nirs4all.pipeline.dagml.full_train_variants module

Expand full-training candidates without manufacturing cross-validation evidence.

nirs4all.pipeline.dagml.full_train_variants.expand_full_train_variants(pipeline: list[Any], name: str = '') → list[tuple[list[Any], str]]

Return concrete step lists and their standard PipelineConfigs names.

Reuse the public generator grammar, expansion limit and naming contract. Deserialization precedes caller-side splitter detection: canonical Studio definitions may contain splitters as class paths, including inside choices. Each candidate must execute as its own real DAG run, retaining its fitted artifacts. This helper neither fits nor selects a candidate and produces no scores or synthetic folds.