Grouping and aggregation Updated November 2026

transform

df.groupby("k")["v"].transform("sum") computes the sum per group and returns it on every original row, so the result aligns with df and can be assigned as a new column. Use it for shares of a group total, group-wise normalization or filling by group.

Syntax

grouped.transform(func)   |   Series.transform(func)

Part	What it is
`func`	A function name string ("sum", "mean", "max") or a callable that returns a same-length Series or a scalar

Examples

import pandas as pd

df = pd.DataFrame({"dept": ["Sales", "IT", "Sales", "IT"],
                   "salary": [50000, 62000, 58000, 70000]})
df["dept_total"] = df.groupby("dept")["salary"].transform("sum")
df

Result: dept salary dept_total 0 Sales 50000 108000 1 IT 62000 132000 2 Sales 58000 108000 3 IT 70000 132000

df["share"] = (df["salary"] / df["dept_total"]).round(3)
df["share"].tolist()

Result: [0.463, 0.47, 0.537, 0.53]

df["vs_dept_mean"] = df["salary"] - df.groupby("dept")["salary"].transform("mean")
df["vs_dept_mean"].tolist()

Result: [-4000.0, -4000.0, 4000.0, 4000.0]

Common mistakes

Using agg and trying to assign to a column gives a length mismatch: agg returns one row per group; transform returns one per original row.
A custom function returns a scalar for some groups and a Series for others: return consistently, either a scalar or a same-length Series.
transform("size") or "count" to number rows: use groupby(...).cumcount() for a running counter.

Practice transform

These drills check your answer on live data, free, no account needed.

Total by groupReturn the total `revenue` for each region as a Series indexed by region.
Running totalReturn the running total of `signups`, day by day.

Questions

transform vs apply in groupby?

transform must return output the same length as each group (or a scalar to broadcast) and is faster. apply can return any shape, including a reduced frame per group.

How do I filter groups by a group statistic?

Build the mask with transform: df[df.groupby("k")["v"].transform("sum") > 100]. groupby(...).filter(lambda g: g.v.sum() > 100) does the same.

groupbyHow groupby works in pandas: group by one or more columns, apply sum, mean, count or size, get a flat table with as_index=False, and iterate over groups.
aggHow to use agg in pandas to compute several statistics per group, different functions per column, named aggregations with flat column names, and lambdas.
applyHow to use apply in pandas on a Series or DataFrame, with axis=1 for row-wise logic, lambdas, returning several columns, and why vectorized code is faster.
fillnaHow to fill NaN values in pandas with fillna: a constant, a different value per column, the column mean or median, forward and backward fill, and by group.