Skip to content
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

Jump to bottom

covariance matrix cannot be estimated using Wishart with NUTS, HMC or HMCDA samplers #1861

Closed
DDoyle1066 opened this issue Jul 20, 2022 · 2 comments
Closed

covariance matrix cannot be estimated using Wishart with NUTS, HMC or HMCDA samplers #1861

DDoyle1066 opened this issue Jul 20, 2022 · 2 comments

Comments

@DDoyle1066
Copy link

DDoyle1066 commented Jul 20, 2022 •
edited
Loading

I ran into an issue when trying to estimate a covariance matrix using an inverse Wishart or a wishart distibution (which I believe is the appropriate prior). Below is a minimal example:

using Turing
true_cov = [0.7 0.5; 0.5 0.8]
true_mu = [-0.3, 0.2]
true_dist = MvNormal(true_mu, true_cov)
rng = MersenneTwister(20220720)
data = rand(rng, true_dist, 10000)
cov(data')

@model function norm_est(data)
  D, N = size(data)
  covar ~ Wishart(3, true_cov)
  mu ~ filldist(Normal(0, 1), D)
  data ~ filldist(MvNormal(mu, Hermitian(covar)), N) # tried to force it to be hermitian which was a different error but still got positive definite issue
end

est = norm_est(data)
chain_that_does_not_work = sample(est, NUTS(), 500)
chain_that_works = sample(est, IS(), 500)

The error I get says that the matrix I am using is not positive definite which I believe is some sort of tolerance error. it only affects NUTS, HMC and HMCDA samplers as far as I can tell because of a gradient calculation. I am reluctant to use the other samplers because they did not give accurate results for this example at least. Full error is below:

PosDefException: matrix is not positive definite; Cholesky factorization failed.
Stacktrace:
[1] checkpositivedefinite
@ /julia-1.7.3/share/julia/stdlib/v1.7/LinearAlgebra/src/factorization.jl:18 [inlined]
[2] #cholesky!#138
@ /julia-1.7.3/share/julia/stdlib/v1.7/LinearAlgebra/src/cholesky.jl:266 [inlined]
[3] cholesky!(A::Matrix{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, ::Val{false}; check::Bool)
@ LinearAlgebra /julia-1.7.3/share/julia/stdlib/v1.7/LinearAlgebra/src/cholesky.jl:298
[4] #cholesky#143
@ /julia-1.7.3/share/julia/stdlib/v1.7/LinearAlgebra/src/cholesky.jl:394 [inlined]
[5] cholesky (repeats 2 times)
@ /julia-1.7.3/share/julia/stdlib/v1.7/LinearAlgebra/src/cholesky.jl:394 [inlined]
[6] pd_logpdf_with_trans(d::Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}, X::Matrix{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, transform::Bool)
@ Bijectors ~/.julia/packages/Bijectors/U0SqN/src/Bijectors.jl:229
[7] logpdf_with_trans
@ ~/.julia/packages/Bijectors/U0SqN/src/Bijectors.jl:132 [inlined]
[8] assume(dist::Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}, vn::AbstractPPL.VarName{:covar, Setfield.IdentityLens}, vi::DynamicPPL.ThreadSafeVarInfo{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Set{DynamicPPL.Selector}}}}}, ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Base.RefValue{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}}})
@ DynamicPPL ~/.julia/packages/DynamicPPL/R7VK9/src/context_implementations.jl:198
[9] assume
@ ~/.julia/packages/Turing/S4Y4B/src/inference/hmc.jl:462 [inlined]
[10] tilde_assume
@ ~/.julia/packages/DynamicPPL/R7VK9/src/context_implementations.jl:49 [inlined]
[11] tilde_assume
@ ~/.julia/packages/DynamicPPL/R7VK9/src/context_implementations.jl:46 [inlined]
[12] tilde_assume
@ ~/.julia/packages/DynamicPPL/R7VK9/src/context_implementations.jl:31 [inlined]
[13] tilde_assume!!(context::DynamicPPL.SamplingContext{DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.DefaultContext, Random._GLOBAL_RNG}, right::Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}, vn::AbstractPPL.VarName{:covar, Setfield.IdentityLens}, vi::DynamicPPL.ThreadSafeVarInfo{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Set{DynamicPPL.Selector}}}}}, ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Base.RefValue{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}}})
@ DynamicPPL ~/.julia/packages/DynamicPPL/R7VK9/src/context_implementations.jl:117
[14] norm_est(model::DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, varinfo::DynamicPPL.ThreadSafeVarInfo{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Set{DynamicPPL.Selector}}}}}, ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}, Vector{Base.RefValue{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}}}, context::DynamicPPL.SamplingContext{DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.DefaultContext, Random._GLOBAL_RNG}, data::Matrix{Float64})
@ Main /app/ESG_Enhancements.jl:58
[15] macro expansion
@ ~/.julia/packages/DynamicPPL/R7VK9/src/model.jl:493 [inlined]
[16] _evaluate!!
@ ~/.julia/packages/DynamicPPL/R7VK9/src/model.jl:476 [inlined]
[17] evaluate_threadsafe!!
@ ~/.julia/packages/DynamicPPL/R7VK9/src/model.jl:467 [inlined]
[18] evaluate!!
@ ~/.julia/packages/DynamicPPL/R7VK9/src/model.jl:402 [inlined]
[19] evaluate!!
@ ~/.julia/packages/DynamicPPL/R7VK9/src/model.jl:415 [inlined]
[20] evaluate!!
@ ~/.julia/packages/DynamicPPL/R7VK9/src/model.jl:423 [inlined]
[21] (::Turing.LogDensityFunction{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.DefaultContext})(θ::Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}})
@ Turing ~/.julia/packages/Turing/S4Y4B/src/Turing.jl:37
[22] vector_mode_dual_eval!
@ ~/.julia/packages/ForwardDiff/wAaVJ/src/apiutils.jl:37 [inlined]
[23] vector_mode_gradient!(result::DiffResults.MutableDiffResult{1, Float64, Tuple{Vector{Float64}}}, f::Turing.LogDensityFunction{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.DefaultContext}, x::Vector{Float64}, cfg::ForwardDiff.GradientConfig{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}})
@ ForwardDiff ~/.julia/packages/ForwardDiff/wAaVJ/src/gradient.jl:113
[24] gradient!
@ ~/.julia/packages/ForwardDiff/wAaVJ/src/gradient.jl:37 [inlined]
[25] gradient!(result::DiffResults.MutableDiffResult{1, Float64, Tuple{Vector{Float64}}}, f::Turing.LogDensityFunction{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.DefaultContext}, x::Vector{Float64}, cfg::ForwardDiff.GradientConfig{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6, Vector{ForwardDiff.Dual{ForwardDiff.Tag{Turing.TuringTag, Float64}, Float64, 6}}})
@ ForwardDiff ~/.julia/packages/ForwardDiff/wAaVJ/src/gradient.jl:35
[26] gradient_logp(ad::Turing.Essential.ForwardDiffAD{0, true}, θ::Vector{Float64}, vi::DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, model::DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, sampler::DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, context::DynamicPPL.DefaultContext)
@ Turing.Essential ~/.julia/packages/Turing/S4Y4B/src/essential/ad.jl:130
[27] gradient_logp (repeats 2 times)
@ ~/.julia/packages/Turing/S4Y4B/src/essential/ad.jl:88 [inlined]
[28] ∂logπ∂θ
@ ~/.julia/packages/Turing/S4Y4B/src/inference/hmc.jl:433 [inlined]
[29] ∂H∂θ
@ ~/.julia/packages/AdvancedHMC/51xgc/src/hamiltonian.jl:31 [inlined]
[30] macro expansion
@ ~/.julia/packages/UnPack/EkESO/src/UnPack.jl:100 [inlined]
[31] step(lf::AdvancedHMC.Leapfrog{Float64}, h::AdvancedHMC.Hamiltonian{AdvancedHMC.UnitEuclideanMetric{Float64, Tuple{Int64}}, Turing.LogDensityFunction{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.DefaultContext}, Turing.Inference.var"#∂logπ∂θ#53"{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}}}, z::AdvancedHMC.PhasePoint{Vector{Float64}, AdvancedHMC.DualValue{Float64, Vector{Float64}}}, n_steps::Int64; fwd::Bool, full_trajectory::Val{false})
@ AdvancedHMC ~/.julia/packages/AdvancedHMC/51xgc/src/integrator.jl:88
[32] step (repeats 2 times)
@ ~/.julia/packages/AdvancedHMC/51xgc/src/integrator.jl:66 [inlined]
[33] A
@ ~/.julia/packages/AdvancedHMC/51xgc/src/trajectory.jl:692 [inlined]
[34] find_good_stepsize(rng::Random._GLOBAL_RNG, h::AdvancedHMC.Hamiltonian{AdvancedHMC.UnitEuclideanMetric{Float64, Tuple{Int64}}, Turing.LogDensityFunction{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.DefaultContext}, Turing.Inference.var"#∂logπ∂θ#53"{DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}, DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}}}, θ::Vector{Float64}; max_n_iters::Int64)
@ AdvancedHMC ~/.julia/packages/AdvancedHMC/51xgc/src/trajectory.jl:714
[35] #find_good_stepsize#19
@ ~/.julia/packages/AdvancedHMC/51xgc/src/trajectory.jl:770 [inlined]
[36] find_good_stepsize
@ ~/.julia/packages/AdvancedHMC/51xgc/src/trajectory.jl:770 [inlined]
[37] initialstep(rng::Random._GLOBAL_RNG, model::DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, spl::DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, vi::DynamicPPL.TypedVarInfo{NamedTuple{(:covar, :mu), Tuple{DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:covar, Setfield.IdentityLens}, Int64}, Vector{Wishart{Float64, PDMats.PDMat{Float64, Matrix{Float64}}, Int64}}, Vector{AbstractPPL.VarName{:covar, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}, DynamicPPL.Metadata{Dict{AbstractPPL.VarName{:mu, Setfield.IdentityLens}, Int64}, Vector{DistributionsAD.TuringScalMvNormal{Vector{Float64}, Float64}}, Vector{AbstractPPL.VarName{:mu, Setfield.IdentityLens}}, Vector{Float64}, Vector{Set{DynamicPPL.Selector}}}}}, Float64}; init_params::Nothing, nadapts::Int64, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
@ Turing.Inference ~/.julia/packages/Turing/S4Y4B/src/inference/hmc.jl:187
[38] step(rng::Random._GLOBAL_RNG, model::DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, spl::DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}; resume_from::Nothing, init_params::Nothing, kwargs::Base.Pairs{Symbol, Int64, Tuple{Symbol}, NamedTuple{(:nadapts,), Tuple{Int64}}})
@ DynamicPPL ~/.julia/packages/DynamicPPL/R7VK9/src/sampler.jl:104
[39] macro expansion
@ ~/.julia/packages/AbstractMCMC/fnRmh/src/sample.jl:120 [inlined]
[40] macro expansion
@ ~/.julia/packages/ProgressLogging/6KXlp/src/ProgressLogging.jl:328 [inlined]
[41] macro expansion
@ ~/.julia/packages/AbstractMCMC/fnRmh/src/logging.jl:9 [inlined]
[42] mcmcsample(rng::Random._GLOBAL_RNG, model::DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, sampler::DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, N::Int64; progress::Bool, progressname::String, callback::Nothing, discard_initial::Int64, thinning::Int64, chain_type::Type, kwargs::Base.Pairs{Symbol, Int64, Tuple{Symbol}, NamedTuple{(:nadapts,), Tuple{Int64}}})
@ AbstractMCMC ~/.julia/packages/AbstractMCMC/fnRmh/src/sample.jl:111
[43] sample(rng::Random._GLOBAL_RNG, model::DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, sampler::DynamicPPL.Sampler{HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}}, N::Int64; chain_type::Type, resume_from::Nothing, progress::Bool, nadapts::Int64, discard_adapt::Bool, discard_initial::Int64, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
@ Turing.Inference ~/.julia/packages/Turing/S4Y4B/src/inference/hmc.jl:133
[44] sample
@ ~/.julia/packages/Turing/S4Y4B/src/inference/hmc.jl:116 [inlined]
[45] #sample#2
@ ~/.julia/packages/Turing/S4Y4B/src/inference/Inference.jl:145 [inlined]
[46] sample
@ ~/.julia/packages/Turing/S4Y4B/src/inference/Inference.jl:145 [inlined]
[47] #sample#1
@ ~/.julia/packages/Turing/S4Y4B/src/inference/Inference.jl:135 [inlined]
[48] sample(model::DynamicPPL.Model{typeof(norm_est), (:data,), (), (), Tuple{Matrix{Float64}}, Tuple{}, DynamicPPL.DefaultContext}, alg::HMCDA{Turing.Essential.ForwardDiffAD{0}, (), AdvancedHMC.UnitEuclideanMetric}, N::Int64)
@ Turing.Inference ~/.julia/packages/Turing/S4Y4B/src/inference/Inference.jl:135
[49] top-level scope
@ /app/ESG_Enhancements.jl:64
@LLTeixeira
Copy link

Unfortunately I am having the same issue on v0.24 but using the Inverse Wishart distribution for the covariance matrix. Only happening with the NUTS, HMC and HMCDA samplers as well and with Gibbs whenever it is composed with one of those.

function MultiNorm_Cov_Model(X,::Type{T}=Float64) where T
    rows = size(X, 1)
    vars = size(X, 2)

    μ = Vector{T}(undef, vars)
    for i in 1:vars
        μ[i] ~ Normal(0, 50)
    end

    Σ ~ InverseWishart(vars + 1, Matrix(I(vars)*1))
    
    for i in 1:rows
        X[i,:] ~ MvNormal(μ, Symmetric(Σ))
    end
end

Using the "Symmetric" function doesn't solve the issue. And I still get:

"PosDefException: matrix is not Hermitian; Cholesky factorization failed."

@yebai
Copy link
Member

yebai commented Jun 6, 2024

Duplicate with #2188 (comment)

@yebai yebai closed this as completed Jun 6, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@yebai @DDoyle1066 @LLTeixeira