probly.method.prior_network.torch.append_activation_torch

probly.method.prior_network.torch.append_activation_torch(obj: Module) → Sequential

Append exp so the model outputs Dirichlet alpha based on [MG18].

Unlike Sensoy’s softplus + 1 parameterization, exp allows alpha values below 1, which the Prior Networks paper uses for very flat Dirichlets on out-of-distribution inputs.