SplitMap

Processor that assigns each event to a dataset split (for example train, validation, or test) using a seeded random map. The weights give the relative proportion of each split, and the assignment is deterministic for a given seed so splits are reproducible.

Configuration

Selected as name: splitmap.

Option	Description	Type	Default
seed	Random seed making the split assignment reproducible.	int	required
range	Number of buckets the weights are spread over (0 to 255).	int	required
weights	Relative proportion of each split. Must have range entries, be non-negative, and sum to a positive value (normalized internally).	list[number]	required

- name: splitmap
  kwargs:
    seed: 2747
    range: 10
    weights: [ 8, 8, 8, 8, 8, 8, 8, 8, 1, 1 ]