Seeding

igl.seed_everything(seed)

Seed python, numpy, and torch global RNGs in one call.

Parameters:

Name	Type	Description	Default
seed	int	The seed applied to all three generators.	required

igl.fit_seeds(factory, x, y=None, *, seeds)

Fit one estimator per seed and aggregate the headline read-outs.

Parameters:

Name	Type	Description	Default
factory	Callable[[int], fit_seeds[EstimatorT]]	Builds a fresh estimator for a given seed (typically lambda seed: IGLRegressor(..., random_state=seed)).	required
x	object	Training inputs, forwarded to fit.	required
y	object | None	Optional training targets, forwarded to fit.	None
seeds	Sequence[int]	The seed list — e.g. a project's canonical seed protocol.	required

Returns:

Type	Description
dict[str, object]	A dict with "seeds", "estimators" (the fitted instances),
dict[str, object]	and "aggregate": for each numeric read-out found on the fitted
dict[str, object]	estimators (effective_dimension_, the final validation metric),
dict[str, object]	its mean, std, median, and IQR across seeds.