Implementation:Eventual Inc Daft Window Specification

Knowledge Sources	Daft Daft Docs
Domains	Data_Engineering, Data_Analysis
Last Updated	2026-02-08 00:00 GMT

Overview

Concrete tool for defining window specifications for window function computations provided by the Daft library.

Description

The Window class describes how to partition data and in what order to apply window functions. It provides a fluent builder API with three main methods: partition_by() to define grouping columns, order_by() to specify row ordering within partitions (with configurable ascending/descending and null placement), and rows_between() to define row-based frame boundaries. The class also provides class-level constants for frame boundaries: unbounded_preceding, unbounded_following, and current_row. Window specifications are applied to expressions using the .over(window) method. A range_between() method is also available for value-based frame boundaries.

Usage

Import and use this class when you need to compute window functions such as running totals, moving averages, rankings, or lead/lag operations over partitioned and ordered data.

Code Reference

Source Location

• Repository: Daft
• File: daft/window.py
• Lines: L12-259

Signature

class Window:
    # Class-level constants
    unbounded_preceding = _PyWindowBoundary.unbounded_preceding()
    unbounded_following = _PyWindowBoundary.unbounded_following()
    current_row = _PyWindowBoundary.offset(0)

    def partition_by(self, *cols: ManyColumnsInputType) -> Window
    def order_by(self, *cols: ManyColumnsInputType, desc: bool | list[bool] = False, nulls_first: bool | list[bool] | None = None) -> Window
    def rows_between(self, start: int | _PyWindowBoundary, end: int | _PyWindowBoundary, min_periods: int = 1) -> Window
    def range_between(self, start: Any, end: Any, min_periods: int = 1) -> Window

Import

from daft import Window

# or
import daft
window = daft.Window()

I/O Contract

Inputs

Name	Type	Required	Description
partition_by cols	ManyColumnsInputType	No	Columns or expressions to partition by; column names as strings or Expression objects
order_by cols	ManyColumnsInputType	No	Columns or expressions to order by within each partition
desc	list[bool]	No	Sort descending (True) or ascending (False); defaults to False
nulls_first	list[bool] | None	No	Position nulls at beginning (True) or end (False); defaults to matching desc
rows_between start	_PyWindowBoundary	No	Start boundary for row-based frame (negative = preceding, 0 = current, positive = following)
rows_between end	_PyWindowBoundary	No	End boundary for row-based frame
min_periods	int	No	Minimum rows required in frame to compute a result; defaults to 1

Outputs

Name	Type	Description
return	Window	A window specification for use with .over(window) on expressions

Usage Examples

Basic Usage

from daft import Window, col

# Basic window aggregation partitioned by category
window_spec = Window().partition_by("category")
df = df.select(
    col("value").sum().over(window_spec).alias("category_total"),
    col("value").mean().over(window_spec).alias("category_avg"),
)

# Percent of category total
window_spec = Window().partition_by("category")
df = df.select(
    (col("value") / col("value").sum().over(window_spec)).alias("pct_of_category")
)

# Sliding window: current row and 2 preceding rows
window_spec = Window().partition_by("cat").order_by("val").rows_between(-2, Window.current_row)

# Cumulative window: from beginning of partition to current row
cum_window = (
    Window()
    .partition_by("cat")
    .order_by("val")
    .rows_between(Window.unbounded_preceding, Window.current_row)
)

Related Pages

Implements Principle

• Principle:Eventual_Inc_Daft_Window_Functions

Requires Environment

• Environment:Eventual_Inc_Daft_Python_PyArrow_Core