David Widmann
David Widmann
I started the registration process for PDMats 0.11.11: https://github.com/JuliaRegistries/General/pull/61420 Let's rerun tests when it's available.
> Do you happen to have a case that you know fails under the newer Roots. I can dig through the test cases in `tests/interface.jl` but if it were easy...
> In the modified example below, `basic_bisection` seems fine when the number of steps is capped at 16, but not 17: Ooh, that's very weird. Ad hoc, I don't know...
> #315 addresses this by reverting to the old behavior, It seems that is sufficient to fix the example above at least. With Roots 2.0.1 I get ```julia julia> using...
This does not fix the issue with `basic_bisection` above but maybe we should also define a dedicated forward-mode rule for ForwardDiff in Bijectors that uses the partial derivatives directly (like...
Ref: https://github.com/TuringLang/Bijectors.jl/pull/219/commits/3641c0ae342476c235ff381cd71b308f1cba18ef (works also with Roots 2.0.0)
Thanks, I'll close this issue then :slightly_smiling_face:
But we already have this in place (implementing our own pairwise), just not with Tullio and unnecessarily restricted to `::PreMetric`? E.g., for `Delta` we already use `broadcast(::Delta, x, y)` if...
> More updates, GPU is working with Tullio :) and is around 250x faster on my machine! I guess this is not surprising as Distances uses scalar indexing which is...
We can't pass additional arguments to `kernel(x, y)`, so implementation-wise the base kernel would have to be a field of `kernel`. More generally, I wonder though if this generalization is...