API reference

Divergence

class rdr.Divergence(value)

Supported $phi$-divergence objectives for RDR estimation.

Trainer

class rdr.RDRTrainer(model: Module, divergence: Divergence | str = Divergence.HELLINGER, optimizer: Optimizer | None = None, lr: float = 0.0005, weight_decay: float = 0.01, device: str | device | None = None, mixture_ratio: float = 0.5, eps: float = 1e-08)

Train a neural network to estimate the relative density ratio $r(x)=2p(x)/(p(x)+q(x))$.

Evaluation helpers

rdr.evaluate_ratio(model: Module, x_real: Tensor, x_gen: Tensor, device: str | device | None = None) → dict

rdr.estimate_relative_density_ratio(model: Module, x_real: Tensor, x_gen: Tensor, device: str | device | None = None) → dict