handle constrained actions

src/sparse_tabular.jl

@@ -179,7 +179,8 @@ function transition_matrix_a_s_sp(mdp::Union{MDP, POMDP})
         end
     end
     transmats_A_S_S2 = [sparse(transmat_row_A[a], transmat_col_A[a], transmat_data_A[a], n_states(mdp), n_states(mdp)) for a in 1:n_actions(mdp)]
-    @assert all(all(sum(transmats_A_S_S2[a], dims=2) .≈ ones(n_states(mdp))) for a in 1:n_actions(mdp)) "Transition probabilities must sum to 1"
+    # if an action is not valid from a state, the transition is 0.0 everywhere
+    # @assert all(all(sum(transmats_A_S_S2[a], dims=2) .≈ ones(n_states(mdp))) for a in 1:n_actions(mdp)) "Transition probabilities must sum to 1"
     return transmats_A_S_S2
 end
 
@@ -251,6 +252,7 @@ POMDPs.n_actions(prob::SparseTabularProblem) = size(prob.T, 1)
 
 POMDPs.states(p::SparseTabularProblem) = 1:n_states(p)
 POMDPs.actions(p::SparseTabularProblem) = 1:n_actions(p)
+POMDPs.actions(p::SparseTabularProblem, s::Int64) = [a for a in actions(p) if sum(transition_matrix(p, a)) ≈ n_states(p)]
 
 POMDPs.stateindex(::SparseTabularProblem, s::Int64) = s
 POMDPs.actionindex(::SparseTabularProblem, a::Int64) = a

test/test_tabular.jl

@@ -66,6 +66,8 @@ smdp3 = SparseTabularMDP(smdp, discount = 0.5)
 gw = SimpleGridWorld()
 sparsegw = SparseTabularMDP(gw)
 @test isterminal(sparsegw, 101)
+@inferred actions(sparsegw, 101)
+@test actions(sparsegw, 101) == collect(actions(sparsegw))
 
 ## POMDP