Data Analysis Data Wrangling Tutorial

groupby() Method: Split Data into Groups, Apply a Function to Groups, Combine the Results

A group by is a process that tyipcally involves splitting the data into groups based on some criteria, applying a function to each group independently, and then combining the outputted results.

Import Modules

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

Get Tips Dataset

Let's get the tips dataset from the seaborn library and assign it to the DataFrame df_tips.

df_tips = sns.load_dataset('tips')

Each row represents a unique meal at a restaurant for a party of people; the dataset contains the following fields:

column name column description
total_bill financial amount of meal in U.S. dollars
tip financial amount of the meal's tip in U.S. dollars
sex gender of server
smoker boolean to represent if server smokes or not
day day of week
time meal name (Lunch or Dinner)
size count of people eating meal

Preview the first 5 rows of df_tips.

total_bill tip sex smoker day time size
0 16.99 1.01 Female No Sun Dinner 2
1 10.34 1.66 Male No Sun Dinner 3
2 21.01 3.50 Male No Sun Dinner 3
3 23.68 3.31 Male No Sun Dinner 2
4 24.59 3.61 Female No Sun Dinner 4

Implement Group bys with Tips Dataset

Group by of a Single Column and Apply a Single Aggregate Method on a Column

The simplest example of a groupby() operation is to compute the size of groups in a single column. By size, the calculation is a count of unique occurences of values in a single column. Here is the official documentation for this operation.

This is the same operation as utilizing the value_counts() method in pandas.

Below, for the df_tips DataFrame, I call the groupby() method, pass in the sex column, and then chain the size() method.

Male      157
Female     87
dtype: int64

To interpret the output above, 157 meals were served by males and 87 meals were served by females.

A note, if there are any NaN or NaT values in the grouped column that would appear in the index, those are automatically excluded in your output (reference here).

In pandas, we can also group by one columm and then perform an aggregate method on a different column.

For example, in our dataset, I want to group by the sex column and then across the total_bill column, find the mean bill size. To do this in pandas, given our df_tips DataFrame, apply the groupby() method and pass in the sex column (that'll be our index), and then reference our ['total_bill'] column (that'll be our returned column) and chain the mean() method.

Meals served by males had a mean bill size of 20.74 while meals served by females had a mean bill size of 18.06.

Male      20.744076
Female    18.056897
Name: total_bill, dtype: float64

We can verify the output above with a query. We get the same result that meals served by males had a mean bill size of 20.74


Another interesting tidbit with the groupby() method is the ability to group by a single column, and call an aggregate method that will apply to all other numeric columns in the DataFrame.

For example, if I group by the sex column and call the mean() method, the mean is calculated for the three other numeric columns in df_tips which are total_bill, tip, and size.

total_bill tip size
Male 20.744076 3.089618 2.630573
Female 18.056897 2.833448 2.459770

Aggregate Methods

Other aggregate methods you could perform with a groupby() method in pandas are:

method description
sum() summation
mean() average
count() count of all values
size() count of non-null values
max() maximum value
min() minimum value
std() standard deviation
median() median

To illustrate the difference between the size() and count() methods, I included this simple example below.

The DataFrame below of df_rides includes Dan and Jamie's ride data.

data = {'person': ['Dan', 'Dan', 'Jamie', 'Jamie'], 
        'ride_duration_minutes': [4, np.NaN, 8, 10]}
df_rides = pd.DataFrame(data)
person ride_duration_minutes
0 Dan 4.0
1 Dan NaN
2 Jamie 8.0
3 Jamie 10.0

For one of Dan's rides, the ride_duration_minutes value is null. However, if we apply the size method, we'll still see a count of 2 rides for Dan. We are 100% sure he took 2 rides but there's only a small issue in our dataset in which the the exact duration of one ride wasn't recorded.

Dan      2
Jamie    2
Name: ride_duration_minutes, dtype: int64

Upon applying the count() method, we only see a count of 1 for Dan because that's the number of non-null values in the ride_duration_minutes field that belongs to him.

Dan      1
Jamie    2
Name: ride_duration_minutes, dtype: int64

Group by of a Single Column and Apply the describe() Method on a Single Column

With grouping of a single column, you can also apply the describe() method to a numerical column. Below, I group by the sex column, reference the total_bill column and apply the describe() method on its values. The describe method outputs many descriptive statistics. Learn more about the describe() method on the official documentation page.

count mean std min 25% 50% 75% max
Male 157.0 20.744076 9.246469 7.25 14.00 18.35 24.71 50.81
Female 87.0 18.056897 8.009209 3.07 12.75 16.40 21.52 44.30

Group by of a Single Column and Apply a Lambda Expression on a Single Column

Most examples in this tutorial involve using simple aggregate methods like calculating the mean, sum or a count. However, with group bys, we have flexibility to apply custom lambda functions.

You can learn more about lambda expressions from the Python 3 documentation and about using instance methods in group bys from the official pandas documentation.

Below, I group by the sex column and apply a lambda expression to the total_bill column. The expression is to find the range of total_bill values. The range is the maximum value subtracted by the minimum value. I also rename the single column returned on output so it's understandable.

df_tips.groupby(by='sex').agg({'total_bill': lambda bill: bill.max() - bill.min()}).rename(columns={'total_bill': "range_total_bill"})
Male 43.56
Female 41.23

In this dataset, males had a bigger range of total_bill values.

Group by of Multiple Columns and Apply a Single Aggregate Method on a Column

We can group by multiple columns too. For example, I want to know the count of meals served by people's gender for each day of the week. So, call the groupby() method and set the by argument to a list of the columns we want to group by.

df_tips.groupby(by=['sex', 'day']).size()
sex     day 
Male    Thur    30
        Fri     10
        Sat     59
        Sun     58
Female  Thur    32
        Fri      9
        Sat     28
        Sun     18
dtype: int64

We can also group by multiple columns and apply an aggregate method on a different column. Below I group by people's gender and day of the week and find the total sum of those groups' bills.

df_tips.groupby(by=['sex', 'day'])['total_bill'].sum()
sex     day 
Male    Thur     561.44
        Fri      198.57
        Sat     1227.35
        Sun     1269.46
Female  Thur     534.89
        Fri      127.31
        Sat      551.05
        Sun      357.70
Name: total_bill, dtype: float64

Group by of a Single Column and Apply Multiple Aggregate Methods on a Column

The agg() method allows us to specify multiple functions to apply to each column. Below, I group by the sex column and then we'll apply multiple aggregate methods to the total_bill column. Inside the agg() method, I pass a dictionary and specify total_bill as the key and a list of aggregate methods as the value.

You can pass various types of syntax inside the argument for the agg() method. I chose a dictionary because that syntax will be helpful when we want to apply aggregate methods to multiple columns later on in this tutorial.

df_tips.groupby(by='sex').agg({'total_bill': ['count', 'mean', 'sum']})
count mean sum
Male 157 20.744076 3256.82
Female 87 18.056897 1570.95

You can learn more about the agg() method on the official pandas documentation page.

The code below performs the same group by operation as above, and additionally I rename columns to have clearer names.

df_tips.groupby(by='sex').agg({'total_bill': ['count', 'mean', 'sum']}).rename(columns={'count': 'count_meals_served', 'mean': 'average_bill_of_meal', 'sum': 'total_bills_of_meals'})
count_meals_served average_bill_of_meal total_bills_of_meals
Male 157 20.744076 3256.82
Female 87 18.056897 1570.95

We can modify the format of the output above through chaining the unstack() and reset_index() methods after our group by operation. This format may be ideal for additional analysis later on.

df_tips.groupby(by='sex').agg({'total_bill': ['count', 'mean', 'sum']}).unstack().reset_index().rename(columns={'level_0': 'aggregated_column', 'level_1': 'aggregate_metric', 'sex': 'grouped_column', 0: 'aggregate_calculation'}).round(2)
aggregated_column aggregate_metric grouped_column aggregate_calculation
0 total_bill count Male 157.00
1 total_bill count Female 87.00
2 total_bill mean Male 20.74
3 total_bill mean Female 18.06
4 total_bill sum Male 3256.82
5 total_bill sum Female 1570.95

Group by of a Single Column and Apply Multiple Aggregate Methods on Multiple Columns

Below, I use the agg() method to apply two different aggregate methods to two different columns. I group by the sex column and for the total_bill column, apply the max method, and for the tip column, apply the min method.

df_tips.groupby(by='sex').agg({'total_bill': 'max', 'tip': 'min'}).rename(columns={'total_bill': 'max_total_bill', 'tip': 'min_tip_amount'})
max_total_bill min_tip_amount
Male 50.81 1.0
Female 44.30 1.0

Group by of Multiple Columns and Apply a Groupwise Calculation on Multiple Columns

In restaurants, common math by guests is to calculate the tip for the waiter/waittress. My mom thinks 20% tip is customary. So, if the bill was 10, you should tip 2 and pay 12 in total. I'm curious what the tip percentages are based on the gender of servers, meal and day of the week. We can perform that calculation with a groupby() and the pipe() method.

The pipe() method allows us to call functions in a chain. So as the groupby() method is called, at the same time, another function is being called to perform data manipulations. You can learn more about pipe() from the official documentation.

To perform this calculation, we need to group by sex, time and day, then call our pipe() method and calculate the tip divided by total_bill multiplied by 100.

df_tips.groupby(by=['sex', 'time', 'day']).pipe(lambda group: group.tip.sum()/group.total_bill.sum()*100)
sex     time    day 
Male    Lunch   Thur    15.925121
                Fri     16.686183
        Dinner  Fri     12.912840
                Sat     14.824622
                Sun     14.713343
Female  Lunch   Thur    15.388192
                Fri     19.691535
        Dinner  Thur    15.974441
                Fri     19.636618
                Sat     14.236458
                Sun     16.944367
dtype: float64

The highest tip percentage has been for females for dinner on Sunday.