Operations

Every operation that changes the schema of a frame has a hook. Each hook computes the result schema from the schema of the input, or reports why it cannot. The operation catalog of the specification gives the rule of each call.

Operation	What the checker computes and checks
select, select_seq	The columns of the expressions. Each name exists, and no two results share a name.
with_columns, with_columns_seq	The columns of the frame, with each result added or replaced.
drop, remove	The columns that are left. Each dropped name exists.
rename	The new names. Each old name exists, and no two results share a name.
filter	The same columns. The condition is a Boolean.
join, join_asof, join_where	The columns of both sides, with the suffix on a right column whose name is also a left column. The keys have data types that join.
group_by(...).agg(...) and the shorthands	The keys and one column for each aggregation.
rolling(...).agg(...), group_by_dynamic(...).agg(...)	The keys, the boundaries, the index and the aggregations. The index and each duration have a data type that the call accepts.
pivot, unpivot, melt	The reshaped columns.
explode, unnest	The inner data type of a List or an Array, and the fields of a Struct.
pl.concat, hstack, vstack, extend, merge_sorted	The combined columns. The frames agree as each strategy needs. A vertical concat gives the plain frame.
with_row_index	A UInt32 index column first.
fill_null, fill_nan	The data type of each column after the fill.
update	The columns of the left frame.
map_columns	The same names. Each mapped column becomes Unchecked.
corr	A Float64 column for each input column.
sum, mean, max and the other whole-frame reducers	One row, with the data type that the reducer gives for each column.
df["a"], get_column	A TypedSeries with the name and the data type of the column.
pl.select	The columns of the expressions, over no frame.
lazy(), collect()	The same schema, as a LazyFrame or as a DataFrame.
pl.read_*, pl.scan_*, pl.from_* with schema=	A typed frame with the given schema.

An operation that keeps the columns, for example sort, head, tail, unique or sample, keeps the schema.

The expressions inside the operations are typed with the typing rules of Polars. See Expressions and data types.