Merging Strategy

Merge results from multiple operator branches.

See Also

• Operators Overview - All operator types
• Branching Strategy - Split processing paths
• Control Flow - DAG merge nodes
• Ensemble Strategy - Ensemble aggregation

---

datarax.operators.strategies.merging

Merging utilities for parallel and ensemble strategies.

logger module-attribute

logger = getLogger(__name__)

merge_output_sequence

merge_output_sequence(outputs: list[PyTree], merge_strategy: str | None, merge_axis: int = 0, merge_fn: Callable | None = None) -> PyTree

Merge parallel outputs based on strategy.

Parameters:

Name	Type	Description	Default
outputs	list[PyTree]	List of outputs from parallel operators	required
merge_strategy	str | None	How to merge ("concat", "stack", "sum", "mean", "dict")	required
merge_axis	int	Axis for stack/concat operations	0
merge_fn	Callable | None	Custom merge function (overrides merge_strategy)	None

Returns:

Type	Description
PyTree	Merged output

merge_outputs_conditional

merge_outputs_conditional(outputs: list[PyTree], conditions: list[Array], merge_strategy: str | None, merge_axis: int = 0, merge_fn: Callable | None = None) -> PyTree

Merge outputs with conditional masking (vmap-compatible).

Parameters:

Name	Type	Description	Default
outputs	list[PyTree]	List of ALL operator outputs (identity for False conditions)	required
conditions	list[Array]	Boolean arrays indicating which operators executed	required
merge_strategy	str | None	How to merge	required
merge_axis	int	Axis used for merge	0
merge_fn	Callable | None	Custom merge function	None

Returns:

Type	Description
PyTree	Merged output with only True-condition outputs