Components¶
Base classes and registration functions of the components a plugin implements. Recipes: Extending.
Methods¶
marathonts.bench.methods.register_method ¶
register_method(name, pool=None, assembler=None, long=None, description='', overwrite=False, joiner=None)
marathonts.bench.methods.register_stitch_suite ¶
register_stitch_suite(prefix: str, pool: str, overwrite: bool = False, extra_assemblers=(), extra_joiners=()) -> list[str]
Register the whole stitching suite (config.STITCH_SUITE: the zseries stitcher, the random path, the
alternative stitchers of BASELINES.md A–B) on one pool generator: Config.smooth_paths / joiners). Every method of the suite shares the pool, so they differ only in the
choice, order and seams of the windows. Idempotent; a name already registered with another
pool / assembler raises unless overwrite. Returns the method names.
marathonts.bench.methods.register_long_method ¶
register_long_method(name: str, long: str | None = None) -> str
A long generator as a method (by default under its own name); idempotent.
Generators¶
marathonts.bench.generators.base.GenContext
dataclass
¶
K
property
¶
K: int
Windows the assembler returns: the series' K, or more for a method with seam processing.
marathonts.bench.generators.base.PoolGenerator ¶
load_state ¶
load_state(arrays: dict[str, ndarray], ctx: 'GenContext')
Restore a fitted generator from state() arrays; returns self.
marathonts.bench.generators.base.LongGenerator ¶
Assembly¶
marathonts.bench.assemblers.Assembler ¶
marathonts.bench.joiners.Joiner ¶
Tasks and models¶
marathonts.bench.tasks.Task
dataclass
¶
marathonts.bench.models.BaseModel ¶
load_state ¶
load_state(arrays: dict[str, ndarray]) -> 'BaseModel'
Inverse of state(): the model predicts as when it was saved.
Fidelity metrics¶
marathonts.bench.fidelity.register_fidelity_group ¶
register_fidelity_group(name: str, spec: dict, overwrite: bool = False, primary_only: bool = False)
Decorator: register fn(ctx) -> dict under name; spec = {metric: orientation} (+1 / −1; a tuple
whose first element is the orientation is accepted too).
Unless primary_only, the @<part> copies of every metric are registered too.
marathonts.bench.fidelity.FidelityContext
dataclass
¶
Hypotheses¶
marathonts.hypotheses.catalog.register_hypothesis ¶
register_hypothesis(hid: str, title: str = '', overwrite: bool = False)
Decorator: fn(tables, cfg) -> {"global": DataFrame, "per_dataset": DataFrame, "extra": dict}.
tables holds the benchmark tables (timing, fidelity, utility, cross_seed, datasets, pools).