Version 08 compat

.travis.yml

@@ -4,8 +4,8 @@ os:
   - linux
 # - osx
 julia:
-  - 0.7
   - 1.0
+  - 1
 notifications:
   email: false
 after_success:

Project.toml

@@ -0,0 +1,31 @@
+name = "ParticleFilters"
+uuid = "c8b314e2-9260-5cf8-ae76-3be7461ca6d0"
+repo = "https://github.com/JuliaPOMDP/ParticleFilters.jl"
+version = "0.4.0"
+
+[deps]
+POMDPModelTools = "08074719-1b2a-587c-a292-00f91cc44415"
+POMDPPolicies = "182e52fb-cfd0-5e46-8c26-fd0667c990f4"
+POMDPs = "a93abf59-7444-517b-a68a-c42f96afdd7d"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
+StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
+
+[compat]
+POMDPs = "0.7.3, 0.8"
+julia = "1"
+
+[extras]
+DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
+Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
+LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+NBInclude = "0db19996-df87-5ea3-a455-e3a50d440464"
+POMDPModels = "355abbd5-f08e-5560-ac9e-8b5f2592a0ca"
+POMDPSimulators = "e0d0a172-29c6-5d4e-96d0-f262df5d01fd"
+Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
+Reel = "71555da5-176e-5e73-a222-aebc6c6e4f2f"
+StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+
+[targets]
+test = ["DelimitedFiles", "Distributions", "LinearAlgebra", "NBInclude", "Plots", "POMDPModels", "POMDPSimulators", "Reel", "StaticArrays", "Test"]

docs/src/beliefs.md

@@ -20,7 +20,6 @@ The following functions from the [POMDPs.jl distributions interface](http://juli
 - [`rand`](@ref)
 - [`pdf`](@ref)
 - [`support`](@ref)
-- [`sampletype`](@ref) (will be replaced with `Random.gentype`)
 - [`mode`](@ref)
 - [`mean`](@ref)
 
@@ -43,7 +42,6 @@ These functions provide access to the particles and weights in the beliefs (clic
 POMDPs.rand
 POMDPs.pdf
 POMDPs.support
-POMDPs.sampletype
 POMDPs.mode
 POMDPs.mean
 n_particles

src/ParticleFilters.jl

@@ -3,10 +3,8 @@ module ParticleFilters
 using POMDPs
 import POMDPs: pdf, mode, update, initialize_belief, support
 import POMDPs: statetype, isterminal, observation
-import POMDPs: generate_s
 import POMDPs: action, value
 import POMDPs: implemented
-import POMDPs: sampletype
 
 import POMDPModelTools: obs_weight
 using StatsBase
@@ -15,7 +13,7 @@ using Statistics
 using POMDPPolicies
 using POMDPModelTools # for weighted_iterator
 
-import Random: rand
+import Random: rand, gentype
 import Statistics: mean
 
 export

src/beliefs.jl

@@ -1,6 +1,5 @@
 abstract type AbstractParticleBelief{T} end
 
-sampletype(::Type{B}) where B<:AbstractParticleBelief{T} where T = T
 Random.gentype(::Type{B}) where B<:AbstractParticleBelief{T} where T = T
 
 ########################

src/policies.jl

@@ -1,26 +1,26 @@
-#=
-Implementation specific to using ParticleBeliefs with AlphaVectorPolicy from POMDPToolbox
-these are more efficient than converting the ParticleBelief to a DiscreteBelief 
-=#
-
-"""
-Given a particle belief, return the unnormalized utility function that weights each state value by the weight of 
-the corresponding particle 
-    unnormalized_util(p::AlphaVectorPolicy, b::AbstractParticleBelief)
-"""
-function unnormalized_util(p::AlphaVectorPolicy, b::AbstractParticleBelief)
-    util = zeros(n_actions(p.pomdp))
-    for (i, s) in enumerate(particles(b))
-        j = stateindex(p.pomdp, s)
-        util += weight(b, i)*getindex.(p.alphas, (j,))
-    end
-    return util
-end
-
-function action(p::AlphaVectorPolicy, b::AbstractParticleBelief)
-    util = unnormalized_util(p, b)
-    ihi = argmax(util)
-    return p.action_map[ihi]
-end
-
-value(p::AlphaVectorPolicy, b::AbstractParticleBelief) = maximum(unnormalized_util(p, b))/weight_sum(b)
+#=
+Implementation specific to using ParticleBeliefs with AlphaVectorPolicy from POMDPToolbox
+these are more efficient than converting the ParticleBelief to a DiscreteBelief
+=#
+
+"""
+Given a particle belief, return the unnormalized utility function that weights each state value by the weight of
+the corresponding particle
+    unnormalized_util(p::AlphaVectorPolicy, b::AbstractParticleBelief)
+"""
+function unnormalized_util(p::AlphaVectorPolicy, b::AbstractParticleBelief)
+    util = p.pomdp |> actions |> length |> zeros
+    for (i, s) in enumerate(particles(b))
+        j = stateindex(p.pomdp, s)
+        util += weight(b, i)*getindex.(p.alphas, (j,))
+    end
+    return util
+end
+
+function action(p::AlphaVectorPolicy, b::AbstractParticleBelief)
+    util = unnormalized_util(p, b)
+    ihi = argmax(util)
+    return p.action_map[ihi]
+end
+
+value(p::AlphaVectorPolicy, b::AbstractParticleBelief) = maximum(unnormalized_util(p, b))/weight_sum(b)

src/pomdps.jl

@@ -18,7 +18,7 @@ function predict!(pm, m::POMDP, b, a, rng)
         s = particle(b, i)
         if !isterminal(m, s)
             all_terminal = false
-            sp = generate_s(m, s, a, rng)
+            sp = gen(DDNOut(:sp), m, s, a, rng)
             pm[i] = sp
         end
     end
@@ -68,7 +68,7 @@ function initialize_belief(up::BasicParticleFilter, d::D) where D
         if ex isa MethodError
             @warn("""
                 Suppressing MethodError in initialize_belief in ParticleFilters.jl. Please file an issue here:
-                
+
                 https://github.com/JuliaPOMDP/ParticleFilters.jl/issues/new
 
                 The error was

src/unweighted.jl

@@ -14,11 +14,11 @@ function UnweightedParticleFilter(model, n::Integer; rng=Base.GLOBAL_RNG)
 end
 
 function update(up::UnweightedParticleFilter, b::ParticleCollection, a, o)
-    new = sampletype(b)[]
+    new = Random.gentype(b)[]
     i = 1
     while i <= up.n
         s = particle(b, mod1(i, n_particles(b)))
-        sp, o_gen = generate_so(up.model, s, a, up.rng)
+        sp, o_gen = gen(DDNOut(:sp, :o), up.model, s, a, up.rng)
         if o_gen == o
             push!(new, sp)
         end

test/domain_specific_resampler.jl

@@ -36,5 +36,5 @@ for bp in stepthrough(m, p2, up, "bp", max_steps=3)
     @test bp isa ParticleCollection{LightDark1DState}
     @test n_particles(bp) == n
 end
-    
+
 end

test/runtests.jl

@@ -58,13 +58,13 @@ struct ContinuousPOMDP <: POMDP{Float64, Float64, Float64} end
         rb2 = @inferred resample(rs, WeightedParticleBelief(particles(b), ones(n_particles(b))), MersenneTwister(3))
         @test all(particles(rb1).==particles(rb2))
     end
-    
+
     @testset "unweighted" begin
         rng = MersenneTwister(47)
         uf = UnweightedParticleFilter(p, 1000, rng)
         ps = @inferred initialize_belief(uf, initialstate_distribution(p))
         a = @inferred rand(rng, actions(p))
-        sp, o = @inferred generate_so(p, rand(rng, ps), a, rng)
+        sp, o = @inferred gen(DDNOut(:sp, :o), p, rand(rng, ps), a, rng)
         bp = @inferred update(uf, ps, a, o)
 
         wp1 = @inferred collect(weighted_particles(ParticleCollection([1,2])))