POMDPs v0.8 Compatibility

Project.toml

@@ -14,7 +14,7 @@ UnicodePlots = "b8865327-cd53-5732-bb35-84acbb429228"
 
 [compat]
 Distributions = ">= 0.17"
-POMDPs = "0.7.3, 0.9.0"
+POMDPs = "0.7.3, 0.8.0"
 julia = "1"
 
 [extras]

src/POMDPModelTools.jl

@@ -22,11 +22,12 @@ include("visualization.jl")
 
 # info interface
 export
-    generate_sri,
-    generate_sori,
+    add_infonode,
     action_info,
     solve_info,
-    update_info
+    update_info,
+    generate_sri,
+    generate_sori
 include("info.jl")
 
 export

src/info.jl

@@ -1,24 +1,6 @@
 # functions for passing out info from simulations, similar to the info return from openai gym
 # maintained by @zsunberg
 
-"""
-Return a tuple containing the next state and reward and information (usually a `NamedTuple`, `Dict` or `nothing`) from that step.
-
-By default, returns `nothing` as info.
-"""
-function generate_sri(p::MDP, s, a, rng::AbstractRNG)
-    return generate_sr(p, s, a, rng)..., nothing
-end
-
-"""
-Return a tuple containing the next state, observation, and reward and information (usually a `NamedTuple`, `Dict` or `nothing`) from that step.
-
-By default, returns `nothing` as info. 
-"""
-function generate_sori(p::POMDP, s, a, rng::AbstractRNG)
-    return generate_sor(p, s, a, rng)..., nothing
-end
-
 """
     a, ai = action_info(policy, x)
 
@@ -51,3 +33,49 @@ By default, returns `nothing` as info.
 function update_info(up::Updater, b, a, o)
     return update(up, b, a, o), nothing
 end
+
+"""
+    add_infonode(ddn::DDNStructure)
+
+Create a new DDNStructure object with a new node labeled :info with parents :s and :a
+"""
+function add_infonode(ddn) # for DDNStructure, but it is not declared in v0.7.3, so there is not annotation
+    add_node(ddn, :info, ConstantDDNNode(nothing), (:s, :a))
+end
+
+function add_infonode(ddn::POMDPs.DDNStructureV7{nodenames}) where nodenames
+    return POMDPs.DDNStructureV7{(nodenames..., :info)}()
+end
+
+###############################################################
+# Note all generate functions will be deprecated in POMDPs v0.8
+###############################################################
+
+
+if DDNStructure(MDP) isa POMDPs.DDNStructureV7
+    """
+    Return a tuple containing the next state and reward and information (usually a `NamedTuple`, `Dict` or `nothing`) from that step.
+
+    By default, returns `nothing` as info.
+    """
+    function generate_sri(p::MDP, s, a, rng::AbstractRNG)
+        return generate_sr(p, s, a, rng)..., nothing
+    end
+
+    """
+    Return a tuple containing the next state, observation, and reward and information (usually a `NamedTuple`, `Dict` or `nothing`) from that step.
+
+    By default, returns `nothing` as info. 
+    """
+    function generate_sori(p::POMDP, s, a, rng::AbstractRNG)
+        return generate_sor(p, s, a, rng)..., nothing
+    end
+
+    POMDPs.gen(::DDNOut{(:sp,:o,:r,:i)}, m, s, a, rng) = generate_sori(m, s, a, rng)
+    POMDPs.gen(::DDNOut{(:sp,:o,:r,:info)}, m, s, a, rng) = generate_sori(m, s, a, rng)
+    POMDPs.gen(::DDNOut{(:sp,:r,:i)}, m, s, a, rng) = generate_sri(m, s, a, rng)
+    POMDPs.gen(::DDNOut{(:sp,:r,:info)}, m, s, a, rng) = generate_sri(m, s, a, rng)
+else
+    @deprecate generate_sri(args...) gen(DDNOut(:sp,:r,:info), args...)
+    @deprecate generate_sori(args...) gen(DDNOut(:sp,:o,:r,:info), args...)
+end

test/runtests.jl

@@ -73,4 +73,4 @@ using SparseArrays
         include("test_tabular.jl")
     end
 
-end
+end

test/test_fully_observable_pomdp.jl

@@ -1,5 +1,5 @@
 let
-    mdp = GridWorld()
+    mdp = SimpleGridWorld()
 
     pomdp = FullyObservablePOMDP(mdp)
 

test/test_info.jl

@@ -23,14 +23,20 @@ let
     rng = MersenneTwister(7)
 
     mdp = LegacyGridWorld()
+    POMDPs.DDNStructure(::Type{typeof(mdp)}) = DDNStructure(MDP) |> add_infonode
+    @test :info in nodenames(DDNStructure(mdp))
     s = initialstate(mdp, rng)
     a = rand(rng, actions(mdp))
-    @inferred generate_sri(mdp, s, a, rng)
+    sp, r, i = @inferred gen(DDNOut(:sp,:r,:info), mdp, s, a, rng)
+    @test i === nothing
 
     pomdp = TigerPOMDP()
+    POMDPs.DDNStructure(::Type{typeof(pomdp)}) = DDNStructure(POMDP) |> add_infonode
+    @test :info in nodenames(DDNStructure(pomdp))
     s = initialstate(pomdp, rng)
     a = rand(rng, actions(pomdp))
-    @inferred generate_sori(pomdp, s, a, rng)
+    sp, o, r, i = @inferred gen(DDNOut(:sp,:o,:r,:info), pomdp, s, a, rng)
+    @test i === nothing
 
     up = VoidUpdater()
     policy = RandomPolicy(rng, pomdp)
@@ -43,6 +49,6 @@ let
     d = initialstate_distribution(pomdp)
     b = initialize_belief(up, d)
     a = action(policy, b)
-    sp, o = generate_so(pomdp, rand(rng, d), a, rng)
+    sp, o, r = gen(DDNOut(:sp,:o,:r), pomdp, rand(rng, d), a, rng)
     @inferred update_info(up, b, a, o)
 end