st--
Despite https://github.com/JuliaDiff/ChainRules.jl/pull/613 this seems to still be broken :(
How about sticking to julia 1.6 for the docs in this PR, and we can sort out the upgrade separately ?
Bigger allocations, but less of them (and only a constant number). Few large allocations is more efficient than lots of small ones, I suppose!
It might be helpful to see how it scales with number of Gauss-Hermite points, 10 vs 100 vs 1000...
Woops, yes, definitely an unintentional bug. Thanks for spotting it!
fixed by #119
I just realised I could've resolved this with #109 in one go :facepalm:
hadn't looked so good to the CI :sweat_smile:
@willtebbutt any idea what's going on with the Julia 1 tests? :thinking:
We could still dispatch analytic implementations of expected_loglik etc on `typeof(exp)` etc, right?