mbi.estimation.interior_gradient_step

mbi.estimation.interior_gradient_step(state: InteriorGradientState, loss_fn: MarginalLossFn, marginal_oracle: MarginalOracle, total: Array | float, inv_lipschitz: float) → InteriorGradientState

Performs a single interior gradient step.

Parameters:

• state – Current algorithm state.

• loss_fn – The marginal loss function.

• marginal_oracle – A marginal oracle with signature (potentials, total) -> marginals.

• total – The known or estimated total number of records.

• inv_lipschitz – Reciprocal of the Lipschitz constant (sigma / lipschitz).

Returns:

Updated InteriorGradientState.