nnpd.core.settings

Explicit one-factor-at-a-time sweeps. Ordinary lists are literal values.

Classes

Choice	An atomic sweep knob: Choice([baseline, alternative, ...]).
Variant

Functions

canonical(→ str)	Stable JSON; reject NaN, infinity, callables, and ambiguous custom objects.
digest(→ str)
at(→ Any)
select(→ dict)
expand_sweep(→ list[Variant])	Baseline once, then one changed knob per run; never a Cartesian product.
single(→ dict)	Settings that must be one value everywhere: reject any Choice.
load_settings(→ dict)	Load a trusted Python settings file; this executes Python, not a sandbox.
default_profile(→ str)

Module Contents

nnpd.core.settings.canonical(value: Any) → str

Stable JSON; reject NaN, infinity, callables, and ambiguous custom objects.

nnpd.core.settings.digest(value: Any) → str

class nnpd.core.settings.Choice

An atomic sweep knob: Choice([baseline, alternative, …]).

Values may themselves be lists or dictionaries (e.g. complete architectures). Nesting Choice inside an option is deliberately unsupported.

values: list[Any] | tuple[Any, ...]

class nnpd.core.settings.Variant

settings: dict

changed: str | None

value: Any = None

property id: str

nnpd.core.settings.at(settings: dict, path: str) → Any

nnpd.core.settings.select(settings: dict, *paths: str) → dict

nnpd.core.settings.expand_sweep(settings: dict) → list[Variant]

Baseline once, then one changed knob per run; never a Cartesian product.

Dictionary insertion order determines presentation order, not run identity. Duplicate resolved settings are removed, including repeated baselines.

nnpd.core.settings.single(settings: dict, name: str) → dict

Settings that must be one value everywhere: reject any Choice.

nnpd.core.settings.load_settings(path: str | pathlib.Path, profile: str | None = None) → dict

Load a trusted Python settings file; this executes Python, not a sandbox.

nnpd.core.settings.default_profile(path: str | pathlib.Path) → str