nirs4all.operators.models.pytorch.mlp module

Small width-agnostic Torch factory for the closed CPU regression profile.

nirs4all.operators.models.pytorch.mlp.structural_mlp(input_shape: tuple[int, ...], hidden_units: int = 8) → Any

Construct a fresh flat-input scalar regressor from existing Torch layers.

The coordinator supplies the actual selected source width. This factory creates no fitted state and retains no samples, folds, or optimizer state.