-
Notifications
You must be signed in to change notification settings - Fork 17
Description
It's ocassionally not possible to test @non_differentiable using test_rrule because finite differencing gives the wrong / unreliable results.
For example, this is the case in the Vector{Floa64}(undef, 5) constructor implemented in JuliaDiff/ChainRules.jl#491 -- in this case, the values of the array produced are unreliable, so finite difference estimates of the gradient aren't consistent between different calls to the function.
One option would be to define a test_non_differentiable suite that looks something like
test_non_differentiable(f, args...; use_finite_differencing=true)in which we prescribe a set of standard tests. In the case that the user sets use_finite_differencing=false, we could just ensure that f does indeed return NoTangents, in addition to standard things like ensuring that the primal agrees with the first output of the rrule being tested.