Merge pull request #69 from JuliaPOMDP/alphavec_utilities

Some helpful functions to work with alpha vector policies

src/POMDPToolbox.jl

@@ -17,7 +17,6 @@ using ProgressMeter
 using StatsBase
 using DataFrames
 
-
 # export commons
 export
     update,
@@ -70,7 +69,9 @@ include("convenience/implementations.jl")
 
 # policies
 export
-    AlphaVectorPolicy
+    AlphaVectorPolicy,
+    belief_vector,
+    unnormalized_util
 include("policies/alpha_vector.jl")
 
 export

src/beliefs/discrete.jl

@@ -16,7 +16,7 @@ struct DiscreteBelief{P<:POMDP, S}
     b::Vector{Float64}
 end
 
-function DiscreteBelief(pomdp, b::Vector{Float64}; check::Bool=true)
+function DiscreteBelief(pomdp::POMDP, state_list::AbstractVector, b::AbstractVector{Float64}, check::Bool=true)
     if check
         if !isapprox(sum(b), 1.0, atol=0.001)
             warn("""
@@ -33,7 +33,21 @@ function DiscreteBelief(pomdp, b::Vector{Float64}; check::Bool=true)
                  """)
         end
     end
-    return DiscreteBelief(pomdp, ordered_states(pomdp), b)
+    return DiscreteBelief(pomdp, state_list, b)
+end
+
+function DiscreteBelief(pomdp::POMDP, b::Vector{Float64}; check::Bool=true)
+    return DiscreteBelief(pomdp, ordered_states(pomdp), b, check)
+end
+
+function DiscreteBelief(pomdp::POMDP, b; check::Bool=true)
+    # convert b to a vector representation 
+    state_list = ordered_states(pomdp)
+    bv = Vector{Float64}(n_states(pomdp))
+    for (i, s) in enumerate(state_list)
+        bv[i] = pdf(b, s)
+    end
+    return DiscreteBelief(pomdp, state_list, bv, check)
 end
 
 

src/policies/alpha_vector.jl

@@ -51,3 +51,11 @@ function Base.push!(p::AlphaVectorPolicy, alpha::Vector{Float64}, a)
     push!(p.alphas, alpha)
     push!(p.action_map, a)
 end
+
+function action(p::AlphaVectorPolicy, b)
+    return action(p, DiscreteBelief(p.pomdp, b))
+end
+
+function value(p::AlphaVectorPolicy, b)
+    return value(p, DiscreteBelief(p.pomdp, b))
+end

test/test_alpha_policy.jl

@@ -19,3 +19,4 @@ push!(policy, [0.0,0.0], true)
 
 @test value(policy, b0) == 0.0
 @test action(policy, b0) == true
+