Windows: framing primer (rowsBetween / rangeBetween)

This recipe is a short mental model for window functions and framing in pydantable.

Key idea

• Unframed windows behave like “compute over partition/order context” (Polars-style).
• Framed windows (rowsBetween, rangeBetween) define an explicit frame around each row.

See WINDOW_SQL_SEMANTICS for the detailed rules, including multi-key rangeBetween axis behavior.

Recipe: rank within partition

from pydantable import DataFrameModel
from pydantable.window_spec import Window
from pydantable.expressions import row_number


class Row(DataFrameModel):
    group: str
    v: int


df = Row({"group": ["a", "a", "b"], "v": [2, 1, 5]})
w = Window.partitionBy("group").orderBy("v")
out = df.with_columns(rn=row_number().over(w)).to_dict()
assert "rn" in out

Pitfalls

• Prefer the named window APIs (Window.partitionBy(...).orderBy(...)) over generic Expr.over(...) (see INTERFACE_CONTRACT).