Ensure unavailable actions have zero probability.

src/solutions/epsilon_greedy.jl

@@ -52,13 +52,15 @@ function get_action_probs(sol::EpsilonGreedyPolicy, state::State)
 end
 
 function get_action_prob(sol::EpsilonGreedyPolicy, state::State, action::Term)
-    n_actions = length(lazy_collect(available(sol.domain, state)))
+    actions = lazy_collect(available(sol.domain, state))
+    n_actions = length(actions)
     if n_actions == 0 return 0.0 end
     prob = sol.epsilon / n_actions
     best_act = best_action(sol.policy, state)
     if action == best_act
         return prob + (1 - sol.epsilon)
-    else
+    elseif action in actions
         return prob
     end
+    return 0.0
 end

test/solutions.jl

@@ -47,6 +47,8 @@ no_op = convert(Term, PDDL.no_op)
 @test get_action_probs(sol, trajectory[2]) == Dict(plan[2] => 1.0)
 @test get_action_probs(sol, trajectory[end]) == Dict(no_op => 1.0)
 @test get_action_prob(sol, trajectory[1], plan[1]) == 1.0
+@test get_action_prob(sol, trajectory[1], pddl"(pick-up b)") == 0.0
+@test get_action_prob(sol, trajectory[1], pddl"(pick-up z)") == 0.0
 
 @test copy(sol) == sol
 
@@ -62,6 +64,7 @@ n_actions = length(bw_init_actions)
 probs = Dict(a => 1.0 / n_actions for a in bw_init_actions)
 @test get_action_probs(sol, bw_state) == probs
 @test get_action_prob(sol, bw_state, pddl"(pick-up a)") == 1.0 / n_actions
+@test get_action_prob(sol, bw_state, pddl"(pick-up z)") == 0.0
 
 @test copy(sol) == sol
 
@@ -83,8 +86,9 @@ sol.Q[hash(bw_state)] = bw_init_q
 
 probs = Dict(a => a == pddl"(pick-up a)" ? 1.0 : 0.0 for a in bw_init_actions)
 @test get_action_probs(sol, bw_state) == probs
-@test get_action_prob(sol, bw_state, pddl"(pick-up a)") == 1.0 
-@test get_action_prob(sol, bw_state, pddl"(pick-up b)") == 0.0 
+@test get_action_prob(sol, bw_state, pddl"(pick-up a)") == 1.0
+@test get_action_prob(sol, bw_state, pddl"(pick-up b)") == 0.0
+@test get_action_prob(sol, bw_state, pddl"(pick-up z)") == 0.0
 
 @test copy(sol) == sol
 
@@ -109,8 +113,9 @@ end
 
 probs = Dict(a => a == pddl"(pick-up a)" ? 1.0 : 0.0 for a in bw_init_actions)
 @test get_action_probs(sol, bw_state) == probs
-@test get_action_prob(sol, bw_state, pddl"(pick-up a)") == 1.0 
-@test get_action_prob(sol, bw_state, pddl"(pick-up b)") == 0.0 
+@test get_action_prob(sol, bw_state, pddl"(pick-up a)") == 1.0
+@test get_action_prob(sol, bw_state, pddl"(pick-up b)") == 0.0
+@test get_action_prob(sol, bw_state, pddl"(pick-up z)") == 0.0
 
 @test copy(sol) == sol
 
@@ -132,7 +137,7 @@ sol = FunctionalVPolicy(heuristic, blocksworld, bw_spec)
 
 probs = Dict(a => a == pddl"(pick-up a)" ? 1.0 : 0.0 for a in bw_init_actions)
 @test get_action_probs(sol, bw_state) == probs
-@test get_action_prob(sol, bw_state, pddl"(pick-up a)") == 1.0 
+@test get_action_prob(sol, bw_state, pddl"(pick-up a)") == 1.0
 @test get_action_prob(sol, bw_state, pddl"(pick-up b)") == 0.0 
 
 @test copy(sol) == sol
@@ -159,6 +164,7 @@ probs = Dict(zip(keys(bw_init_q), probs))
 @test get_action_probs(sol, bw_state) == probs
 act_prob = probs[pddl"(pick-up a)"]
 @test get_action_prob(sol, bw_state, pddl"(pick-up a)") ≈ act_prob
+@test get_action_prob(sol, bw_state, pddl"(pick-up z)") == 0.0
 
 @test copy(sol) == sol
 
@@ -184,6 +190,7 @@ probs[pddl"(pick-up a)"] += 0.9
 @test get_action_probs(sol, bw_state) == probs
 act_prob = probs[pddl"(pick-up a)"]
 @test get_action_prob(sol, bw_state, pddl"(pick-up a)") ≈ act_prob
+@test get_action_prob(sol, bw_state, pddl"(pick-up z)") == 0.0
 
 @test copy(sol) == sol