minor test update

JuliaPOMDP · Mar 19, 2021 · de466a5 · de466a5
1 parent 5312ba2
commit de466a5
Show file tree

Hide file tree

Showing 6 changed files with 22 additions and 47 deletions.
diff --git a/test/Project.toml b/test/Project.toml
@@ -1,5 +1,7 @@
 [deps]
+BeliefUpdaters = "8bb6e9a1-7d73-552c-a44a-e5dc5634aac4"
 DiscreteValueIteration = "4b033969-44f6-5439-a48b-c11fa3648068"
+FiniteHorizonPOMDPs = "8a13bbfe-798e-11e9-2f1c-eba9ee5ef093"
 POMDPLinter = "f3bd98c0-eb40-45e2-9eb1-f2763262d755"
 POMDPModelTools = "08074719-1b2a-587c-a292-00f91cc44415"
 POMDPModels = "355abbd5-f08e-5560-ac9e-8b5f2592a0ca"

diff --git a/test/custom1dtest.jl b/test/custom1dtest.jl
@@ -15,7 +15,7 @@ fhsolver = FiniteHorizonSolver()
 # MDPs initialization
 mdp = CustomFHExample(no_states, _horizon, actions, actionCost, actionsImpact, reward_states, reward, discount_factor, noise)
 
-# check implementation of required methods 
+# check implementation of required methods
 # @POMDPLinter.show_requirements FiniteHorizonPOMDPs.solve(fhsolver, mdp)
 
 # initialize the solver
@@ -34,7 +34,7 @@ FHPolicy = FiniteHorizonPOMDPs.solve(fhsolver, mdp);
 # Compare resulting policies
 @test all((FiniteHorizonPOMDPs.action(FHPolicy, s) == action(VIPolicy, s) for s in states(mdp)))
 
-# Compare FHMDP and IHMDP states 
+# Compare FHMDP and IHMDP states
 fh_states = Iterators.flatten([FiniteHorizonPOMDPs.stage_states(mdp, i) for i=1:FiniteHorizonPOMDPs.horizon(mdp) + 1])
 ih_states = states(mdp)
 

diff --git a/test/fixhorizon.jl b/test/fixhorizon.jl
@@ -29,5 +29,5 @@ end
 @testset "solver" begin
     fhgw = fixhorizon(SimpleGridWorld(), 3)
     # TODO: Change test to pass without boolean value
-    # @test FiniteHorizonPOMDPs.solve(fhgw)
+    # @test test_solver(FiniteHorizonSolver(), fhgw)
 end
diff --git a/test/fixhorizon1dtest.jl b/test/fixhorizon1dtest.jl
@@ -18,7 +18,7 @@ mdp = FHExample(no_states, actions, actionCost, actionsImpact, reward_states, re
 # Wrap mdp into finite horizon wrapper
 fhex = fixhorizon(mdp, _horizon)
 
-# check implementation of required methods 
+# check implementation of required methods
 # @POMDPLinter.show_requirements FiniteHorizonPOMDPs.solve(fhsolver, fhex)
 
 # initialize the solver
@@ -38,13 +38,10 @@ FHPolicy = FiniteHorizonPOMDPs.solve(fhsolver, fhex);
 # Compare resulting policies
 @test all((FiniteHorizonPOMDPs.action(FHPolicy, s) == action(VIPolicy, s) for s in states(fhex)))
 
-# Compare FHMDP and IHMDP states 
+# Compare FHMDP and IHMDP states
 fh_states = Iterators.flatten([FiniteHorizonPOMDPs.stage_states(fhex, i) for i=1:fhex.horizon + 1])
 ih_states = states(fhex)
 
 z = zip(fh_states, ih_states)
 
 @test all((fh == ih for (fh, ih) in z))
-
-
-
diff --git a/test/instances/Pyramid.jl b/test/instances/Pyramid.jl
@@ -5,7 +5,7 @@ using FiniteHorizonPOMDPs
 #####################
 # MDP and State types
 #####################
-struct PyramidState 
+struct PyramidState
     position::Int64
     epoch::Int64
 end
@@ -31,57 +31,35 @@ FiniteHorizonPOMDPs.horizon(mdp::PyramidMDP) = mdp.horizon
 # changed elements of POMDPs interface
 ###################################
 
-# Creates (horizon(mdp) - 1) * mdp.no_states states to be evaluated and mdp.no_states sink states
-function POMDPs.states(mdp::PyramidMDP)::Array{PyramidState}
-    mdp_states = PyramidState[]
-    for e=1:horizon(mdp) + 1
-        for i=1:e
-            push!(mdp_states, PyramidState(i, e))
-        end
-    end
-
-    return mdp_states
+# Creates (horizon(mdp) + 1) * mdp.no_states states to be evaluated and mdp.no_states sink states
+function POMDPs.states(mdp::PyramidMDP)
+    return [PyramidState(i, e) for e in 1:horizon(mdp) + 1 for i in 1:e]
 end
 
 POMDPs.stateindex(mdp::PyramidMDP, ss::PyramidState)::Int64 = sum(1:ss.epoch - 1) + ss.position
-POMDPs.isterminal(mdp::PyramidMDP, ss::PyramidState) = FiniteHorizonPOMDPs.stage(mdp, ss) > horizon(mdp) || [ss.epoch, ss.position] in mdp.reward_states
+POMDPs.isterminal(mdp::PyramidMDP, ss::PyramidState) = FiniteHorizonPOMDPs.stage(mdp, ss) > horizon(mdp) || (ss.epoch, ss.position) in mdp.reward_states
 
 # returns transition distributions - works only for 1D Gridworld with possible moves to left and to right
-function POMDPs.transition(mdp::PyramidMDP, ss::PyramidState, a::Symbol)::SparseCat{Vector{PyramidState},Vector{Float64}}    
-    sp = PyramidState[]
-    prob = Float64[]
-
-    # add original transition target and probability
-    position = ss.position + mdp.actionsImpact[a]
-    push!(sp, PyramidState(position, ss.epoch + 1))
-    push!(prob, 1. - mdp.noise)
-
-    # add noise transition target and probability
-    noise_action = a == :l ? :r : :l
-    position = ss.position + mdp.actionsImpact[noise_action]
-    push!(sp, PyramidState(position, ss.epoch + 1))
-    push!(prob, mdp.noise)
+function POMDPs.transition(mdp::PyramidMDP, ss::PyramidState, a::Symbol)::SparseCat
+    sp = (  PyramidState(ss.position + mdp.actionsImpact[a], ss.epoch + 1),
+            PyramidState(ss.position + mdp.actionsImpact[a == :l ? :r : :l], ss.epoch + 1))
+    prob = (1. - mdp.noise, mdp.noise)
 
     return SparseCat(sp, prob)
 end
 
 
 POMDPs.actions(mdp::PyramidMDP)::Vector{Symbol} = mdp.actions
 POMDPs.actions(mdp::PyramidMDP, ss::PyramidState) = mdp.actions
-POMDPs.actionindex(mdp::PyramidMDP, a::Symbol)::Int64 = findall(x->x==a, POMDPs.actions(mdp))[1]
+POMDPs.actionindex(mdp::PyramidMDP, a::Symbol)::Int64 = findfirst(x->x==a, POMDPs.actions(mdp))
 
 ###############################
 # FiniteHorizonPOMDPs interface
 ###############################
 FiniteHorizonPOMDPs.stage(mdp::PyramidMDP, ss::PyramidState) = ss.epoch
 
 function FiniteHorizonPOMDPs.stage_states(mdp::PyramidMDP, stage::Int)
-    mdp_states = PyramidState[]
-    for i=1:stage
-        push!(mdp_states, PyramidState(i, stage))
-    end
-
-    return mdp_states
+    return (PyramidState(i, stage) for i in 1:stage)
 end
 
 FiniteHorizonPOMDPs.stage_stateindex(mdp::PyramidMDP, ss::PyramidState) = ss.position
@@ -90,7 +68,7 @@ FiniteHorizonPOMDPs.stage_stateindex(mdp::PyramidMDP, ss::PyramidState) = ss.pos
 # Forwarded parts of POMDPs interface
 ###############################
 function isreward(mdp::PyramidMDP, ss::PyramidState)::Bool
-    return [ss.epoch, ss.position] in mdp.reward_states
+    return (ss.epoch, ss.position) in mdp.reward_states
 end
 
 function POMDPs.reward(mdp::PyramidMDP, ss::PyramidState, a::Symbol, sp::PyramidState)::Float64

diff --git a/test/pyramidtest.jl b/test/pyramidtest.jl
@@ -1,5 +1,3 @@
-using POMDPModelTools
-
 _horizon = 10
 actions = [:l, :r]
 actionCost = 1.
@@ -14,7 +12,7 @@ fhsolver = FiniteHorizonSolver()
 # MDPs initialization
 mdp = PyramidMDP(_horizon, actions, actionCost, actionsImpact, reward_states, reward, discount_factor, noise)
 
-# check implementation of required methods 
+# check implementation of required methods
 # @POMDPLinter.show_requirements FiniteHorizonPOMDPs.solve(fhsolver, mdp)
 
 # initialize the solver
@@ -33,10 +31,10 @@ FHPolicy = FiniteHorizonPOMDPs.solve(fhsolver, mdp);
 # Compare resulting policies
 @test all((FiniteHorizonPOMDPs.action(FHPolicy, s) == action(VIPolicy, s) for s in states(mdp)))
 
-# Compare FHMDP and IHMDP states 
+# Compare FHMDP and IHMDP states
 fh_states = Iterators.flatten([FiniteHorizonPOMDPs.stage_states(mdp, i) for i=1:FiniteHorizonPOMDPs.horizon(mdp) + 1])
 ih_states = states(mdp)
 
 z = zip(fh_states, ih_states)
 
-@test all((fh == ih for (fh, ih) in z))
+@test all((fh == ih for (fh, ih) in z))