Add getmodel and setmodel from/to LogDensityFunction
#626
Conversation
sunxd3
changed the title
setmodel and getmodel from/to LogDensityFunction getmodel and setmodel from/to LogDensityFunction
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
ext/DynamicPPLReverseDiffExt.jl
Outdated
| @@ -23,4 +25,17 @@ function LogDensityProblemsAD.ADgradient( | |||
| ) | |||
| end | |||
|
|
|||
| function DynamicPPL.setmodel( | |||
| f::LogDensityProblemsAD.ReverseDiffLogDensity{L,Nothing}, model::DynamicPPL.Model | |||
There was a problem hiding this comment.
@torfjelde @yebai this will error because ReverseDiffLogDensity can't be found. It is defined under package extension(https://github.com/tpapp/LogDensityProblemsAD.jl/blob/449e5661bc2667f7bef061e148a6ea5526cbb427/ext/LogDensityProblemsADReverseDiffExt.jl#L25), do you know how to handle it here?
There was a problem hiding this comment.
You can't dispatch on types from package extensions. IIRC, that is a limitation.
There was a problem hiding this comment.
Hmm, yeah this is annoying 😕
Don't really see how can avoid this tbh. One alternative is to just define our own AD log density model which contains the ADTypes.AbstractADType instead and uses DifferentiationInterface.jl or something. Or we raise an issue in LogDensityProblemsAD.jl to make the structs part of the main package instead of the extension to allow this stuff.
There was a problem hiding this comment.
It makes sense to move these structs into the main package.
|
@yebai @torfjelde I don't think we need this PR now. The motivation of this PR was that we want customized |
|
@sunxd3 let's still transfer these methods from |
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Pull Request Test Coverage Report for Build 9992511738Details
💛 - Coveralls |
| @@ -1,6 +1,7 @@ | |||
| [deps] | |||
| Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f" | |||
| DynamicPPL = "366bfd00-2699-11ea-058f-f148b4cae6d8" | |||
| HypothesisTests = "09f84164-cd44-5f33-b23f-e6b0d136a0d5" | |||
There was a problem hiding this comment.
It is -- without this, Turing integration tests fail (ref https://github.com/TuringLang/Turing.jl/blob/29a134245b2499d59fa992420eba37ab2b9f5945/test/test_utils/numerical_tests.jl#L6)
There was a problem hiding this comment.
But maybe better solution? 🤔
|
@torfjelde another look? 🙏 |
|
LGTM:) |
ref: TuringLang/Turing.jl#2231 (comment)