SymbolicRegression stuck at version 0.22.2

Specific Domains Modelling & Simulations
,
1

Hi all. Just playing around with SymbolicRegression.jl and notice the version kept getting stuck on v0.22.2 instead of the latest v1.11.1. And when I force install that version, the ForwardDiff.jl went from 1.0.1 down to 0.10.38. I’m currently on Julia 1.11.5.

Since I use Turing a bit, which as I understand uses ForwradDiff as default, I’m just a tad worried that this would have some unintended consequences there. @MilesCranmer is this something that can be updated your end or does that need ForwardDiff to be refreshed as well?

1 Like
2

It is likely my fault but I don’t see where yet, so tips appreciated. The main dependency relying on ForwardDiff is DynamicDiff.jl but that seems to be compatible with v1: DynamicDiff.jl/Project.toml at 274736f413c97c6fc408dd18d596e8f4562624dc · MilesCranmer/DynamicDiff.jl · GitHub.

3

I just installed ForwardDiff.jl and SymbolicRegression.jl and they are both up-to-date. Adding Turing.jl makes it downgrade ForwardDiff.jl. However, Turing.jl is also compatible with v1 Turing.jl/Project.toml at 8be9094cc3773a488f76a234900e4b977781dc05 · TuringLang/Turing.jl · GitHub

So I’m not sure :face_with_spiral_eyes:

4

Ah. Confirmed my fault. It was this PR: CompatHelper: bump compat for DifferentiationInterface to 0.7, (keep existing compat) by github-actions[bot] · Pull Request #451 · MilesCranmer/SymbolicRegression.jl · GitHub. Will fix now.

4 Likes
5

That seems to have worked - I have ForwardDiff v1.0.1 now side-by-side SymbolicRegression. But seems like I need to force install a version rather than auto-install the latest?

The order that I installed the packages were:

IJulia, Turing, and SymbolicRegression

but then I only get SymbolicRegression v0.22.2:

image
image708×150 13.8 KB

I can then either run up or force add SymbolicRegression@1.11.2 and installed ok:

image
image678×645 83.7 KB

6

Ok now it looks to be something other than SymbolicRegression.jl.

I think it’s this: LogDensityProblemsAD.jl/Project.toml at 6c344c9d312d114509801b2756f52aaa614bc67e · tpapp/LogDensityProblemsAD.jl · GitHub

FYI @Tamas_Papp @devmotion I think the CompatHelper is failing for this repo for some reason?

I see this:

(@v1.11) pkg> activate --temp
  Activating new project at `/var/folders/1h/xyppkvx52cl6w3_h8bw_gdqh0000gr/T/jl_rm482Q`

(jl_rm482Q) pkg> add IJulia Turing SymbolicRegression
...
    Updating `/private/var/folders/1h/xyppkvx52cl6w3_h8bw_gdqh0000gr/T/jl_rm482Q/Project.toml`
  [7073ff75] + IJulia v1.29.0
  [8254be44] + SymbolicRegression v1.11.2
  [fce5fe82] + Turing v0.39.1
...
⌃ [f6369f11] + ForwardDiff v0.10.38

So indeed it’s still defaulting to the older ForwardDiff.jl. If I force that I see:

(jl_rm482Q) pkg> add ForwardDiff@1.0.1
   Resolving package versions...
    Updating `/private/var/folders/1h/xyppkvx52cl6w3_h8bw_gdqh0000gr/T/jl_rm482Q/Project.toml`
  [f6369f11] + ForwardDiff v1.0.1
    Updating `/private/var/folders/1h/xyppkvx52cl6w3_h8bw_gdqh0000gr/T/jl_rm482Q/Manifest.toml`
  [f6369f11] ↑ ForwardDiff v0.10.38 ⇒ v1.0.1
⌃ [996a588d] ↓ LogDensityProblemsAD v1.11.0 ⇒ v1.9.2
  [ce78b400] + SimpleUnPack v1.1.0

So looks like an older version of LogDensityProblemsAD.jl implicitly support it, but the explicit support (now) rules it out.

Looks like it was only added as an explicit dependency in Drop support for Julia < 1.9 by devmotion · Pull Request #36 · tpapp/LogDensityProblemsAD.jl · GitHub. So the earlier version of the library is used instead because it has fewer dependencies.

7

yes, it is the usual 60-day timeout. I am dusting off actions now, will bump versions and release.

1 Like