Add logratio_proposal_density and remove is_symmetric_proposal (#54)

* Add `logratio_proposal_density` and remove `is_symmetric_proposal`

* Bump version

* Add `issymmetric` for `StaticProposal` and add aliases

* Fix type inference problems

* Remove accidentally included code

Author: devmotion

Project.toml

@@ -1,6 +1,6 @@
 name = "AdvancedMH"
 uuid = "5b7e9947-ddc0-4b3f-9b55-0d8042f74170"
-version = "0.5.9"
+version = "0.6.0"
 
 [deps]
 AbstractMCMC = "80f14c24-f653-4e6a-9b94-39d6b0f70001"

src/AdvancedMH.jl

@@ -8,8 +8,18 @@ using Requires
 import Random
 
 # Exports
-export MetropolisHastings, DensityModel, RWMH, StaticMH, StaticProposal, 
-    RandomWalkProposal, Ensemble, StretchProposal, MALA
+export
+    MetropolisHastings,
+    DensityModel,
+    RWMH,
+    StaticMH,
+    StaticProposal,
+    SymmetricStaticProposal,
+    RandomWalkProposal,
+    SymmetricRandomWalkProposal,
+    Ensemble,
+    StretchProposal,
+    MALA
 
 # Reexports
 export sample, MCMCThreads, MCMCDistributed

src/MALA.jl

@@ -45,6 +45,11 @@ function q(
     return q(spl.proposal(-t_cond.gradient), t.params, t_cond.params)
 end
 
+function logratio_proposal_density(
+    sampler::MALA{<:Proposal}, state::GradientTransition, candidate::GradientTransition
+)
+    return q(sampler, state, candidate) - q(sampler, candidate, state)
+end
 
 """
     logdensity_and_gradient(model::DensityModel, params)

src/mh-core.jl

@@ -138,37 +138,6 @@ end
     return expr
 end
 
-# Evaluate the likelihood of t conditional on t_cond.
-function q(
-    spl::MetropolisHastings{<:AbstractArray},
-    t::Transition,
-    t_cond::Transition
-)
-    # mapreduce with multiple iterators requires Julia 1.2 or later
-    return mapreduce(+, 1:length(spl.proposal)) do i
-        q(spl.proposal[i], t.params[i], t_cond.params[i])
-    end
-end
-
-function q(
-    spl::MetropolisHastings{<:Proposal},
-    t::Transition,
-    t_cond::Transition
-)
-    return q(spl.proposal, t.params, t_cond.params)
-end
-
-function q(
-    spl::MetropolisHastings{<:NamedTuple},
-    t::Transition,
-    t_cond::Transition
-)
-    # mapreduce with multiple iterators requires Julia 1.2 or later
-    return mapreduce(+, keys(t.params)) do k
-        q(spl.proposal[k], t.params[k], t_cond.params[k])
-    end
-end
-
 transition(sampler, model, params) = transition(model, params)
 transition(model, params) = Transition(model, params)
 
@@ -191,31 +160,6 @@ function AbstractMCMC.step(
     return transition, transition
 end
 
-"""
-    is_symmetric_proposal(proposal)::Bool
-
-Implementing this for a custom proposal will allow `AbstractMCMC.step` to avoid
-computing the "Hastings" part of the Metropolis-Hasting log acceptance
-probability (if the proposal is indeed symmetric). By default,
-`is_symmetric_proposal(proposal)` returns `false`.  The user is responsible for
-determining whether a custom proposal distribution is indeed symmetric.  As
-noted in `MetropolisHastings`, `proposal` is a `Proposal`, `NamedTuple` of
-`Proposal`, or `Array{Proposal}` in the shape of your data.
-"""
-is_symmetric_proposal(proposal) = false
-
-# The following univariate random walk proposals are symmetric.
-is_symmetric_proposal(::RandomWalkProposal{<:Normal}) = true
-is_symmetric_proposal(::RandomWalkProposal{<:MvNormal}) = true
-is_symmetric_proposal(::RandomWalkProposal{<:TDist}) = true
-is_symmetric_proposal(::RandomWalkProposal{<:Cauchy}) = true
-
-# The following multivariate random walk proposals are symmetric.
-is_symmetric_proposal(::RandomWalkProposal{<:AbstractArray{<:Normal}}) = true
-is_symmetric_proposal(::RandomWalkProposal{<:AbstractArray{<:MvNormal}}) = true
-is_symmetric_proposal(::RandomWalkProposal{<:AbstractArray{<:TDist}}) = true
-is_symmetric_proposal(::RandomWalkProposal{<:AbstractArray{<:Cauchy}}) = true
-
 # Define the other sampling steps.
 # Return a 2-tuple consisting of the next sample and the the next state.
 # In this case they are identical, and either a new proposal (if accepted)
@@ -231,12 +175,8 @@ function AbstractMCMC.step(
     params = propose(rng, spl, model, params_prev)
 
     # Calculate the log acceptance probability.
-    logα = logdensity(model, params) - logdensity(model, params_prev)
-
-    # Compute Hastings portion of ratio if proposal is not symmetric.
-    if !is_symmetric_proposal(spl.proposal)
-      logα += q(spl, params_prev, params) - q(spl, params, params_prev)
-    end
+    logα = logdensity(model, params) - logdensity(model, params_prev) +
+        logratio_proposal_density(spl, params_prev, params)
 
     # Decide whether to return the previous params or the new one.
     if -Random.randexp(rng) < logα
@@ -245,3 +185,9 @@ function AbstractMCMC.step(
         return params_prev, params_prev
     end
 end
+
+function logratio_proposal_density(
+    sampler::MetropolisHastings, params_prev::Transition, params::Transition
+)
+    return logratio_proposal_density(sampler.proposal, params_prev.params, params.params)
+end

src/proposal.jl

@@ -1,12 +1,24 @@
 abstract type Proposal{P} end
 
-struct StaticProposal{P} <: Proposal{P}
+struct StaticProposal{issymmetric,P} <: Proposal{P}
     proposal::P
 end
+const SymmetricStaticProposal{P} = StaticProposal{true,P}
 
-struct RandomWalkProposal{P} <: Proposal{P}
+StaticProposal(proposal) = StaticProposal{false}(proposal)
+function StaticProposal{issymmetric}(proposal) where {issymmetric}
+    return StaticProposal{issymmetric,typeof(proposal)}(proposal)
+end
+
+struct RandomWalkProposal{issymmetric,P} <: Proposal{P}
     proposal::P
 end
+const SymmetricRandomWalkProposal{P} = RandomWalkProposal{true,P}
+
+RandomWalkProposal(proposal) = RandomWalkProposal{false}(proposal)
+function RandomWalkProposal{issymmetric}(proposal) where {issymmetric}
+    return RandomWalkProposal{issymmetric,typeof(proposal)}(proposal)
+end
 
 # Random draws
 Base.rand(p::Proposal, args...) = rand(Random.GLOBAL_RNG, p, args...)
@@ -26,24 +38,28 @@ end
 # Random Walk #
 ###############
 
-function propose(rng::Random.AbstractRNG, p::RandomWalkProposal, m::DensityModel)
-    return propose(rng, StaticProposal(p.proposal), m)
+function propose(
+    rng::Random.AbstractRNG,
+    proposal::RandomWalkProposal{issymmetric,<:Union{Distribution,AbstractArray}},
+    ::DensityModel
+) where {issymmetric}
+    return rand(rng, proposal)
 end
 
 function propose(
     rng::Random.AbstractRNG,
-    proposal::RandomWalkProposal{<:Union{Distribution,AbstractArray}}, 
-    model::DensityModel, 
+    proposal::RandomWalkProposal{issymmetric,<:Union{Distribution,AbstractArray}},
+    model::DensityModel,
     t
-)
+) where {issymmetric}
     return t + rand(rng, proposal)
 end
 
 function q(
-    proposal::RandomWalkProposal{<:Union{Distribution,AbstractArray}}, 
+    proposal::RandomWalkProposal{issymmetric,<:Union{Distribution,AbstractArray}},
     t,
     t_cond
-)
+) where {issymmetric}
     return logpdf(proposal, t - t_cond)
 end
 
@@ -53,18 +69,18 @@ end
 
 function propose(
     rng::Random.AbstractRNG,
-    proposal::StaticProposal{<:Union{Distribution,AbstractArray}},
+    proposal::StaticProposal{issymmetric,<:Union{Distribution,AbstractArray}},
     model::DensityModel,
     t=nothing
-)
+) where {issymmetric}
     return rand(rng, proposal)
 end
 
 function q(
-    proposal::StaticProposal{<:Union{Distribution,AbstractArray}},
+    proposal::StaticProposal{issymmetric,<:Union{Distribution,AbstractArray}},
     t,
     t_cond
-)
+) where {issymmetric}
     return logpdf(proposal, t)
 end
 
@@ -73,10 +89,14 @@ end
 ############
 
 # function definition with abstract types requires Julia 1.3 or later
-for T in (StaticProposal, RandomWalkProposal)
+for T in (:StaticProposal, :RandomWalkProposal)
     @eval begin
-        (p::$T{<:Function})() = $T(p.proposal())
-        (p::$T{<:Function})(t) = $T(p.proposal(t))
+        function (p::$T{issymmetric,<:Function})() where {issymmetric}
+            return $T{issymmetric}(p.proposal())
+        end
+        function (p::$T{issymmetric,<:Function})(t) where {issymmetric}
+            return $T{issymmetric}(p.proposal(t))
+        end
     end
 end
 
@@ -103,4 +123,69 @@ function q(
     t_cond
 )
     return q(proposal(t_cond), t, t_cond)
-end
+end
+
+"""
+    logratio_proposal_density(proposal, state, candidate)
+
+Compute the log-ratio of the proposal densities in the Metropolis-Hastings algorithm.
+
+The log-ratio of the proposal densities is defined as
+```math
+\\log \\frac{g(x | x')}{g(x' | x)},
+```
+where ``x`` is the current state, ``x'`` is the proposed candidate for the next state,
+and ``g(y' | y)`` is the conditional probability of proposing state ``y'`` given state
+``y`` (proposal density).
+"""
+function logratio_proposal_density(proposal::Proposal, state, candidate)
+    return q(proposal, state, candidate) - q(proposal, candidate, state)
+end
+
+# ratio is always 0 for symmetric proposals
+logratio_proposal_density(::RandomWalkProposal{true}, state, candidate) = 0
+logratio_proposal_density(::StaticProposal{true}, state, candidate) = 0
+
+# type stable implementation for `NamedTuple`s
+function logratio_proposal_density(
+    proposals::NamedTuple{names}, states::NamedTuple, candidates::NamedTuple
+) where {names}
+    if @generated
+        args = map(names) do name
+            :(logratio_proposal_density(
+                proposals[$(QuoteNode(name))],
+                states[$(QuoteNode(name))],
+                candidates[$(QuoteNode(name))],
+            ))
+        end
+        return :(+($(args...)))
+    else
+        return sum(names) do name
+            return logratio_proposal_density(
+                proposals[name], states[name], candidates[name]
+            )
+        end
+    end
+end
+
+# use recursion for `Tuple`s to ensure type stability
+logratio_proposal_density(proposals::Tuple{}, states::Tuple, candidates::Tuple) = 0
+function logratio_proposal_density(
+    proposals::Tuple{<:Proposal}, states::Tuple, candidates::Tuple
+)
+    return logratio_proposal_density(first(proposals), first(states), first(candidates))
+end
+function logratio_proposal_density(proposals::Tuple, states::Tuple, candidates::Tuple)
+    valfirst = logratio_proposal_density(first(proposals), first(states), first(candidates))
+    valtail = logratio_proposal_density(
+        Base.tail(proposals), Base.tail(states), Base.tail(candidates)
+    )
+    return valfirst + valtail
+end
+
+# fallback for general iterators (arrays etc.) - possibly not type stable!
+function logratio_proposal_density(proposals, states, candidates)
+    return sum(zip(proposals, states, candidates)) do (proposal, state, candidate)
+        return logratio_proposal_density(proposal, state, candidate)
+    end
+end