search

JuliaLinearAlgebra / Octavian.jl

Multi-threaded BLAS-like library that provides pure Julia matrix multiplication
https://julialinearalgebra.github.io/Octavian.jl/stable/
Other
226 stars 18 forks source link

Add HyperDualNumbersExt #179

Closed CheukHinHoJerry closed 1 year ago

CheukHinHoJerry commented 1 year ago

Ref https://github.com/JuliaLinearAlgebra/Octavian.jl/issues/178#issue-1758088520

Basically I just replace the type constrain of 'ForwardDiff.Dual' to 'HyperDualNumbers.Hyper' and create an extra module. Haven't done any serious test on it but it works okay on my end.

I guess it might be possible to combine into a single ext with ForwardDiff.Dual but I will just leave something workable for now since I am not sure where/how should I do this properly.

codecov[bot] commented 1 year ago

Codecov Report

Patch coverage: 86.95% and project coverage change: +3.97 :tada:

Comparison is base (00d50b3) 86.42% compared to head (1a10ea4) 90.40%.

Additional details and impacted files ```diff @@ Coverage Diff @@ ## master #179 +/- ## ========================================== + Coverage 86.42% 90.40% +3.97% ========================================== Files 13 14 +1 Lines 1002 1146 +144 ========================================== + Hits 866 1036 +170 + Misses 136 110 -26 ``` | [Impacted Files](https://app.codecov.io/gh/JuliaLinearAlgebra/Octavian.jl/pull/179?src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=JuliaLinearAlgebra) | Coverage Δ | | |---|---|---| | [src/Octavian.jl](https://app.codecov.io/gh/JuliaLinearAlgebra/Octavian.jl/pull/179?src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=JuliaLinearAlgebra#diff-c3JjL09jdGF2aWFuLmps) | `100.00% <ø> (+100.00%)` | :arrow_up: | | [ext/HyperDualNumbersExt.jl](https://app.codecov.io/gh/JuliaLinearAlgebra/Octavian.jl/pull/179?src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=JuliaLinearAlgebra#diff-ZXh0L0h5cGVyRHVhbE51bWJlcnNFeHQuamw=) | `86.95% <86.95%> (ø)` | | ... and [6 files with indirect coverage changes](https://app.codecov.io/gh/JuliaLinearAlgebra/Octavian.jl/pull/179/indirect-changes?src=pr&el=tree-more&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=JuliaLinearAlgebra)

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.

chriselrod commented 1 year ago

Mind adding tests, adding hyper duals as a test dependency? You can compare the answers with a naive or base multiplication. I'd reinterpret to Float64 before comparing, because comparisons tend to only check the value, and not any of the partials.

CheukHinHoJerry commented 1 year ago

Stucked on a bug on two dual arrays since I was not notcing it would be a bit different as in ForwardDiff.Dual but just got that fix and it should be fine now. Basically I checked:

Now I check also the correectness of the epsilons and the reinterpredHD function is now a sanity check which I use LinearAlgebra.mul! to calcualte only the val part of the result. I think this can be removed but it doesn't harm if we keep that.

CheukHinHoJerry commented 1 year ago

Okay, great! One last thing: mind bumping the version, so we can tag a new release?

Just bumped to 0.3.23. Thank you for your review.