Skip to content

Commit

Permalink
Fix for issue #95 (#96)
Browse files Browse the repository at this point in the history 
* fix for issue #95

* fixed failing tests

* remove usage of stack and stop logging so much progress

* removed more progress logging in test suite

* removed more progress logging

* bump patch version
  • Loading branch information
@torfjelde
torfjelde authored Jun 5, 2024
1 parent b7829cb commit 5aa196c
Show file tree
Hide file tree
Showing 4 changed files with 94 additions and 45 deletions.
2 changes: 1 addition & 1 deletion Project.toml
View file
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
name = "AdvancedMH"
uuid = "5b7e9947-ddc0-4b3f-9b55-0d8042f74170"
version = "0.8.1"
version = "0.8.2"

[deps]
AbstractMCMC = "80f14c24-f653-4e6a-9b94-39d6b0f70001"
Expand Down
9 changes: 3 additions & 6 deletions src/MALA.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -62,8 +62,8 @@ function AbstractMCMC.step(

# Compute the log ratio of proposal densities.
logratio_proposal_density = q(
proposal(-gradient_logdensity_candidate), state, candidate
) - q(proposal(-gradient_logdensity_state), candidate, state)
proposal(gradient_logdensity_candidate), state, candidate
) - q(proposal(gradient_logdensity_state), candidate, state)

# Compute the log acceptance probability.
logα = logdensity_candidate - logdensity_state + logratio_proposal_density
Expand All @@ -72,10 +72,7 @@ function AbstractMCMC.step(
transition = if -Random.randexp(rng) < logα
GradientTransition(candidate, logdensity_candidate, gradient_logdensity_candidate, true)
else
candidate = transition_prev.params
lp = transition_prev.lp
gradient = transition_prev.gradient
GradientTransition(candidate, lp, gradient, false)
GradientTransition(transition_prev.params, transition_prev.lp, transition_prev.gradient, false)
end

return transition, transition
Expand Down
6 changes: 4 additions & 2 deletions test/emcee.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@
sampler = Ensemble(1_000, StretchProposal([InverseGamma(2, 3), Normal(0, 1)]))

chain = sample(model, sampler, 1_000;
param_names = ["s", "m"], chain_type = Chains)
param_names = ["s", "m"], chain_type = Chains, progress=false)
@test chain isa Chains
@test range(chain) == 1:1_000
@test mean(chain["s"]) ≈ 49/24 atol=0.1
Expand All @@ -33,6 +33,7 @@
chain_type = Chains,
discard_initial=25,
thinning=4,
progress=false
)
@test chain2 isa Chains
@test range(chain2) == range(26; step=4, length=1_000)
Expand All @@ -59,7 +60,7 @@
Random.seed!(100)
sampler = Ensemble(1_000, StretchProposal(MvNormal(zeros(2), I)))
chain = sample(model, sampler, 1_000;
param_names = ["logs", "m"], chain_type = Chains)
param_names = ["logs", "m"], chain_type = Chains, progress=false)
@test chain isa Chains
@test range(chain) == 1:1_000
@test mean(exp, chain["logs"]) ≈ 49/24 atol=0.1
Expand All @@ -73,6 +74,7 @@
chain_type = Chains,
discard_initial=25,
thinning=4,
progress=false
)
@test chain2 isa Chains
@test range(chain2) == range(26; step=4, length=1_000)
Expand Down
122 changes: 86 additions & 36 deletions test/runtests.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -40,9 +40,9 @@ include("util.jl")
spl3 = StaticMH(2)

# Sample from the posterior.
chain1 = sample(model, spl1, 100000; chain_type=StructArray, param_names=["μ", "σ"])
chain2 = sample(model, spl2, 100000; chain_type=StructArray, param_names=["μ", "σ"])
chain3 = sample(model, spl3, 100000; chain_type=StructArray, param_names=["μ", "σ"])
chain1 = sample(model, spl1, 100000; chain_type=StructArray, param_names=["μ", "σ"], progress=false)
chain2 = sample(model, spl2, 100000; chain_type=StructArray, param_names=["μ", "σ"], progress=false)
chain3 = sample(model, spl3, 100000; chain_type=StructArray, param_names=["μ", "σ"], progress=false)

# chn_mean ≈ dist_mean atol=atol_v
@test mean(chain1.μ) ≈ 0.0 atol=0.1
Expand All @@ -60,9 +60,9 @@ include("util.jl")
spl3 = RWMH(2)

# Sample from the posterior.
chain1 = sample(model, spl1, 100000; chain_type=StructArray, param_names=["μ", "σ"])
chain2 = sample(model, spl2, 100000; chain_type=StructArray, param_names=["μ", "σ"])
chain3 = sample(model, spl3, 200000; chain_type=StructArray, param_names=["μ", "σ"])
chain1 = sample(model, spl1, 100000; chain_type=StructArray, param_names=["μ", "σ"], progress=false)
chain2 = sample(model, spl2, 100000; chain_type=StructArray, param_names=["μ", "σ"], progress=false)
chain3 = sample(model, spl3, 200000; chain_type=StructArray, param_names=["μ", "σ"], progress=false)

# chn_mean ≈ dist_mean atol=atol_v
@test mean(chain1.μ) ≈ 0.0 atol=0.1
Expand All @@ -77,13 +77,13 @@ include("util.jl")
spl1 = StaticMH([Normal(0,1), Normal(0, 1)])

chain1 = sample(model, spl1, MCMCDistributed(), 10000, 4;
param_names=["μ", "σ"], chain_type=Chains)
param_names=["μ", "σ"], chain_type=Chains, progress=false)
@test mean(chain1["μ"]) ≈ 0.0 atol=0.1
@test mean(chain1["σ"]) ≈ 1.0 atol=0.1

if VERSION >= v"1.3"
chain2 = sample(model, spl1, MCMCThreads(), 10000, 4;
param_names=["μ", "σ"], chain_type=Chains)
param_names=["μ", "σ"], chain_type=Chains, progress=false)
@test mean(chain2["μ"]) ≈ 0.0 atol=0.1
@test mean(chain2["σ"]) ≈ 1.0 atol=0.1
end
Expand All @@ -93,7 +93,7 @@ include("util.jl")
# Array of parameters
chain1 = sample(
model, StaticMH([Normal(0,1), Normal(0, 1)]), 10_000;
param_names=["μ", "σ"], chain_type=Chains
param_names=["μ", "σ"], chain_type=Chains, progress=false
)
@test chain1 isa Chains
@test range(chain1) == 1:10_000
Expand All @@ -103,6 +103,7 @@ include("util.jl")
chain1b = sample(
model, StaticMH([Normal(0,1), Normal(0, 1)]), 10_000;
param_names=["μ", "σ"], chain_type=Chains, discard_initial=25, thinning=4,
progress=false
)
@test chain1b isa Chains
@test range(chain1b) == range(26; step=4, length=10_000)
Expand All @@ -115,7 +116,8 @@ include("util.jl")
MetropolisHastings(
(μ = StaticProposal(Normal(0,1)), σ = StaticProposal(Normal(0, 1)))
), 10_000;
chain_type=Chains
chain_type=Chains,
progress=false
)
@test chain2 isa Chains
@test range(chain2) == 1:10_000
Expand All @@ -128,6 +130,7 @@ include("util.jl")
(μ = StaticProposal(Normal(0,1)), σ = StaticProposal(Normal(0, 1)))
), 10_000;
chain_type=Chains, discard_initial=25, thinning=4,
progress=false
)
@test chain2b isa Chains
@test range(chain2b) == range(26; step=4, length=10_000)
Expand All @@ -137,7 +140,8 @@ include("util.jl")
# Scalar parameter
chain3 = sample(
DensityModel(x -> loglikelihood(Normal(x, 1), data)),
StaticMH(Normal(0, 1)), 10_000; param_names=["μ"], chain_type=Chains
StaticMH(Normal(0, 1)), 10_000; param_names=["μ"], chain_type=Chains,
progress=false
)
@test chain3 isa Chains
@test range(chain3) == 1:10_000
Expand All @@ -147,6 +151,7 @@ include("util.jl")
DensityModel(x -> loglikelihood(Normal(x, 1), data)),
StaticMH(Normal(0, 1)), 10_000;
param_names=["μ"], chain_type=Chains, discard_initial=25, thinning=4,
progress=false
)
@test chain3b isa Chains
@test range(chain3b) == range(26; step=4, length=10_000)
Expand All @@ -164,10 +169,10 @@ include("util.jl")
p3 = (a=StaticProposal(Normal(0,1)), b=StaticProposal(InverseGamma(2,3)))
p4 = StaticProposal((x=1.0) -> Normal(x, 1))

c1 = sample(m1, MetropolisHastings(p1), 100; chain_type=Vector{NamedTuple})
c2 = sample(m2, MetropolisHastings(p2), 100; chain_type=Vector{NamedTuple})
c3 = sample(m3, MetropolisHastings(p3), 100; chain_type=Vector{NamedTuple})
c4 = sample(m4, MetropolisHastings(p4), 100; chain_type=Vector{NamedTuple})
c1 = sample(m1, MetropolisHastings(p1), 100; chain_type=Vector{NamedTuple}, progress=false)
c2 = sample(m2, MetropolisHastings(p2), 100; chain_type=Vector{NamedTuple}, progress=false)
c3 = sample(m3, MetropolisHastings(p3), 100; chain_type=Vector{NamedTuple}, progress=false)
c4 = sample(m4, MetropolisHastings(p4), 100; chain_type=Vector{NamedTuple}, progress=false)

@test keys(c1[1]) == (:param_1, :lp)
@test keys(c2[1]) == (:param_1, :param_2, :lp)
Expand All @@ -182,7 +187,7 @@ include("util.jl")
val = [0.4, 1.2]

# Sample from the posterior.
chain1 = sample(model, spl1, 10, initial_params = val)
chain1 = sample(model, spl1, 10, initial_params = val, progress=false)

@test chain1[1].params == val
end
Expand All @@ -199,12 +204,12 @@ include("util.jl")
p1 = RandomWalkProposal(CustomNormal())
@test p1 isa RandomWalkProposal{false}
@test_throws MethodError AdvancedMH.logratio_proposal_density(p1, randn(), randn())
@test_throws MethodError sample(m1, MetropolisHastings(p1), 10)
@test_throws MethodError sample(m1, MetropolisHastings(p1), 10, progress=false)

p1 = StaticProposal(x -> CustomNormal(x))
@test p1 isa StaticProposal{false}
@test_throws MethodError AdvancedMH.logratio_proposal_density(p1, randn(), randn())
@test_throws MethodError sample(m1, MetropolisHastings(p1), 10)
@test_throws MethodError sample(m1, MetropolisHastings(p1), 10, progress=false)

# If the proposal is declared to be symmetric, the log ratio of the proposal
# density is not evaluated.
Expand All @@ -227,7 +232,8 @@ include("util.jl")
))
chain1 = sample(
m1, MetropolisHastings(p2), 100000;
chain_type=StructArray, param_names=["x"]
chain_type=StructArray, param_names=["x"],
progress=false
)
@test mean(chain1.x) ≈ mean(d1) atol=0.05
@test std(chain1.x) ≈ std(d1) atol=0.05
Expand Down Expand Up @@ -260,29 +266,73 @@ include("util.jl")
end

@testset "MALA" begin
# Set up the sampler.
σ² = 0.01
spl1 = MALA(x -> MvNormal((σ² / 2) .* x, σ² * I))
@testset "basic" begin
# Set up the sampler.
σ² = 1e-3
spl1 = MALA(x -> MvNormal((σ² / 2) .* x, σ² * I))

# Sample from the posterior with initial parameters.
chain1 = sample(model, spl1, 100000; initial_params=ones(2), chain_type=StructArray, param_names=["μ", "σ"])
# Sample from the posterior with initial parameters.
chain1 = sample(
model, spl1, 1000;
initial_params=ones(2),
chain_type=StructArray,
param_names=["μ", "σ"],
discard_initial=100,
progress=false
)

@test mean(chain1.μ) ≈ 0.0 atol=0.1
@test mean(chain1.σ) ≈ 1.0 atol=0.1
@test mean(chain1.μ) ≈ 0.0 atol = 0.1
@test mean(chain1.σ) ≈ 1.0 atol = 0.1

@testset "LogDensityProblems interface" begin
admodel = LogDensityProblemsAD.ADgradient(Val(:ForwardDiff), density)
chain2 = sample(
admodel,
spl1,
1000;
initial_params=ones(2),
chain_type=StructArray,
param_names=["μ", "σ"],
discard_initial=100,
progress=false
)

@test mean(chain2.μ) ≈ 0.0 atol = 0.1
@test mean(chain2.σ) ≈ 1.0 atol = 0.1
end
end

@testset "LogDensityProblems interface" begin
admodel = LogDensityProblemsAD.ADgradient(Val(:ForwardDiff), density)
chain2 = sample(
admodel,
spl1,
100000;
@testset "issue #95" begin
struct TheNormalLogDensity{M}
A::M
end

# can do gradient
LogDensityProblems.capabilities(::Type{<:TheNormalLogDensity}) = LogDensityProblems.LogDensityOrder{1}()

LogDensityProblems.dimension(d::TheNormalLogDensity) = size(d.A, 1)
LogDensityProblems.logdensity(d::TheNormalLogDensity, x) = -x' * d.A * x / 2

function LogDensityProblems.logdensity_and_gradient(d::TheNormalLogDensity, x)
return -x' * d.A * x / 2, -d.A * x
end

Σ = [1.5 0.35; 0.35 1.0]
σ² = 0.5
spl = AdvancedMH.MALA(g -> Distributions.MvNormal((σ² / 2) .* g, σ² * I))

chain = sample(
TheNormalLogDensity(inv(Σ)),
spl,
500000;
initial_params=ones(2),
chain_type=StructArray,
param_names=["μ", "σ"]
progress=false
)
data = mapreduce(Base.Fix2(getproperty, :params), hcat, chain)
Σ_est = cov(data, dims=2)

@test mean(chain2.μ) ≈ 0.0 atol=0.1
@test mean(chain2.σ) ≈ 1.0 atol=0.1
@test mean(data, dims=2) ≈ zeros(2) atol = 0.1
@test Σ ≈ Σ_est atol = 2e-1
end
end

Expand Down

2 comments on commit 5aa196c

@devmotion
Copy link
Member

Choose a reason for hiding this comment

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

@JuliaRegistrator register

@JuliaRegistrator
Copy link

Choose a reason for hiding this comment

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

Registration pull request created: JuliaRegistries/General/108325

Tip: Release Notes

Did you know you can add release notes too? Just add markdown formatted text underneath the comment after the text
"Release notes:" and it will be added to the registry PR, and if TagBot is installed it will also be added to the
release that TagBot creates. i.e.

@JuliaRegistrator register

Release notes:

## Breaking changes

- blah

To add them here just re-invoke and the PR will be updated.

Tagging

After the above pull request is merged, it is recommended that a tag is created on this repository for the registered package version.

This will be done automatically if the Julia TagBot GitHub Action is installed, or can be done manually through the github interface, or via:

git tag -a v0.8.2 -m "<description of version>" 5aa196c412c2405541eb4c6980909cdb4be4745a
git push origin v0.8.2

Please sign in to comment.