got rid of parents, split up label in DPWObsNode

JuliaPOMDP · Jul 20, 2016 · 5797f23 · 5797f23
1 parent e8fff98
commit 5797f23
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 22 deletions.
diff --git a/src/solver.jl b/src/solver.jl
@@ -79,8 +79,7 @@ function simulate{S,A,O,B}(pomcp::POMCPPlanner{S,A,O,POMCPSolver{B}}, h::BeliefN
 			h.children[a] = ActNode(a,
                                     init_N(pomcp.problem, h, a),
                                     init_V(pomcp.problem, h, a),
-                                    h,
-                                    Dict{O,BeliefNode{S,A,O,B}}())
+                                    Dict{O,ObsNode{S,A,O,B}}())
 		end
 
 		return POMDPs.discount(pomcp.problem)^depth * estimate_value(pomcp, pomcp.problem, s, h)
@@ -107,10 +106,10 @@ function simulate{S,A,O,B}(pomcp::POMCPPlanner{S,A,O,POMCPSolver{B}}, h::BeliefN
         hao = best_node.children[o]
     else
         if isa(pomcp.solver.node_belief_updater, ParticleReinvigorator)
-            hao = ObsNode(o, 0, ParticleCollection{S}(), best_node, Dict{A,ActNode{S,A,O,B}}())
+            hao = ObsNode(o, 0, ParticleCollection{S}(), Dict{A,ActNode{A,O,ObsNode{S,A,O,B}}}())
         else
             new_belief = update(pomcp.solver.node_belief_updater, h.B, a, o) # this relies on h.B not being modified
-            hao = ObsNode(o, 0, new_belief, best_node, Dict{A,ActNode{S,A,O,B}}())
+            hao = ObsNode(o, 0, new_belief, Dict{A,ActNode{A,O,ObsNode{S,A,O,B}}}())
         end
         best_node.children[o]=hao
     end
@@ -142,8 +141,7 @@ function simulate{S,A,O,B}(pomcp::POMCPPlanner{S,A,O,POMCPDPWSolver{B}}, h::Beli
             h.children[a] = ActNode(a,
                                     init_N(pomcp.problem, h, a),
                                     init_V(pomcp.problem, h, a),
-                                    h,
-                                    Dict{O,BeliefNode{S,A,O,B}}())
+                                    Dict{O,DPWObsNode{S,A,O,B}}())
         end
         if length(h.children) <= 1
             return POMDPs.discount(pomcp.problem)^depth * estimate_value(pomcp, pomcp.problem, s, h)
@@ -174,20 +172,20 @@ function simulate{S,A,O,B}(pomcp::POMCPPlanner{S,A,O,POMCPDPWSolver{B}}, h::Beli
             hao = best_node.children[o]
         else
             if isa(pomcp.solver.node_belief_updater, ParticleReinvigorator)
-                hao = ObsNodeDPW((o, sp, r,), 0, ParticleCollection{S}(), best_node, Dict{A,ActNode{S,A,O,B}}())
+                hao = DPWObsNode(o, sp, r, 0, ParticleCollection{S}(), Dict{A,ActNode{A,O,DPWObsNode{S,A,O,B}}}())
             else
                 new_belief = update(pomcp.solver.node_belief_updater, h.B, a, o) # this relies on h.B not being modified
-                hao = ObsNodeDPW((o, sp, r,), 0, new_belief, best_node, Dict{Any,ActNode{S,A,O,B}}())
+                hao = DPWObsNode(o, sp, r, 0, new_belief, Dict{A,ActNode{A,O,DPWObsNode{S,A,O,B}}}())
             end
             best_node.children[o]=hao
         end
     else
         # otherwise sample nodes
-        os = collect(values(best_node.children))
-        wv = WeightVec(Int[node.N for node in os])
+        os = collect(values(best_node.children)) # XXX allocation
+        wv = WeightVec(Int[node.N for node in os]) # XXX allocation
         hao = sample(pomcp.solver.rng, os, wv)
-        sp = hao.label[2]
-        r = hao.label[3]
+        sp = hao.state
+        r = hao.reward
     end
 
     R = r + POMDPs.discount(pomcp.problem)*simulate(pomcp, hao, sp, depth+1)

diff --git a/src/tree.jl b/src/tree.jl
@@ -1,27 +1,29 @@
 abstract BeliefNode{S,A,O,B}
 
-type ActNode{S,A,O,B} # Need A, O, everything in belief
+# Note: links to parents were taken out because they hadn't been used in anything we've done so far
+# Note: the label is really only important for visualization
+
+type ActNode{A, O, BNodeType <: BeliefNode} # Need A, O, everything in belief
     label::A # for keeping track of which action this corresponds to
     N::Int64
     V::Float64
-    parent::BeliefNode
-    children::Dict{O,BeliefNode{S,A,O,B}} # maps observations to ObsNodes
+    children::Dict{O, BNodeType} # maps observations to ObsNodes
 end
 
-type ObsNodeDPW{S,A,O,Belief} <: BeliefNode{S,A,O,Belief}
-    label::Tuple{O,S,Float64} # observation, state, reward
+type DPWObsNode{S,A,O,Belief} <: BeliefNode{S,A,O,Belief}
+    label::O
+    state::S
+    reward::Float64
     N::Int64
     B::Belief # belief/state distribution
-    parent::ActNode{S,A,O,Belief}
-    children::Dict{A,ActNode{S,A,O,Belief}}
+    children::Dict{A,ActNode{A,O,DPWObsNode{S,A,O,Belief}}}
 end
 
 type ObsNode{S,A,O,Belief} <: BeliefNode{S,A,O,Belief}
     label::O
     N::Int64
-    B::Belief # belief/state distribution # perhaps should be nullable in the future
-    parent::ActNode{S,A,O,Belief}
-    children::Dict{A,ActNode{S,A,O,Belief}}
+    B::Belief # belief/state distribution
+    children::Dict{A,ActNode{A,O,ObsNode{S,A,O,Belief}}}
 end
 
 type RootNode{RootBelief} <: BeliefNode