Agrupar por varias claves

Pass a list of keys to groupby and pandas forms a group for every unique combination of those columns.

Two cities and three years can yield up to six groups, one per existing combination found in your data.

Aggregation works exactly the same; the math just runs inside each combination instead of a single key.

Grouping on several keys gives the result a MultiIndex, with one index level for each grouping column.

Each output row is labeled by a tuple like (city, year), so the index tells you exactly which group it is.

Use loc with a tuple to pull one specific combination straight out of the grouped result.

Prefer flat columns? reset_index turns each index level back into an ordinary column you can filter.

By default pandas skips combinations with no rows, so your result shows only combinations that actually occur.

The key order you list sets the index level order, which changes how the result sorts and reads.

Call unstack to lift one key into columns, turning a long grouped result into a readable grid.

Reach for multi-key grouping whenever a question mentions two dimensions, like sales by city and by month.

You group by ["city", "year"]. What labels each result row?

You can now slice data along several keys at once, read the MultiIndex, and reshape it for clear cross-tabs. 📊