Closed maho3 closed 11 months ago
It turns out that all torch.distributions
require Tensor inputs. This is cause for a more general solution, wherein we catch all the arguments of the prior distribution and convert them to Tensors first. Closing this PR and trying again...
Created a new
ili.utils.distributions
module which contains backend-agnostic distribution functions. In later iterations, it can be used to merge priors fromtorch.distributions
andpydelfi
.Also updated any example which needs to call BoxUniform, to now call the wrapper in
ili.utils