SFI.inference.sparse.base module

SFI.inference.sparse.base — Strategy protocol

Every sparse-selection algorithm subclasses SparsityStrategy and implements run(), which receives a SparseScorer and returns a SparsityResult.

class SFI.inference.sparse.base.SparsityStrategy

Bases: ABC

Abstract base class for sparsity search strategies.

Subclasses must implement run().

name: str = 'base'

Short identifier used in SparsityResult.method.

abstractmethod run(scorer, *, max_k, **kwargs)

Execute the search and return a SparsityResult.

Parameters:

• scorer (SparseScorer) – Provides info_and_coeffs / vmap_info for evaluating candidate supports.

• max_k (int) – Maximum model size to explore.

Return type:

SparsityResult