Module laplace.laplace

Functions

def Laplace(model, likelihood, subset_of_weights='last_layer', hessian_structure='kron', *args, **kwargs)

Simplified Laplace access using strings instead of different classes.

Parameters

model : torch.nn.Module

 

likelihood : {'classification', 'regression'}

 

subset_of_weights : {'last_layer', 'subnetwork', 'all'}, default='last_layer'

subset of weights to consider for inference

hessian_structure : {'diag', 'kron', 'full', 'lowrank'}, default='kron'

structure of the Hessian approximation

Returns

laplace : ParametricLaplace

chosen subclass of ParametricLaplace instantiated with additional arguments