Reshapes

Apply

Perform a custom deterministic rows-to-rows reshape.

Apply is the escape hatch for a deterministic whole-Space reshape that does not fit a named operation.

Worked example

Apply(
    name="latest-behavior-per-user",
    input_spaces=("behaviors",),
    output_space="latest-behaviors",
    fn=PythonFunction(
        name="latest_row_per_user",
        source="""def latest_row_per_user(rows):
    latest = None
    for row in rows:
        if latest is None or row.get("time", "") > latest.get("time", ""):
            latest = row
    return [] if latest is None else [latest]
""",
    ),
    group_by="user_id",
    emits=("user_id", "history"),
    output_fields=("user_id", "history"),
)

Parameter	Default	Meaning
`input_spaces`	required	The Space or Spaces read by the reshape.
`name`	required	A stable node name shown in the pipeline and session.
`output_space`	required	The Space receiving returned rows.
`fn`	required	A callable receiving one flat list containing the ordinary mappings gathered from every input Space, and returning an iterable of ordinary mappings.
`emits`	none	Descriptive names of fields produced by the callable.
`output_fields`	none	Complete output schema when the callable builds new rows instead of retaining all fields from its input rows.
`group_by`	none	Optional column or composite key. Relifold invokes the callable once per exact key group and records only that group's rows as support for its outputs.
`max_output_rows`	none	Optional positive global hard ceiling enforced across all returned rows, not once per group. Use it after deterministic whole-hand ranking to bound downstream work and its forecast cost.

Output

Exactly the ordinary row mappings returned by the callable.

Design guidance

Prefer a named reshape when one expresses the operation. Apply is appropriate for explicit aggregations and projections, not semantic decisions. Set group_by for per-entity reductions; reserve whole-hand Apply for genuinely global operations.