Skip to content

more test fixes #2491

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 12 commits into from
Feb 25, 2025
Merged

more test fixes #2491

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
b21bc4b
Remove LogDensityProblemsAD, part 1
penelopeysm Feb 20, 2025
fe46018
update Optimisation code to not use LogDensityProblemsAD
penelopeysm Feb 20, 2025
f83d695
Fix field name change
penelopeysm Feb 20, 2025
68ceb39
Don't put chunksize=0
penelopeysm Feb 20, 2025
71547c9
Remove LogDensityProblemsAD dep
penelopeysm Feb 20, 2025
62ab4b1
Improve OptimLogDensity docstring
penelopeysm Feb 20, 2025
48dac18
Remove unneeded model argument to _optimize
penelopeysm Feb 20, 2025
56997c1
Fix more tests
penelopeysm Feb 20, 2025
ef7328f
Remove essential/ad from the list of CI groups
penelopeysm Feb 20, 2025
7c32e3e
Fix HMC function
penelopeysm Feb 20, 2025
cb95ee1
More test fixes
penelopeysm Feb 20, 2025
ea00f86
Merge branch 'mhauru/dppl-0.35' into py/more-dppl-fixes
penelopeysm Feb 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion src/mcmc/gibbs.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -438,7 +438,9 @@ function setparams_varinfo!!(
state::TuringState,
params::AbstractVarInfo,
)
logdensity = DynamicPPL.setmodel(state.ldf, model, sampler.alg.adtype)
logdensity = DynamicPPL.LogDensityFunction(
model, state.ldf.varinfo, state.ldf.context; adtype=sampler.alg.adtype
)
Comment on lines -441 to +443
Copy link
Member Author

@penelopeysm penelopeysm Feb 20, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

setmodel technically still exists, but it doesn't take the adtype argument anymore, and i think it's clearer to reconstruct the LDF (which is what setmodel does anyway)

new_inner_state = setparams_varinfo!!(
AbstractMCMC.LogDensityModel(logdensity), sampler, state.state, params
)
Expand Down
6 changes: 3 additions & 3 deletions src/mcmc/sghmc.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,7 @@ function DynamicPPL.initialstep(
DynamicPPL.SamplingContext(spl, DynamicPPL.DefaultContext());
adtype=spl.alg.adtype,
)
state = SGHMCState(ℓ, vi, zero(vi[spl]))
state = SGHMCState(ℓ, vi, zero(vi[:]))

return sample, state
end
Expand All @@ -87,7 +87,7 @@ function AbstractMCMC.step(
# Compute gradient of log density.
ℓ = state.logdensity
vi = state.vi
θ = vi[spl]
θ = vi[:]
grad = last(LogDensityProblems.logdensity_and_gradient(ℓ, θ))

# Update latent variables and velocity according to
Expand Down Expand Up @@ -246,7 +246,7 @@ function AbstractMCMC.step(
# Perform gradient step.
ℓ = state.logdensity
vi = state.vi
θ = vi[spl]
θ = vi[:]
grad = last(LogDensityProblems.logdensity_and_gradient(ℓ, θ))
step = state.step
stepsize = spl.alg.stepsize(step)
Expand Down
4 changes: 2 additions & 2 deletions test/mcmc/Inference.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -512,7 +512,7 @@ using Turing

@model function vdemo2(x)
μ ~ MvNormal(zeros(size(x, 1)), I)
return x .~ MvNormal(μ, I)
return x ~ filldist(MvNormal(μ, I), size(x, 2))
end

D = 2
Expand Down Expand Up @@ -560,7 +560,7 @@ using Turing

@model function vdemo7()
x = Array{Real}(undef, N, N)
return x .~ [InverseGamma(2, 3) for i in 1:N]
return x ~ filldist(InverseGamma(2, 3), N, N)
Comment on lines 561 to +563
Copy link
Member Author

@penelopeysm penelopeysm Feb 20, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm kind of unsure as to whether we really need to keep the vdemoN models around. As far as I can tell, they're just legacy tests that try out different 'forms' of vectorisation and presumably the timings that are printed are meant to be some marker of their performance. But surely this is covered by DynamicPPL demo models already? I personally wouldn't mind if we just deleted the entire block

end

sample(StableRNG(seed), vdemo7(), alg, 10)
Expand Down
2 changes: 1 addition & 1 deletion test/mcmc/hmc.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -218,7 +218,7 @@ using Turing
# https://github.com/TuringLang/Turing.jl/issues/1308
@model function mwe3(::Type{T}=Array{Float64}) where {T}
m = T(undef, 2, 3)
return m .~ MvNormal(zeros(2), I)
return m ~ filldist(MvNormal(zeros(2), I), 3)
end
@test sample(StableRNG(seed), mwe3(), HMC(0.2, 4; adtype=adbackend), 100) isa Chains
end
Expand Down
2 changes: 1 addition & 1 deletion test/mcmc/mh.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -238,7 +238,7 @@ GKernel(var) = (x) -> Normal(x, sqrt.(var))

# Link if proposal is `AdvancedHM.RandomWalkProposal`
vi = deepcopy(vi_base)
d = length(vi_base[DynamicPPL.SampleFromPrior()])
d = length(vi_base[:])
alg = MH(AdvancedMH.RandomWalkProposal(MvNormal(zeros(d), I)))
spl = DynamicPPL.Sampler(alg)
vi = Turing.Inference.maybe_link!!(vi, spl, alg.proposals, gdemo_default)
Expand Down
Loading