GroupBy with Single and Multiple Keys

The simplest GroupBy call uses a single column as the grouping key. You call df.groupby('column_name') and chain an aggregation. Pandas creates one group for each unique value in that column and applies the aggregation to every numeric column (or the selected column). This is the most common form of GroupBy in day-to-day analysis.

After calling groupby() you can select one column with bracket notation to get a SeriesGroupBy, or select multiple columns with a list to get a DataFrameGroupBy. If you skip selection entirely, Pandas aggregates all numeric columns, which is convenient but can produce unexpected results when you have irrelevant numeric columns.

Pandas supports a full set of built-in aggregation methods on GroupBy objects: sum(), mean(), median(), min(), max(), count(), std(), var(), first(), and last(). Each returns a scalar per group, giving you a clean summary table indexed by the grouping key.

Pass a list of column names to groupby() to group by more than one dimension at once. Each unique combination of values in those columns becomes its own group. The result has a MultiIndex with one level per grouping column, allowing you to drill into cross-dimensional summaries in a single step.

When you group by multiple keys the result index is a MultiIndex. You can access a specific outer level with .loc['value'] and navigate inner levels with tuples. Calling reset_index() flattens the MultiIndex into regular columns, which is often more convenient for further operations or export.

By default, the grouping columns become the index of the result. Passing as_index=False keeps them as regular columns instead, giving you a flat DataFrame that is easier to pass to further Pandas operations or to export. This is equivalent to calling reset_index() on the result.

count() returns the number of non-null values in each group. If you want the total number of rows per group regardless of nulls, use size() instead. This distinction matters when your data has missing values, because count() will undercount groups that have NaN entries.

By default, GroupBy sorts the result by the group key. You can disable sorting with sort=False to preserve the original order of first appearance, which is slightly faster on large datasets. Once you have the result as a DataFrame, you can sort it further with sort_values() on any column.

GroupBy results are regular Series or DataFrames, so you can chain further Pandas methods on them immediately. A common pattern is to aggregate, then reset_index(), then rename columns, then sort_values() — all in a single readable method chain without storing intermediate variables.

You frequently need more than one statistic per group. Pass a list of function names to .agg() to compute several at once. The result is a DataFrame with a column for each function. This is more efficient than calling each aggregation separately because Pandas processes all of them in a single pass over the data.

When a grouping column has the Categorical dtype, Pandas includes all categories in the result by default, even those with no rows. This can be useful to ensure your summary table always shows every category, but it creates rows with NaN or 0 for empty groups. Control this with the observed parameter (set to True to skip empty categories).

Test your understanding of GroupBy with single and multiple keys from this lesson.

In this lesson you learned: how to group by a single column and apply common aggregations, how to group by multiple columns to produce cross-dimensional summaries, and the difference between count() and size() when nulls are present. Next up we explore the powerful agg() method for computing multiple different statistics in a single call.