When we perform a groupby across multiple columns, we often want to change how our data is stored. For instance, recall the example where we are running a chain of stores and have data about the number of sales at different locations on different days:

We suspected that there might be different sales on different days of the week at different stores, so we performed a groupby across two different columns (Location and Day of Week). This gave us results that looked like this: In order to test our hypothesis, it would be more useful if the table was formatted like this:

Reorganizing a table in this way is called pivoting. The new table is called a pivot table.

In Pandas, the command for pivot is:

df.pivot(columns='ColumnToPivot',
         index='ColumnToBeRows',
         values='ColumnToBeValues')

For our specific example, we would write the command like this:

# First use the groupby statement:
unpivoted = df.groupby(['Location', 'Day of Week'])['Total Sales'].mean().reset_index()
# Now pivot the table
pivoted = unpivoted.pivot(
    columns='Day of Week',
    index='Location',
    values='Total Sales')

Just like with groupby, the output of a pivot command is a new DataFrame, but the indexing tends to be “weird”, so we usually follow up with .reset_index().

For more on the pivot function, see the pandas documentation.