Shrinkage

shrinkage.WorkingSetPolicy

Shrinkage for Python · python · 0.2.0

WorkingSetPolicy(
    kind: typing.Literal["auto", "disabled", "ever_active", "adaptive"],
    initial_size: int = 10,
    prune: bool = True
)

Heuristic feature selection, independent of certified zero screening.

Every policy retains current nonzeros. Excluded zeros may reenter after full checks. Only the original full-space mapping certifies convergence.

kind

Source: packages/python/src/shrinkage/solver.py

kind: typing.Literal["auto", "disabled", "ever_active", "adaptive"]

initial_size

Source: packages/python/src/shrinkage/solver.py

initial_size: int = 10

prune

Source: packages/python/src/shrinkage/solver.py

prune: bool = True

auto

Source: packages/python/src/shrinkage/solver.py

auto() -> shrinkage.WorkingSetPolicy

Use adaptive sets for single fits, or ever-active sets for paths.

disabled

Source: packages/python/src/shrinkage/solver.py

disabled() -> shrinkage.WorkingSetPolicy

Update every feature allowed by independently configured screening.

ever_active

Source: packages/python/src/shrinkage/solver.py

ever_active() -> shrinkage.WorkingSetPolicy

Retain features activated at accepted iterates.

adaptive

Source: packages/python/src/shrinkage/solver.py

adaptive(
    initial_size: int = 10,
    *,
    prune: bool = True
) -> shrinkage.WorkingSetPolicy

Rank violations, growing the set and optionally pruning it.