No glue code #319

JaimeRZP · 2023-03-09T20:07:04Z

Hi All!

My name is Jaime Ruiz Zapatero and I am 3rd year PhD in astrophysics at Oxford.

@yebai and I have been discussing possible ways of interfacing Turing and AdvancedHMC in a more general without the current glue code currently present inside Turing.
This motivated by the limitations of the current interface already discussed in this PR by @sethaxen.
The fundamental idea is to extract a LogDensityProblem object from a generic Turing model which then is used to build a Hamiltonian object for AdvancedHMC.
This can be done as follows:

ctxt = model.context
vi = DynamicPPL.VarInfo(model, ctxt)
ℓ = LogDensityProblemsAD.ADgradient(DynamicPPL.LogDensityFunction(vi, model, ctxt))
hamiltonian = AdvancedHMC.Hamiltonian(metric, ℓ)

Where model is a conditioned Turing model.

I have written a Neal's funnel example of how this interface can work which you can find in this notebook. Here I have also drafted some place holder functions to wrap the interface such that it is user-friendly. Essentially, I have added an additional signature to sample and created a wrapper structure for the sampler ingridients. To do so I have tried to follow the idea proposed in this issue by @yebai.

This example is already fully functional. However, it is missing an important aspect. Turing models can have priors with hard boundaries. However, samplers work best in continuous spaces. The solution is map the bounded prior space to the unbounded sampling space. Turing does this internally but to avoid relying too much on Turing one can also do the following:

ctxt = model.context
vi = DynamicPPL.VarInfo(model, ctxt)
vi_t = Turing.link!!(vi, model) # This transforms the variables to the unbounded space

# By passing vi_t as opposed to vi to ℓ, the log density function will input and output the transformed variables
ℓ = LogDensityProblemsAD.ADgradient(DynamicPPL.LogDensityFunction(vi_t, model, ctxt))
hamiltonian = AdvancedHMC.Hamiltonian(metric, ℓ)

Then the generated samples can be transformed back to the prior space using:

function _get_dists(vi)
    mds = values(vi.metadata)
    return [md.dists[1] for md in mds]
end

dists = _get_dists(vi)
dist_lengths = [length(dist) for dist in dists]
vsyms = _name_variables(vi, dist_lengths)

function _reshape_params(x::AbstractVector)
    xx = []
    idx = 0
    for dist_length in dist_lengths
        append!(xx, [x[idx+1:idx+dist_length]])
        idx += dist_length
    end
    return xx
end

function transform(x)
    x = _reshape_params(x)
    xt = [Bijectors.link(dist, par) for (dist, par) in zip(dists, x)]
    return vcat(xt...)
end

function inv_transform(xt)
    xt = _reshape_params(xt)
    x = [Bijectors.invlink(dist, par) for (dist, par) in zip(dists, xt)]
    return vcat(x...)
end

Note that these already account for the jacobian of the transformation. Also, this might not be the most elegant way of doing the transformation.

Question: Where would you suggest writing these functions into the code? My guess would be within sampler but I don't want to mess around with the design idea.

Question: Where would you suggest incorporating the transformation functions?

Once we have interfaced with the internal sampling method, we should also be able to use AbstractMCMC to do the sample following what Turing does. I already have written something like this for a micro-canonical HMC sampler I have been working on, MicrocanonicalHMC.jl. The most involved step is to overload the AbstractMCMC.step function with AdvancedHMC equivalent.

All the best,
Jaime

github-actions · 2023-03-10T19:31:38Z

src/sampler.jl

+function sample(model::DynamicPPL.Model, ϵ::Number, TAP::Number, n_samples::Int, n_adapts::Int;
+                initial_θ=initial_θ, progress=true, kwargs...)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function sample(model::DynamicPPL.Model, ϵ::Number, TAP::Number, n_samples::Int, n_adapts::Int;

initial_θ=initial_θ, progress=true, kwargs...)

function sample(

model::DynamicPPL.Model,

ϵ::Number,

TAP::Number,

n_samples::Int,

n_adapts::Int;

initial_θ = initial_θ,

progress = true,

kwargs...,

)

github-actions · 2023-03-10T19:31:38Z

src/sampler.jl

+    vsyms = _name_variables(vi, dist_lengths)
+    d = length(vsyms)
+
+    metric =  DiagEuclideanMetric(d)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric = DiagEuclideanMetric(d)

metric = DiagEuclideanMetric(d)

github-actions · 2023-03-10T19:31:38Z

src/sampler.jl

+
+    metric =  DiagEuclideanMetric(d)
+    integrator = Leapfrog(ϵ)
+    proposal = NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

proposal = NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)

proposal = NUTS{MultinomialTS,GeneralisedNoUTurn}(integrator)

github-actions · 2023-03-10T19:31:39Z

src/sampler.jl

+    return sample(model, metric, proposal, initial_θ, n_samples, adaptor, n_adapts;
+                  progress=progress, kwargs...)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return sample(model, metric, proposal, initial_θ, n_samples, adaptor, n_adapts;

progress=progress, kwargs...)

return sample(

model,

metric,

proposal,

initial_θ,

n_samples,

adaptor,

n_adapts;

progress = progress,

kwargs...,

)

github-actions · 2023-03-10T19:31:39Z

src/turing_utils.jl

+    vsyms = keys(vi)
+    names = []
+    for (vsym, dist_length) in zip(vsyms, dist_lengths)
+        if dist_length==1


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

if dist_length==1

if dist_length == 1

github-actions · 2023-03-10T19:31:39Z

src/turing_utils.jl

+            name = [vsym]
+            append!(names, name)
+        else
+            name = [DynamicPPL.VarName(Symbol(vsym, i,)) for i in 1:dist_length]


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

name = [DynamicPPL.VarName(Symbol(vsym, i,)) for i in 1:dist_length]

name = [DynamicPPL.VarName(Symbol(vsym, i)) for i = 1:dist_length]

github-actions · 2023-03-10T19:31:40Z

src/turing_utils.jl

+        else
+            name = [DynamicPPL.VarName(Symbol(vsym, i,)) for i in 1:dist_length]
+            append!(names, name)
+         end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-03-10T19:31:40Z

src/turing_utils.jl

+         end
+    end
+    return names
+end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

JaimeRZP · 2023-03-10T19:35:38Z

I have now coded the changes into AdvancedHMC src.
Users should be able to sample a Turing model using AdvancedHMC directly by using:

n_samples, n_adapts = 10_000, 1_000
sample(model, metric, proposal, initial_θ, n_samples, adaptor, n_adapts)

or even simpler

samples, stats = sample(model, 0.1, 0.95, n_samples, n_adapts; initial_θ=initial_θ)

which will use NUTS by defautl.

The unbounded to bounded space transforms are still missing.

github-actions · 2023-05-30T17:07:21Z

src/abstractmcmc.jl

+    d = length(vsyms)
+
+    # wrap metric, kernel and adaptor into HMCSampler
+    metric =  DiagEuclideanMetric(d)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric = DiagEuclideanMetric(d)

metric = DiagEuclideanMetric(d)

github-actions · 2023-05-30T17:07:21Z

src/abstractmcmc.jl

+    kernel = AdvancedHMC.NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)
+    adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(settings.TAP, integrator))


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

kernel = AdvancedHMC.NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)

adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(settings.TAP, integrator))

kernel = AdvancedHMC.NUTS{MultinomialTS,GeneralisedNoUTurn}(integrator)

adaptor =

StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(settings.TAP, integrator))

github-actions · 2023-06-01T15:43:14Z

src/abstractmcmc.jl

@@ -25,6 +25,15 @@ struct HMCSampler{K,M,A} <: AbstractMCMC.AbstractSampler
 end
 HMCSampler(kernel, metric) = HMCSampler(kernel, metric, Adaptation.NoAdaptation())

+# Convinience constructor
+function NUTSSampler(ϵ::Float64, TAP::Float64, d::Int)
+    metric =  DiagEuclideanMetric(d)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric = DiagEuclideanMetric(d)

metric = DiagEuclideanMetric(d)

github-actions · 2023-06-01T15:43:14Z

src/abstractmcmc.jl

+function NUTSSampler(ϵ::Float64, TAP::Float64, d::Int)
+    metric =  DiagEuclideanMetric(d)
+    integrator = Leapfrog(ϵ)
+    kernel = AdvancedHMC.NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

kernel = AdvancedHMC.NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)

kernel = AdvancedHMC.NUTS{MultinomialTS,GeneralisedNoUTurn}(integrator)

github-actions · 2023-06-01T15:43:14Z

src/abstractmcmc.jl

+    kernel = AdvancedHMC.NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)
+    adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(TAP, integrator))
+    return HMCSampler(kernel, metric, adaptor)
+end    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-01T15:43:14Z

src/abstractmcmc.jl

+# No glue code #
+################
+function AbstractMCMC.sample(
+    model::DynamicPPL.Model, 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

model::DynamicPPL.Model,

model::DynamicPPL.Model,

github-actions · 2023-06-05T13:04:27Z

src/abstractmcmc.jl

    progress = true,
    verbose = false,
    callback = nothing,
    kwargs...,
-)
-    sampler = HMCSampler(kernel, metric, adaptor)
+)   


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

)

)

github-actions · 2023-06-05T13:04:27Z

src/abstractmcmc.jl

+    ℓ = LogDensityProblemsAD.ADgradient(DynamicPPL.LogDensityFunction(vi, model, ctxt))
+    d = LogDensityProblems.dimension(ℓ)
+    model = AbstractMCMC.LogDensityModel(ℓ)
+


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

github-actions · 2023-06-05T13:04:28Z

src/abstractmcmc.jl

-
-    if init_params === nothing
-        init_params = randn(rng, size(metric, 1))
+)   


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

)

)

github-actions · 2023-06-05T13:04:28Z

src/abstractmcmc.jl

+    if spl.metric == nothing
+        metric = DiagEuclideanMetric(d)
+    else
+        metric = spl.metric    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric = spl.metric

metric = spl.metric

github-actions · 2023-06-05T13:04:28Z

src/abstractmcmc.jl

+    return AbstractMCMC.step(
+        rng,
+        model,
+        spl,
+        state;
+        n_adapts = n_adapts,
+        kwargs...)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return AbstractMCMC.step(

rng,

model,

spl,

state;

n_adapts = n_adapts,

kwargs...)

return AbstractMCMC.step(rng, model, spl, state; n_adapts = n_adapts, kwargs...)

github-actions · 2023-06-05T13:04:28Z

src/constructors.jl

+    metric
+    integrator
+    kernel
+    adaptor


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric

integrator

kernel

adaptor

metric::Any

integrator::Any

kernel::Any

adaptor::Any

github-actions · 2023-06-05T13:04:28Z

src/constructors.jl

+    max_depth::Int=10,
+    Δ_max::Float64=1000.0,
+    init_ϵ::Float64=0.0,
+    metric=nothing,
+    integrator=Leapfrog,
+    kernel = NUTS_kernel{MultinomialTS, GeneralisedNoUTurn}
+)   


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

max_depth::Int=10,

Δ_max::Float64=1000.0,

init_ϵ::Float64=0.0,

metric=nothing,

integrator=Leapfrog,

kernel = NUTS_kernel{MultinomialTS, GeneralisedNoUTurn}

)

max_depth::Int = 10,

Δ_max::Float64 = 1000.0,

init_ϵ::Float64 = 0.0,

metric = nothing,

integrator = Leapfrog,

kernel = NUTS_kernel{MultinomialTS,GeneralisedNoUTurn},

)

github-actions · 2023-06-05T13:04:29Z

src/constructors.jl

+        return StanHMCAdaptor(MassMatrixAdaptor(metric),
+                              StepSizeAdaptor(TAP, integrator))
+    end                          


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return StanHMCAdaptor(MassMatrixAdaptor(metric),

StepSizeAdaptor(TAP, integrator))

end

return StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(TAP, integrator))

end

github-actions · 2023-06-05T13:04:29Z

src/constructors.jl

+                              StepSizeAdaptor(TAP, integrator))
+    end                          
+    NUTS(n_adapts, TAP, max_depth, Δ_max, init_ϵ, metric, integrator, kernel, adaptor)
+end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-05T15:51:21Z

src/constructors.jl

+
+function NUTS_kernel(integrator)
+    return HMCKernel(Trajectory{MultinomialTS}(integrator, GeneralisedNoUTurn()))
+end    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-05T15:51:21Z

src/constructors.jl

+    max_depth::Int=10,
+    Δ_max::Float64=1000.0,
+    init_ϵ::Float64=0.0,
+    metric=nothing,
+    integrator=Leapfrog)   


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

max_depth::Int=10,

Δ_max::Float64=1000.0,

init_ϵ::Float64=0.0,

metric=nothing,

integrator=Leapfrog)

max_depth::Int = 10,

Δ_max::Float64 = 1000.0,

init_ϵ::Float64 = 0.0,

metric = nothing,

integrator = Leapfrog,

)

github-actions · 2023-06-05T15:51:21Z

src/constructors.jl

+    NUTS(n_adapts, TAP, max_depth, Δ_max, init_ϵ, metric, integrator, NUTS_kernel, adaptor)
+end 
+
+export NUTS 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

export NUTS

export NUTS

github-actions · 2023-06-05T15:51:21Z

src/constructors.jl

+    function kernel(integrator)
+        return HMCKernel(Trajectory{EndPointTS}(integrator, FixedNSteps(n_leapfrog)))
+    end
+    return kernel    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return kernel

return kernel

github-actions · 2023-06-05T15:51:22Z

src/constructors.jl

+    metric
+    integrator
+    kernel


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric

integrator

kernel

metric::Any

integrator::Any

kernel::Any

github-actions · 2023-06-05T15:51:22Z

src/constructors.jl

+    function kernel(integrator)
+        return HMCKernel(Trajectory{EndPointTS}(integrator, FixedIntegrationTime(λ)))
+    end
+    return kernel    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return kernel

return kernel

github-actions · 2023-06-05T15:51:22Z

src/constructors.jl

+    n_adapts    ::  Int         # number of samples with adaption for ϵ
+    TAP         ::  Float64     # target accept rate
+    λ           ::  Float64     # target leapfrog length
+    ϵ           ::  Float64     # (initial) step size
+    metric
+    integrator
+    kernel


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

n_adapts :: Int # number of samples with adaption for ϵ

TAP :: Float64 # target accept rate

λ :: Float64 # target leapfrog length

ϵ :: Float64 # (initial) step size

metric

integrator

kernel

n_adapts::Int # number of samples with adaption for ϵ

TAP::Float64 # target accept rate

λ::Float64 # target leapfrog length

ϵ::Float64 # (initial) step size

metric::Any

integrator::Any

kernel::Any

github-actions · 2023-06-05T15:51:22Z

src/constructors.jl

+    ϵ::Float64=0.0,
+    metric=nothing,
+    integrator=Leapfrog)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

ϵ::Float64=0.0,

metric=nothing,

integrator=Leapfrog)

ϵ::Float64 = 0.0,

metric = nothing,

integrator = Leapfrog,

)

github-actions · 2023-06-05T15:51:23Z

src/constructors.jl

+        return StanHMCAdaptor(MassMatrixAdaptor(metric),
+                              StepSizeAdaptor(TAP, integrator))
+    end    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return StanHMCAdaptor(MassMatrixAdaptor(metric),

StepSizeAdaptor(TAP, integrator))

end

return StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(TAP, integrator))

end

github-actions · 2023-06-05T15:51:23Z

src/constructors.jl

+    return HMCDA(n_adapts, TAP, λ, ϵ, metric, integrator, kernel, adaptor)
+end
+
+export HMCDA


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

export HMCDA

export HMCDA

github-actions · 2023-06-06T08:59:44Z

src/abstractmcmc.jl

-        progress = false # don't use AMCMC's progress-funtionality
+)   
+    vi = kwargs[:vi]
+    d = kwargs[:d]       


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

d = kwargs[:d]

d = kwargs[:d]

github-actions · 2023-06-06T08:59:44Z

src/constructors.jl

+    metric
+    integrator
+    kernel
+    adaptor


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric

integrator

kernel

adaptor

metric::Any

integrator::Any

kernel::Any

adaptor::Any

github-actions · 2023-06-06T08:59:44Z

src/constructors.jl

+    n_adapts    ::  Int         # number of samples with adaption for ϵ
+    TAP         ::  Float64     # target accept rate
+    λ           ::  Float64     # target leapfrog length
+    ϵ           ::  Float64     # (initial) step size
+    metric
+    integrator
+    kernel
+    adaptor


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

n_adapts :: Int # number of samples with adaption for ϵ

TAP :: Float64 # target accept rate

λ :: Float64 # target leapfrog length

ϵ :: Float64 # (initial) step size

metric

integrator

kernel

adaptor

n_adapts::Int # number of samples with adaption for ϵ

TAP::Float64 # target accept rate

λ::Float64 # target leapfrog length

ϵ::Float64 # (initial) step size

metric::Any

integrator::Any

kernel::Any

adaptor::Any

github-actions · 2023-06-07T11:29:10Z

src/sampler.jl

@@ -246,4 +246,4 @@ function sample(
        @info "Finished $n_samples sampling steps for $n_chains chains in $time (s)" h κ EBFMI_est average_acceptance_rate
    end
    return θs, stats
-end
+end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-08T10:24:29Z

src/abstractmcmc.jl

+        spl,
+        state;
+        n_adapts = n_adapts,
+        kwargs...)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

kwargs...)

kwargs...,

)

github-actions · 2023-06-08T11:20:45Z

src/abstractmcmc.jl

    state::HMCState;
    nadapts::Int = 0,
    kwargs...,
-)
+)   


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

)

)

github-actions · 2023-06-08T11:20:45Z

src/abstractmcmc.jl

+getmodel(f::LogDensityProblemsAD.ADGradientWrapper) = getmodel(parent(f))
+function getdimensions(f::AbstractMCMC.LogDensityModel)
+    return LogDensityProblems.dimension(f.logdensity)
+end    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-08T13:30:45Z

src/abstractmcmc.jl

+    println(typeof(hamiltonian)<:Hamiltonian)
+    return AbstractMCMC.step(
+        rng,
+        model,
+        spl,
+        state;
+        n_adapts = n_adapts,
+        kwargs...)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

println(typeof(hamiltonian)<:Hamiltonian)

return AbstractMCMC.step(

rng,

model,

spl,

state;

n_adapts = n_adapts,

kwargs...)

println(typeof(hamiltonian) <: Hamiltonian)

return AbstractMCMC.step(rng, model, spl, state; n_adapts = n_adapts, kwargs...)

src/AdvancedHMC.jl

github-actions · 2023-06-08T16:10:49Z

src/constructors.jl

+        return StanHMCAdaptor(MassMatrixAdaptor(metric),
+                              StepSizeAdaptor(TAP, integrator))
+    end                          
+    AHMC_NUTS(n_adapts, TAP, max_depth, Δ_max, init_ϵ, metric, integrator, NUTS_kernel, adaptor)
+end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return StanHMCAdaptor(MassMatrixAdaptor(metric),

StepSizeAdaptor(TAP, integrator))

end

AHMC_NUTS(n_adapts, TAP, max_depth, Δ_max, init_ϵ, metric, integrator, NUTS_kernel, adaptor)

end

return StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(TAP, integrator))

end

AHMC_NUTS(

n_adapts,

TAP,

max_depth,

Δ_max,

init_ϵ,

metric,

integrator,

NUTS_kernel,

adaptor,

)

end

github-actions · 2023-06-08T16:10:49Z

src/constructors.jl

+    AHMC_NUTS(n_adapts, TAP, max_depth, Δ_max, init_ϵ, metric, integrator, NUTS_kernel, adaptor)
+end 
+
+export AHMC_NUTS 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

export AHMC_NUTS

export AHMC_NUTS

github-actions · 2023-06-08T16:10:49Z

src/constructors.jl

+    return AHMC_HMCDA(n_adapts, TAP, λ, ϵ, metric, integrator, kernel, adaptor)
+end
+
+export AHMC_HMCDA


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

export AHMC_HMCDA

export AHMC_HMCDA

github-actions · 2023-06-09T11:00:45Z

src/abstractmcmc.jl

+            adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),
+                                     StepSizeAdaptor(spl.alg.TAP, integrator))


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),

StepSizeAdaptor(spl.alg.TAP, integrator))

adaptor = StanHMCAdaptor(

MassMatrixAdaptor(metric),

StepSizeAdaptor(spl.alg.TAP, integrator),

)

github-actions · 2023-06-09T11:00:45Z

src/abstractmcmc.jl

+    else
+        adaptor = spl.adaptor
+        n_adapts = kwargs[:n_adapts]
+    end        


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-09T11:00:45Z

src/constructors.jl

+# Basic use
+HMCSampler(algorithm) = HMCSampler(algorithm, nothing, nothing, nothing, nothing)
+# Expert use
+HMCSampler(integrator, kernel, metric, adaptor) = HMCSampler(Custom_alg, integrator, kernel, metric, adaptor)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

HMCSampler(integrator, kernel, metric, adaptor) = HMCSampler(Custom_alg, integrator, kernel, metric, adaptor)

HMCSampler(integrator, kernel, metric, adaptor) =

HMCSampler(Custom_alg, integrator, kernel, metric, adaptor)

github-actions · 2023-06-09T11:00:45Z

src/constructors.jl

+##########
+# Custom #
+##########
+struct Custom_alg<:SamplingAlgorithm end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

struct Custom_alg<:SamplingAlgorithm end

struct Custom_alg <: SamplingAlgorithm end

github-actions · 2023-06-09T11:00:45Z

src/constructors.jl

+    max_depth::Int=10,
+    Δ_max::Float64=1000.0,
+    ϵ::Float64=0.0)   


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

max_depth::Int=10,

Δ_max::Float64=1000.0,

ϵ::Float64=0.0)

max_depth::Int = 10,

Δ_max::Float64 = 1000.0,

ϵ::Float64 = 0.0,

)

github-actions · 2023-06-09T11:00:46Z

src/constructors.jl

+function HMCDA(
+    n_adapts::Int,
+    TAP::Float64,
+    λ::Float64;
+    ϵ::Float64=0.0) 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function HMCDA(

n_adapts::Int,

TAP::Float64,

λ::Float64;

ϵ::Float64=0.0)

function HMCDA(n_adapts::Int, TAP::Float64, λ::Float64; ϵ::Float64 = 0.0)

github-actions · 2023-06-09T11:00:47Z

src/constructors.jl

+    return StanHMCAdaptor(MassMatrixAdaptor(metric, integrator),
+                          StepSizeAdaptor(alg.TAP, integrator))
+ end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return StanHMCAdaptor(MassMatrixAdaptor(metric, integrator),

StepSizeAdaptor(alg.TAP, integrator))

end

return StanHMCAdaptor(

MassMatrixAdaptor(metric, integrator),

StepSizeAdaptor(alg.TAP, integrator),

)

end

github-actions · 2023-06-09T11:00:47Z

src/constructors.jl

+
+function make_kernel(alg::NUTS_alg, integrator)
+    return HMCKernel(Trajectory{MultinomialTS}(integrator, GeneralisedNoUTurn()))
+end    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-09T11:00:47Z

src/constructors.jl

+
+function make_kernel(alg::HMC_alg, integrator)
+    return HMCKernel(Trajectory{EndPointTS}(integrator, FixedNSteps(alg.n_leapfrog)))
+end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-09T11:00:47Z

src/constructors.jl

+
+function make_kernel(alg::HMCDA_alg, integrator)
+    return HMCKernel(Trajectory{EndPointTS}(integrator, FixedIntegrationTime(alg.λ)))
+end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-12T12:59:09Z

src/abstractmcmc.jl

+    return AbstractMCMC.step(
+        rng,
+        model,
+        spl,
+        state;
+        n_adapts=n_adapts,
+        kwargs...)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return AbstractMCMC.step(

rng,

model,

spl,

state;

n_adapts=n_adapts,

kwargs...)

return AbstractMCMC.step(rng, model, spl, state; n_adapts = n_adapts, kwargs...)

github-actions · 2023-06-12T12:59:09Z

src/abstractmcmc.jl

    state::HMCState;
-    nadapts::Int = 0,
+    nadapts::Int=0,


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

nadapts::Int=0,

nadapts::Int = 0,

github-actions · 2023-06-12T12:59:10Z

src/abstractmcmc.jl

@@ -217,14 +143,19 @@ function AbstractMCMC.step(
    h, κ, isadapted = adapt!(h, κ, adaptor, i, nadapts, t.z.θ, tstat.acceptance_rate)
    tstat = merge(tstat, (is_adapt = isadapted,))

+    # Convert variables back
+    vii_t = DynamicPPL.unflatten(vi_t, t.z.θ)  


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

vii_t = DynamicPPL.unflatten(vi_t, t.z.θ)

vii_t = DynamicPPL.unflatten(vi_t, t.z.θ)

yebai · 2023-06-13T10:04:00Z

src/abstractmcmc.jl

+    vi = DynamicPPL.VarInfo(model, ctxt)
+    vi_t = DynamicPPL.link!!(vi, model)
+    logdensityfunction = DynamicPPL.LogDensityFunction(vi_t, model, ctxt)
+    logdensityproblem = LogDensityProblemsAD.ADgradient(logdensityfunction)
+    logdensitymodel = AbstractMCMC.LogDensityModel(logdensityproblem)


Ideally, we should be independent of DynamicPPL APIs or internals from MCMC libraries. Maybe this should be transferred to DynamicPPL/AbstractMCMC.

github-actions · 2023-06-20T16:32:51Z

src/abstractmcmc.jl

+            adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),
+                StepSizeAdaptor(spl.alg.δ, integrator))


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),

StepSizeAdaptor(spl.alg.δ, integrator))

adaptor = StanHMCAdaptor(

MassMatrixAdaptor(metric),

StepSizeAdaptor(spl.alg.δ, integrator),

)

github-actions · 2023-06-20T16:32:51Z

src/abstractmcmc.jl

-    state = HMCState(0, t, h.metric, κ, adaptor)
-
+    state = HMCState(0, t, metric, κ, adaptor)
+    


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

github-actions · 2023-06-20T16:32:51Z

src/abstractmcmc.jl

@@ -302,4 +312,4 @@ function (cb::HMCProgressCallback)(rng, model, spl, t, state, i; nadapts = 0, kw
    elseif verbose && isadapted && i == nadapts
        @info "Finished $nadapts adapation steps" adaptor κ.τ.integrator metric
    end
-end
+end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-20T16:32:51Z

src/constructors.jl

+    n_adapts    ::  Int         # number of samples with adaption for ϵ
+    δ         ::  Float64     # target accept rate
+    λ           ::  Float64     # target leapfrog length
+    ϵ           ::  Float64     # (initial) step size


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

n_adapts :: Int # number of samples with adaption for ϵ

δ :: Float64 # target accept rate

λ :: Float64 # target leapfrog length

ϵ :: Float64 # (initial) step size

n_adapts::Int # number of samples with adaption for ϵ

δ::Float64 # target accept rate

λ::Float64 # target leapfrog length

ϵ::Float64 # (initial) step size

github-actions · 2023-06-20T16:32:51Z

src/constructors.jl

+function HMCDA(
+    n_adapts::Int,
+    δ::Float64,
+    λ::Float64;
+    ϵ::Float64=0.0) 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function HMCDA(

n_adapts::Int,

δ::Float64,

λ::Float64;

ϵ::Float64=0.0)

function HMCDA(n_adapts::Int, δ::Float64, λ::Float64; ϵ::Float64 = 0.0)

github-actions · 2023-06-20T16:32:52Z

src/constructors.jl

+    return StanHMCAdaptor(MassMatrixAdaptor(metric, integrator),
+                          StepSizeAdaptor(alg.δ, integrator))
+ end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return StanHMCAdaptor(MassMatrixAdaptor(metric, integrator),

StepSizeAdaptor(alg.δ, integrator))

end

return StanHMCAdaptor(

MassMatrixAdaptor(metric, integrator),

StepSizeAdaptor(alg.δ, integrator),

)

end

github-actions · 2023-06-20T16:44:14Z

src/constructors.jl

+Base.@kwdef struct HMCSampler{I,K,M,A} <: AbstractMCMC.AbstractSampler
+    alg::SamplingAlgorithm
+    "[`integrator`](@ref)."
+    integrator::I=nothing


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

integrator::I=nothing

integrator::I = nothing

github-actions · 2023-06-20T16:44:14Z

src/constructors.jl

+    "[`integrator`](@ref)."
+    integrator::I=nothing
+    "[`AbstractMCMCKernel`](@ref)."
+    kernel::K=nothing


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

kernel::K=nothing

kernel::K = nothing

github-actions · 2023-06-20T16:44:14Z

src/constructors.jl

+    "[`AbstractMCMCKernel`](@ref)."
+    kernel::K=nothing
+    "[`AbstractMetric`](@ref)."
+    metric::M=nothing


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

metric::M=nothing

metric::M = nothing

github-actions · 2023-06-20T16:44:14Z

src/constructors.jl

+    "[`AbstractMetric`](@ref)."
+    metric::M=nothing
+    "[`AbstractAdaptor`](@ref)."
+    adaptor::A=nothing


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

adaptor::A=nothing

adaptor::A = nothing

github-actions · 2023-06-21T09:12:59Z

src/constructors.jl

+function HMCDA(
+    n_adapts::Int,
+    δ::Float64,
+    λ::Float64;
+    ϵ::Float64=0.0)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function HMCDA(

n_adapts::Int,

δ::Float64,

λ::Float64;

ϵ::Float64=0.0)

function HMCDA(n_adapts::Int, δ::Float64, λ::Float64; ϵ::Float64 = 0.0)

github-actions · 2023-06-21T10:21:36Z

src/constructors.jl

+To access the updated fields use the resulting [`HMCState`](@ref).
+"""
+Base.@kwdef struct HMCSampler{I,K,M,A} <: AbstractMCMC.AbstractSampler
+    alg::HMCAlgorithm=Custom_alg


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

alg::HMCAlgorithm=Custom_alg

alg::HMCAlgorithm = Custom_alg

github-actions · 2023-06-21T10:21:37Z

src/constructors.jl

+##########
+# Custom #
+##########
+struct Custom_alg<:HMCAlgorithm end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

struct Custom_alg<:HMCAlgorithm end

struct Custom_alg <: HMCAlgorithm end

github-actions · 2023-06-21T10:21:37Z

src/constructors.jl

+    max_depth::Int=10,
+    Δ_max::Float64=1000.0,
+    ϵ::Float64=0.0)   
+    return HMCSampler(;alg=NUTS_alg(n_adapts, δ, max_depth, Δ_max, ϵ))
+end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

max_depth::Int=10,

Δ_max::Float64=1000.0,

ϵ::Float64=0.0)

return HMCSampler(;alg=NUTS_alg(n_adapts, δ, max_depth, Δ_max, ϵ))

end

max_depth::Int = 10,

Δ_max::Float64 = 1000.0,

ϵ::Float64 = 0.0,

)

return HMCSampler(; alg = NUTS_alg(n_adapts, δ, max_depth, Δ_max, ϵ))

end

github-actions · 2023-06-21T10:21:37Z

src/constructors.jl

+    ϵ::Float64,
+    n_leapfrog::Int)
+
+    return HMCSampler(;alg=HMC_alg(ϵ, n_leapfrog))


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

return HMCSampler(;alg=HMC_alg(ϵ, n_leapfrog))

return HMCSampler(; alg = HMC_alg(ϵ, n_leapfrog))

github-actions · 2023-06-21T10:21:37Z

src/constructors.jl

+function HMCDA(
+    n_adapts::Int,
+    δ::Float64,
+    λ::Float64;
+    ϵ::Float64=0.0)
+    return HMCSampler(;alg=HMCDA_alg(n_adapts, δ, λ, ϵ))


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function HMCDA(

n_adapts::Int,

δ::Float64,

λ::Float64;

ϵ::Float64=0.0)

return HMCSampler(;alg=HMCDA_alg(n_adapts, δ, λ, ϵ))

function HMCDA(n_adapts::Int, δ::Float64, λ::Float64; ϵ::Float64 = 0.0)

return HMCSampler(; alg = HMCDA_alg(n_adapts, δ, λ, ϵ))

github-actions · 2023-06-27T15:17:07Z

src/abstractmcmc.jl

+    d = d=LogDensityProblems.dimension(logdensity)
+    metric = make_metric(spl; d=d)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

d = d=LogDensityProblems.dimension(logdensity)

metric = make_metric(spl; d=d)

d = d = LogDensityProblems.dimension(logdensity)

metric = make_metric(spl; d = d)

github-actions · 2023-06-27T15:17:07Z

src/abstractmcmc.jl

+    integrator = make_integrator(spl;
+        rng=rng,
+        hamiltonian=hamiltonian,
+        init_params=init_params)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

integrator = make_integrator(spl;

rng=rng,

hamiltonian=hamiltonian,

init_params=init_params)

integrator = make_integrator(

spl;

rng = rng,

hamiltonian = hamiltonian,

init_params = init_params,

)

github-actions · 2023-06-27T15:17:07Z

src/constructors.jl

@@ -0,0 +1,204 @@
+abstract type AbstractHMCSampler <:AbstractMCMC.AbstractSampler end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

abstract type AbstractHMCSampler <:AbstractMCMC.AbstractSampler end

abstract type AbstractHMCSampler <: AbstractMCMC.AbstractSampler end

github-actions · 2023-06-27T15:17:07Z

src/constructors.jl

+"""
+Base.@kwdef struct CustomHMC{I,K,M,A} <: AbstractMCMC.AbstractSampler
+    "[`integrator`](@ref)."
+    integrator::I=Leapfrog


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

integrator::I=Leapfrog

integrator::I = Leapfrog

github-actions · 2023-06-27T15:17:08Z

src/constructors.jl

+    max_depth::Int=10                # maximum tree depth
+    Δ_max::Float64=1000.0            # maximum error
+    init_ϵ::Float64=0.0              # (initial) step size
+    integrator_method=Leapfrog       # integrator method
+    metric_type=DiagEuclideanMetric  # metric type


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

max_depth::Int=10 # maximum tree depth

Δ_max::Float64=1000.0 # maximum error

init_ϵ::Float64=0.0 # (initial) step size

integrator_method=Leapfrog # integrator method

metric_type=DiagEuclideanMetric # metric type

max_depth::Int = 10 # maximum tree depth

Δ_max::Float64 = 1000.0 # maximum error

init_ϵ::Float64 = 0.0 # (initial) step size

integrator_method = Leapfrog # integrator method

metric_type = DiagEuclideanMetric # metric type

github-actions · 2023-06-27T15:17:08Z

src/constructors.jl

+function make_adaptor(spl::Union{NUTS_alg, HMCDA_alg}, metric, integrator)
+    adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),
+                             StepSizeAdaptor(spl.δ, integrator))


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function make_adaptor(spl::Union{NUTS_alg, HMCDA_alg}, metric, integrator)

adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),

StepSizeAdaptor(spl.δ, integrator))

function make_adaptor(spl::Union{NUTS_alg,HMCDA_alg}, metric, integrator)

adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(spl.δ, integrator))

github-actions · 2023-06-27T15:17:09Z

src/constructors.jl

+                             StepSizeAdaptor(spl.δ, integrator))
+    n_adapts = spl.n_adapts
+    return n_adapts, adaptor
+ end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-27T15:17:09Z

src/constructors.jl

+
+function make_adaptor(spl::HMC_alg, metric, integrator)
+    return 0, NoAdaptation()
+ end 


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-27T15:17:09Z

src/constructors.jl

+    return 0, NoAdaptation()
+ end 
+
+ function make_adaptor(spl::CustomHMC, metric, integrator)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function make_adaptor(spl::CustomHMC, metric, integrator)

function make_adaptor(spl::CustomHMC, metric, integrator)

github-actions · 2023-06-27T15:17:10Z

src/constructors.jl

+
+ function make_adaptor(spl::CustomHMC, metric, integrator)
+    return spl.n_adapts, spl.adaptor
+ end


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

end

end

github-actions · 2023-06-27T15:41:06Z

src/constructors.jl

+function make_integrator(rng, spl::Union{HMC_alg, NUTS_alg, HMCDA_alg},
+    hamiltonian, init_params)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function make_integrator(rng, spl::Union{HMC_alg, NUTS_alg, HMCDA_alg},

hamiltonian, init_params)

function make_integrator(

rng,

spl::Union{HMC_alg,NUTS_alg,HMCDA_alg},

hamiltonian,

init_params,

)

github-actions · 2023-06-27T15:41:06Z

src/constructors.jl

+
+#########
+
+function make_metric(spl::Union{HMC_alg, NUTS_alg, HMCDA_alg}, logdensity)


[JuliaFormatter] _{reported by reviewdog 🐶}

Suggested change

function make_metric(spl::Union{HMC_alg, NUTS_alg, HMCDA_alg}, logdensity)

function make_metric(spl::Union{HMC_alg,NUTS_alg,HMCDA_alg}, logdensity)

JaimeRZP added 2 commits March 9, 2023 10:38

Lab

43cfe65

first draft

3158bb6

JaimeRZP linked an issue Mar 9, 2023 that may be closed by this pull request

More friendly default sample interface #313

Closed

JaimeRZP added 4 commits March 9, 2023 13:54

cleaning Lab a lil

37d6831

Coded interface in

6fe6436

Coded interface in

304d401

Merge branch 'master' into no_glue_code

46f0803

github-actions bot reviewed Mar 10, 2023

View reviewed changes

JaimeRZP mentioned this pull request May 30, 2023

No glue code TuringLang/Turing.jl#1997

Closed

5 tasks

JaimeRZP added 2 commits May 30, 2023 18:04

working on no glue code from the other end

2f6f2c1

trying stuff

dfd5e74

github-actions bot reviewed May 30, 2023

View reviewed changes

JaimeRZP added 3 commits May 31, 2023 13:11

no glue code abstract mcmc

3038441

Merge branch 'master' into no_glue_code

deb0555

AbstractMCMC working

00b837a

github-actions bot reviewed Jun 1, 2023

View reviewed changes

JaimeRZP added 2 commits June 2, 2023 17:03

constructors + moving stuff to init_step

ce96cac

Scaling things

1f8c5a7

github-actions bot reviewed Jun 5, 2023

View reviewed changes

axing hmc

612e10b

github-actions bot reviewed Jun 5, 2023

View reviewed changes

HMC + HMCDA

3bbc668

github-actions bot reviewed Jun 6, 2023

View reviewed changes

return sampler to master

1bffe99

github-actions bot reviewed Jun 7, 2023

View reviewed changes

getmodel

b941529

github-actions bot reviewed Jun 8, 2023

View reviewed changes

small step forward

8b1f962

github-actions bot reviewed Jun 8, 2023

View reviewed changes

big step forward

0f45cc8

github-actions bot reviewed Jun 8, 2023

View reviewed changes

JaimeRZP added 2 commits June 8, 2023 14:45

huge step forward

3e1c403

fixed constructors

8fa9fcb

github-actions bot reviewed Jun 8, 2023

View reviewed changes

constructors reworked

c582abf

github-actions bot reviewed Jun 9, 2023

View reviewed changes

linking

39941fd

github-actions bot reviewed Jun 12, 2023

View reviewed changes

yebai reviewed Jun 13, 2023

View reviewed changes

constructors for tors PR

3684a1e

github-actions bot reviewed Jun 20, 2023

View reviewed changes

constructors for tors PR

62c2096

github-actions bot reviewed Jun 20, 2023

View reviewed changes

no sampling needed anymore

1893e3f

github-actions bot reviewed Jun 21, 2023

View reviewed changes

no Dynamic PPL

1af622c

github-actions bot reviewed Jun 21, 2023

View reviewed changes

convinience constructors consensus

acaa289

github-actions bot reviewed Jun 27, 2023

View reviewed changes

kwargs --> args

80f0d8d

github-actions bot reviewed Jun 27, 2023

View reviewed changes

JaimeRZP closed this Jun 27, 2023

JaimeRZP deleted the no_glue_code branch June 27, 2023 15:44

		function sample(model::DynamicPPL.Model, ϵ::Number, TAP::Number, n_samples::Int, n_adapts::Int;
		initial_θ=initial_θ, progress=true, kwargs...)

	metric = DiagEuclideanMetric(d)
	metric = DiagEuclideanMetric(d)

	proposal = NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)
	proposal = NUTS{MultinomialTS,GeneralisedNoUTurn}(integrator)

		return sample(model, metric, proposal, initial_θ, n_samples, adaptor, n_adapts;
		progress=progress, kwargs...)

	name = [DynamicPPL.VarName(Symbol(vsym, i,)) for i in 1:dist_length]
	name = [DynamicPPL.VarName(Symbol(vsym, i)) for i = 1:dist_length]

		kernel = AdvancedHMC.NUTS{MultinomialTS, GeneralisedNoUTurn}(integrator)
		adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric), StepSizeAdaptor(settings.TAP, integrator))

-    metric
-    integrator
-    kernel
-    adaptor
+    metric::Any
+    integrator::Any
+    kernel::Any
+    adaptor::Any

		adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),
		StepSizeAdaptor(spl.alg.TAP, integrator))

	HMCSampler(integrator, kernel, metric, adaptor) = HMCSampler(Custom_alg, integrator, kernel, metric, adaptor)
	HMCSampler(integrator, kernel, metric, adaptor) =
	HMCSampler(Custom_alg, integrator, kernel, metric, adaptor)

	struct Custom_alg<:SamplingAlgorithm end
	struct Custom_alg <: SamplingAlgorithm end

	vii_t = DynamicPPL.unflatten(vi_t, t.z.θ)
	vii_t = DynamicPPL.unflatten(vi_t, t.z.θ)

		adaptor = StanHMCAdaptor(MassMatrixAdaptor(metric),
		StepSizeAdaptor(spl.alg.δ, integrator))

		state = HMCState(0, t, h.metric, κ, adaptor)

		state = HMCState(0, t, metric, κ, adaptor)

	struct Custom_alg<:HMCAlgorithm end
	struct Custom_alg <: HMCAlgorithm end

	return HMCSampler(;alg=HMC_alg(ϵ, n_leapfrog))
	return HMCSampler(; alg = HMC_alg(ϵ, n_leapfrog))

		d = d=LogDensityProblems.dimension(logdensity)
		metric = make_metric(spl; d=d)

		@@ -0,0 +1,204 @@
		abstract type AbstractHMCSampler <:AbstractMCMC.AbstractSampler end

	abstract type AbstractHMCSampler <:AbstractMCMC.AbstractSampler end
	abstract type AbstractHMCSampler <: AbstractMCMC.AbstractSampler end

	function make_adaptor(spl::CustomHMC, metric, integrator)
	function make_adaptor(spl::CustomHMC, metric, integrator)

		function make_integrator(rng, spl::Union{HMC_alg, NUTS_alg, HMCDA_alg},
		hamiltonian, init_params)

No glue code #319

No glue code #319

Conversation

JaimeRZP commented Mar 9, 2023 • edited Loading

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

github-actions bot Mar 10, 2023

Choose a reason for hiding this comment

JaimeRZP commented Mar 10, 2023

github-actions bot May 30, 2023

Choose a reason for hiding this comment

github-actions bot May 30, 2023

Choose a reason for hiding this comment

github-actions bot Jun 1, 2023

Choose a reason for hiding this comment

github-actions bot Jun 1, 2023

Choose a reason for hiding this comment

github-actions bot Jun 1, 2023

Choose a reason for hiding this comment

github-actions bot Jun 1, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 5, 2023

Choose a reason for hiding this comment

github-actions bot Jun 6, 2023

Choose a reason for hiding this comment

github-actions bot Jun 6, 2023

Choose a reason for hiding this comment

github-actions bot Jun 6, 2023

Choose a reason for hiding this comment

github-actions bot Jun 7, 2023

Choose a reason for hiding this comment

github-actions bot Jun 8, 2023

JaimeRZP commented Mar 9, 2023 •

edited

Loading


		#########

		function make_metric(spl::Union{HMC_alg, NUTS_alg, HMCDA_alg}, logdensity)