added obs_weight (JuliaPOMDP/POMDPs.jl#172)

src/POMDPToolbox.jl

@@ -10,8 +10,8 @@ import POMDPs: Simulator, simulate
 import POMDPs: action, value, solve
 import POMDPs: actions, action_index, state_index, obs_index, iterator, sampletype, states, n_actions, n_states, observations, n_observations, discount, isterminal
 import POMDPs: generate_sr, initial_state
+import POMDPs: implemented
 import Base: rand, rand!, mean, ==
-import DataStructures: CircularBuffer, isfull, capacity, push!, append!
 
 using ProgressMeter
 using StatsBase
@@ -178,6 +178,9 @@ include("model/underlying_mdp.jl")
 # tools for distributions
 include("distributions/distributions_jl.jl")
 
+export obs_weight
+include("model/obs_weight.jl")
+
 export
     weighted_iterator
 include("distributions/weighted_iteration.jl")

src/model/obs_weight.jl

@@ -0,0 +1,87 @@
+# obs_weight is a shortcut function for getting the relative likelihood of an observation without having to construct the observation distribution. Useful for particle filtering
+# maintained by @zsunberg
+
+"""
+    obs_weight(pomdp, sp, o)
+    obs_weight(pomdp, a, sp, o)
+    obs_weight(pomdp, s, a, sp, o)
+
+Return a weight proportional to the likelihood of receiving observation o from state sp (and a and s if they are present).
+
+This is a useful shortcut for particle filtering so that the observation distribution does not have to be represented.
+"""
+function obs_weight end
+
+@generated function obs_weight(p, s, a, sp, o)
+    ow_impl = :(obs_weight(p, a, sp, o))
+    o_impl = :(pdf(observation(p, s, a, sp), o))
+    if implemented(obs_weight, Tuple{p, a, sp, o})
+        return ow_impl
+    elseif implemented(observation, Tuple{p, s, a, sp})
+        return o_impl
+    else
+        return quote
+            try # trick to get the compiler to put the right backedges in
+                $ow_impl
+                $o_impl
+            catch
+                throw(MethodError(obs_weight, (p,s,a,sp,o)))
+            end
+        end
+    end
+end
+
+@generated function obs_weight(p, a, sp, o)
+    ow_impl = :(obs_weight(p, sp, o))
+    o_impl = :(pdf(observation(p, a, sp), o))
+    if implemented(obs_weight, Tuple{p, sp, o})
+        return ow_impl
+    elseif implemented(observation, Tuple{p, a, sp})
+        return o_impl
+    else
+        return quote
+            try # trick to get the compiler to put the right backedges in
+                $ow_impl
+                $o_impl
+            catch
+                throw(MethodError(obs_weight, (p, a, sp, o)))
+            end
+        end
+    end
+end
+
+@generated function obs_weight(p, sp, o)
+    impl = :(pdf(observation(p, sp), o))
+    if implemented(observation, Tuple{p, sp})
+        return impl
+    else
+         return quote
+            try # trick to get the compiler to put the right backedges in
+                $impl
+            catch
+                return :(throw(MethodError(obs_weight, (p, sp, o))))
+            end
+        end
+    end
+end
+
+function implemented(f::typeof(obs_weight), TT::Type)
+    m = which(f, TT)
+    if length(TT.parameters) == 5
+        P, S, A, _, O = TT.parameters
+        reqs_met = implemented(observation, Tuple{P,S,A,S}) || implemented(obs_weight, Tuple{P,A,S,O})
+    elseif length(TT.parameters) == 4
+        P, A, S, O = TT.parameters
+        reqs_met = implemented(observation, Tuple{P,A,S}) || implemented(obs_weight, Tuple{P,S,O})
+    elseif length(TT.parameters) == 3
+        P, S, O = TT.parameters
+        reqs_met = implemented(observation, Tuple{P,S})
+    else
+        return method_exists(f, TT)
+    end
+    if m.module == POMDPToolbox && !reqs_met
+        return false
+    else
+        true
+    end
+end

src/simulators/parallel.jl

@@ -127,7 +127,7 @@ function run_parallel(process::Function, queue::AbstractVector;
         warn("""
              run_parallel(...) was started with only 1 process, so simulations will be run in serial. 
 
-             To supress this warning, use run_parallel(..., proc_warn=false).
+             To suppress this warning, use run_parallel(..., proc_warn=false).
 
              To use multiple processes, use addprocs() or the -p option (e.g. julia -p 4).
              """)

test/runtests.jl

@@ -29,3 +29,4 @@ include("test_info.jl")
 include("test_k_previous_observations_belief.jl")
 include("test_fully_observable_pomdp.jl")
 include("test_underlying_mdp.jl")
+include("test_obs_weight.jl")

test/test_obs_weight.jl

@@ -0,0 +1,18 @@
+import POMDPToolbox: obs_weight
+import POMDPs: observation
+
+struct P <: POMDP{Void, Void, Void} end
+
+@test !@implemented obs_weight(::P, ::Void, ::Void, ::Void, ::Void)
+@test !@implemented obs_weight(::P, ::Void, ::Void, ::Void)
+@test !@implemented obs_weight(::P, ::Void, ::Void)
+
+obs_weight(::P, ::Void, ::Void, ::Void) = 1.0
+@test @implemented obs_weight(::P, ::Void, ::Void, ::Void)
+@test @implemented obs_weight(::P, ::Void, ::Void, ::Void, ::Void)
+@test !@implemented obs_weight(::P, ::Void, ::Void)
+
+@test obs_weight(P(), nothing, nothing, nothing, nothing) == 1.0
+
+observation(::P, ::Void) = nothing
+@test @implemented obs_weight(::P, ::Void, ::Void)