nirs4all.pipeline.config.refit_params module

Refit parameter resolution utilities.

Provides resolve_refit_params() which merges refit_params on top of train_params for the refit execution phase. Parameters not specified in refit_params inherit from train_params.

nirs4all.pipeline.config.refit_params.resolve_refit_params(model_config: dict[str, Any]) → dict[str, Any]

Merge refit_params on top of train_params.

Produces the effective training parameters for the refit phase. refit_params values override train_params on conflicts; unspecified parameters inherit from train_params.

Special refit_params keys (stored but not acted on until Phase 4):

• warm_start (bool): Whether to warm-start from CV weights.

• warm_start_fold (str): Which fold to warm-start from ("best", "last", or "fold_N").

Parameters:

model_config – Model step configuration dict, typically containing "train_params" and/or "refit_params" keys.

Returns:

Merged parameter dict. If neither train_params nor refit_params is present, returns an empty dict.

Example

>>> config = {
...     "train_params": {"verbose": 0, "n_jobs": 4},
...     "refit_params": {"verbose": 1, "warm_start": True},
... }
>>> resolve_refit_params(config)
{'verbose': 1, 'n_jobs': 4, 'warm_start': True}