David Widmann
David Widmann
What do the benchmark shows with the latest iteration of this PR? In https://github.com/JuliaMath/SpecialFunctions.jl/pull/491#issuecomment-2766902350 performance with `Float32` seemed to regress.
With the updated version in #1930, I get for `μ = LinearAlgebra.normalize!([1.0, t])` with `t` around zero ```julia julia> using Distributions, LinearAlgebra, AlgebraOfGraphics, CairoMakie julia> ts = range(-1e-7, 1e-7; length=1001);...
Fixed by #1930.
> But when autodiff tries to differentiate the logpdf with respect to κ, it treats this stored value as a constant, breaking the derivative chain. AFAICT that's not the problem,...
The numerical approximations used by the `besselix` implementation (or rather the C libraries Julia calls) break down if the arguments become too large, and then an error is thrown. I...
Sorry, I've been swamped with tasks this week. I like the proposal a lot, I think this will resolve one of my main feature requests (apart from fixing `rand`/`eltype` and...
Locally (macOS aarch64) and on MacOS (aarch64) in CI all tests pass, but on ubuntu and windows 4 tests fail...
Bump 🙂
IMO the main problem (which became apparent in recent releases but has existed for a long time) is that there exists a definition of `eigvals(::Hermitian{
IMO the example is misleading/ill-posed and ForwardDiff 0.10, Zygote, Mooncake and Enzyme are incorrect (as well): The function $f: \mathbb{R}^2 \to \mathbb{R}$ of the second output component is not differentiable...