Following the discussion in #34, this PR brings a correction to the treatment of ScorePrior. In particular, there is no ScorePrior.transform anymore because these transformations are already applied before the prior gets to see the parameters. Autodiff then takes care of the Jacobians, so we simply don't need to track them.
In addition, I streamlined the padding code. And the score model implementation can now also deal with arbitrary dimensions, so we can specify priors on SED, etc.
Following the discussion in #34, this PR brings a correction to the treatment of
ScorePrior
. In particular, there is noScorePrior.transform
anymore because these transformations are already applied before the prior gets to see the parameters. Autodiff then takes care of the Jacobians, so we simply don't need to track them.In addition, I streamlined the padding code. And the score model implementation can now also deal with arbitrary dimensions, so we can specify priors on SED, etc.