Open devmotion opened 3 years ago
In DistributionsAD we introduced a custom way to specify mappings from R^n to the desired subspace to avoid these issues in our tests (see, e.g., https://github.com/TuringLang/DistributionsAD.jl/blob/d5a479acc90acd9fe42f1abd207788c164ac85af/test/ad/distributions.jl#L252)
Related to https://github.com/JuliaDiff/FiniteDifferences.jl/issues/52, it seems there is currently no way to specify constraints on function arguments to FiniteDifferences. For instance, when testing the ChainRule adjoint of (an extension of)
binomlogpdf(n, p, x)
inputp
has to be constrained to the interval [0, 1] since otherwise the forward pass errors.