percentage uneven groups sns barplot

Question

I am trying to show relative percentage by group as well as total frequency in an sns barplot. The two groups I am comparing are very different in size, which is why I show the percentage by group in the function below.

Here is syntax for a sample dataframe I created that has similar relative group sizes to my data ('groups') among the target categorical variable ('item'). 'rand' is just a variable I use to make the df.

# import pandas and seaborn
import pandas as pd
import seaborn as sns
import numpy as np

# create dataframe
foobar = pd.DataFrame(np.random.randn(100, 3), columns=('groups', 'item', 'rand'))

# get relative groupsizes
for row, val in enumerate(foobar.rand) :
    if  val > -1.2 :
        foobar.loc[row, 'groups'] = 'A'
    else: 
        foobar.loc[row, 'groups'] = 'B'

    # assign categories that I am comparing graphically
    if row < 20:
        foobar.loc[row, 'item'] = 'Z'
    elif row < 40:
        foobar.loc[row, 'item'] = 'Y'
    elif row < 60:
        foobar.loc[row, 'item'] = 'X'
    elif row < 80:
        foobar.loc[row, 'item'] = 'W'
    else:
        foobar.loc[row, 'item'] = 'V'

Here is the function I wrote that compares relative frequencies by group. It has some default variables, but I've reassigned them for this question.

def percent_categorical(item, df=IA, grouper='Active Status') :
    # plot categorical responses to an item ('column name')
    # by percent by group ('diff column name w categorical data')
    # select a data frame (default is IA)
    # 'Active Status' is default grouper

    # create df of item grouped by status
    grouped = (df.groupby(grouper)[item]
    # convert to percentage by group rather than total count
                .value_counts(normalize=True)
                # rename column 
                .rename('percentage')
                # multiple by 100 for easier interpretation
                .mul(100)
                # change order from value to name
                .reset_index()
            .sort_values(item))

    # create plot
    PercPlot = sns.barplot(x=item,
                         y='percentage',
                         hue=grouper,
                         data=grouped,
                         palette='RdBu'
                         ).set_xticklabels(
                                 labels = grouped[item
                                      ].value_counts().index.tolist(), rotation=90)
    #show plot
    return PercPlot

The function and resulting graph follow:

percent_categorical('item', df=foobar, grouper='groups')

This is good, because it allows me show the relative percentage by group. However, I also want to display the absolute numbers for each group, preferably in the legend. In this case, I would want it to show that there are 89 total members of group A and 11 total members of group B.

Thank you in advance for any help.

Answer 1

I solved this by splitting out the groupby operation: one to get your percentages and one to count the number of objects.

I adjusted your percent_catergorical function as follows:

def percent_categorical(item, df=IA, grouper='Active Status') :
    # plot categorical responses to an item ('column name')
    # by percent by group ('diff column name w categorical data')
    # select a data frame (default is IA)
    # 'Active Status' is default grouper

    # create groupby of item grouped by status
    groupbase = df.groupby(grouper)[item]
    # count the number of occurences
    groupcount = groupbase.count()       
    # convert to percentage by group rather than total count           
    groupper = (groupbase.value_counts(normalize=True)
                # rename column 
                .rename('percentage')
                # multiple by 100 for easier interpretation
                .mul(100)
                # change order from value to name
                .reset_index()
                .sort_values(item))

    # create plot
    fig, ax = plt.subplots()
    brplt = sns.barplot(x=item,
                         y='percentage',
                         hue=groupper,
                         data=groupper,
                         palette='RdBu',
                         ax=ax).set_xticklabels(
                                 labels = grouper[item
                                      ].value_counts().index.tolist(), rotation=90)
    # get the handles and the labels of the legend
    # these are the bars and the corresponding text in the legend
    thehandles, thelabels = ax.get_legend_handles_labels()
    # for each label, add the total number of occurences
    # you can get this from groupcount as the labels in the figure have
    # the same name as in the values in column of your df
    for counter, label in enumerate(thelabels):
        # the new label looks like this (dummy name and value)
        # 'XYZ (42)'
        thelabels[counter] = label + ' ({})'.format(groupcount[label])
    # add the new legend to the figure
    ax.legend(thehandles, thelabels)
    #show plot
    return fig, ax, brplt

To get your figure:

fig, ax, brplt = percent_categorical('item', df=foobar, grouper='groups')

The resulting graph looks like this:

You can change the look of this legend how you want, I just added parentheses as an example.