SparseModelMemoryProduct: Fixed incorrect computation of state-action rewards under a randomized policy.

6 years ago · 7ffe322e06
1 changed files with 1 additions and 1 deletions
--- a/src/storm/storage/memorystructure/SparseModelMemoryProduct.cpp
+++ b/src/storm/storage/memorystructure/SparseModelMemoryProduct.cpp
@ -410,7 +410,7 @@ namespace storm {
                                if (isStateReachable(modelState, memoryState)) {
                                    if (scheduler && scheduler->getChoice(modelState, memoryState).isDefined()) {
                                        ValueType factor = scheduler->getChoice(modelState, memoryState).getChoiceAsDistribution().getProbability(rowOffset);
-                                        stateActionRewards.get()[resultTransitionMatrix.getRowGroupIndices()[getResultState(modelState, memoryState)]] = factor * modelStateActionReward;
+                                        stateActionRewards.get()[resultTransitionMatrix.getRowGroupIndices()[getResultState(modelState, memoryState)]] += factor * modelStateActionReward;
                                    } else {
                                        stateActionRewards.get()[resultTransitionMatrix.getRowGroupIndices()[getResultState(modelState, memoryState)] + rowOffset] = modelStateActionReward;
                                    }