discopt.solvers.mip_nlp_candidates

Fixed-integer NLP candidate management for MIP-NLP decomposition.

Classes

FixedNLPCandidate	One candidate integer assignment for a fixed-integer NLP solve.
FixedNLPCandidateManager	Queue and schedule fixed-NLP candidates.

Module Contents

class discopt.solvers.mip_nlp_candidates.FixedNLPCandidate

One candidate integer assignment for a fixed-integer NLP solve.

point: numpy.ndarray

source: str

objective: float | None = None

iteration: int | None = None

nlp_source: str = 'active'

provider: str | None = None

sequence: int = 0

integer_assignment: tuple[float, ...] = ()

trace_dict() → dict[str, object]

class discopt.solvers.mip_nlp_candidates.FixedNLPCandidateManager(*, n_vars: int, int_indices: list[int], lb, ub, strategy: str = 'always', iteration_frequency: int = 1, time_frequency: float = 0.0, candidate_limit: int | None = None, deduplicate_used_assignments: bool = False)

Queue and schedule fixed-NLP candidates.

The manager keeps ordering deterministic, deduplicates integer assignments, and exposes a small scheduler that mirrors SHOT’s always/iteration/time/pool activation modes without coupling to a particular MIP or NLP backend.

n_vars

int_indices

lb

ub

strategy = ''

iteration_frequency

time_frequency

candidate_limit = None

deduplicate_used_assignments = False

added_source_counts: collections.Counter[str]

skipped_duplicate_count = 0

property pending_count: int

assignment_key(point) → tuple[float, ...]

add(point, *, source: str, objective: float | None = None, iteration: int | None = None, nlp_source: str = 'active', provider: str | None = None) → bool

add_external_candidates(candidates, *, iteration: int | None = None, provider: str | None = None, nlp_source: str = 'active') → int

add_master_result(master_result: Any, *, iteration: int, solution_pool: bool, limit: int, nlp_source: str = 'active') → int

should_call(*, iteration: int, elapsed: float | None = None, has_solution_pool_candidate: bool = False) → bool

take_ready(*, iteration: int, elapsed: float | None = None, has_solution_pool_candidate: bool = False) → list[FixedNLPCandidate]

record_call_result(candidate: FixedNLPCandidate, *, iteration: int, elapsed: float | None, success: bool) → None

state_signature() → tuple[object, ...]

Everything that decides what a later take_ready returns.

Used by OA’s fixed-point exit (#1710) to tell whether the scheduler would behave identically on a repeated iteration. The call clock and iteration are excluded deliberately: the exit applies only under the always strategy, whose should_call reads neither.

scheduler_trace() → dict[str, object]