neuralset.events.transforms.utils.DeterministicSplitter¶

class neuralset.events.transforms.utils.DeterministicSplitter(ratios: dict[str, float], seed: float = 0.0)[source][source]¶

Hash-based splitter that assigns a deterministic train/val/test split.

Hashes each sample’s unique ID to a float in [0, 1) and maps it to a split name according to cumulative ratios. The assignment is stable across runs and independent of dataset order.

Parameters:

ratios (dict[str, float]) – Mapping from split name to proportion (must sum to 1).
seed (float) – Added to the hash to produce different splits.