sammo.search

Module Contents

Classes

Optimizer
BeamSearch
RegularizedEvolution
SequentialSearch
EnumerativeSearch

Data

logger

API

sammo.search.logger = 'getLogger(...)'

class sammo.search.Optimizer(runner: sammo.base.Runner, search_space: collections.abc.Callable[[], sammo.components.Output] | None, objective: collections.abc.Callable[[sammo.data.DataTable, sammo.data.DataTable, bool], float], maximize: bool = False)

Initialization

REPORT_COLUMNS = ('objective', 'costs')

argbest(x, key='objective')

argsort(x, key='objective')

break_even(baseline_costs, weights=None)

fit(dataset: sammo.data.DataTable)

fit_transform(dataset: sammo.data.DataTable) → sammo.data.DataTable

score(dataset: sammo.data.DataTable, **kwargs) → dict

transform(dataset: sammo.data.DataTable, **kwargs) → sammo.data.DataTable

property best: dict

property best_score

property best_prompt: sammo.components.Output

property fit_costs

save(fname: str | pathlib.Path | None = None, **extra_info)

show_report()

async evaluate(candidates: list[sammo.components.Output], runner: sammo.base.Runner, objective: collections.abc.Callable[[sammo.data.DataTable, sammo.data.DataTable], sammo.base.EvaluationScore], dataset: sammo.data.DataTable, colbar: sammo.compactbars.CompactProgressBars | None = None) → list[dict]

validate(dataset: sammo.data.DataTable, k_best=5)

class sammo.search.BeamSearch(runner: sammo.base.Runner, mutator: sammo.mutators.Mutator, objective: collections.abc.Callable[[sammo.data.DataTable, sammo.data.DataTable, bool], float], maximize: bool = True, beam_width: int = 4, depth: int = 6, mutations_per_beam: int = 8, n_initial_candidates: int = 1, add_previous: bool = False, priors: beartype.typing.Literal[uniform] | dict = 'uniform', max_evals: int | None = None)

Bases: sammo.search.Optimizer

Initialization

REPORT_COLUMNS = ('iteration', 'action', 'objective', 'costs', 'parse_errors', 'prev_actions')

log(depth, items)

async afit_transform(dataset: sammo.data.DataTable) → sammo.data.DataTable

class sammo.search.RegularizedEvolution(runner: sammo.base.Runner, mutator: sammo.mutators.Mutator, objective: collections.abc.Callable[[sammo.data.DataTable, sammo.data.DataTable, bool], float], maximize: bool = True, beam_width: int = 4, depth: int = 6, mutations_per_beam: int = 8, n_initial_candidates: int = 1, add_previous: bool = False, priors: beartype.typing.Literal[uniform] | dict = 'uniform', max_evals: int | None = None)

Bases: sammo.search.BeamSearch

Initialization

class sammo.search.SequentialSearch(runner: sammo.base.Runner, mutator: sammo.mutators.Mutator, objective: collections.abc.Callable[[sammo.data.DataTable, sammo.data.DataTable, bool], float], maximize: bool = True, depth: int = 25)

Bases: sammo.search.BeamSearch

Initialization

class sammo.search.EnumerativeSearch(runner: sammo.base.Runner, search_space: collections.abc.Callable[[], sammo.components.Output], objective: collections.abc.Callable[[sammo.data.DataTable, sammo.data.DataTable], sammo.base.EvaluationScore], maximize: bool = True, algorithm: beartype.typing.Literal[grid, random] = 'grid', max_candidates: int | None = None, n_evals_parallel: int = 2, mutate_from: sammo.components.Output | None = None, random_state: int = 42)

Bases: sammo.search.Optimizer

Initialization

REPORT_COLUMNS = ('iteration', 'action', 'objective', 'costs', 'parse_errors')

async afit_transform(dataset: sammo.data.DataTable) → sammo.data.DataTable