sp
/
tempest
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
Browse Source

Removal of unused extractActions function

tempestpy_adaptions
Alexander Bork 5 years ago
parent
e1bd87b91a
commit
c74a6ea201
2 changed files with 8 additions and 143 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 91
      src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.cpp
  2. 46
      src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.h

91
src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.cpp
View File

@ -884,96 +884,6 @@ namespace storm {
return smb.build();
}
template<typename ValueType, typename RewardModelType>
std::vector<uint64_t> ApproximatePOMDPModelchecker<ValueType, RewardModelType>::extractBestActions(
storm::models::sparse::Pomdp<ValueType, RewardModelType> const &pomdp,
std::vector<storm::pomdp::Belief<ValueType>> &beliefList,
std::vector<bool> &beliefIsTarget,
std::set<uint32_t> const &targetObservations,
std::map<uint64_t, std::vector<std::map<uint32_t, ValueType>>> &observationProbabilities,
std::map<uint64_t, std::vector<std::map<uint32_t, uint64_t>>> &nextBelieves,
std::map<uint64_t, ValueType> &result,
uint64_t gridResolution, uint64_t currentBeliefId, uint64_t nextId, bool min) {
storm::pomdp::Belief<ValueType> currentBelief = beliefList[currentBeliefId];
//TODO put this in extra function
std::vector<std::map<uint32_t, ValueType>> observationProbabilitiesInAction;
std::vector<std::map<uint32_t, uint64_t>> nextBelievesInAction;
uint64_t numChoices = pomdp.getNumberOfChoices(
pomdp.getStatesWithObservation(currentBelief.observation).front());
for (uint64_t action = 0; action < numChoices; ++action) {
std::map<uint32_t, ValueType> actionObservationProbabilities = computeObservationProbabilitiesAfterAction(
pomdp, currentBelief, action);
std::map<uint32_t, uint64_t> actionObservationBelieves;
for (auto iter = actionObservationProbabilities.begin(); iter != actionObservationProbabilities.end(); ++iter) {
uint32_t observation = iter->first;
actionObservationBelieves[observation] = getBeliefAfterActionAndObservation(pomdp, beliefList, beliefIsTarget, targetObservations, currentBelief,
action, observation, nextId);
nextId = beliefList.size();
}
observationProbabilitiesInAction.push_back(actionObservationProbabilities);
nextBelievesInAction.push_back(actionObservationBelieves);
}
observationProbabilities.emplace(std::make_pair(currentBeliefId, observationProbabilitiesInAction));
nextBelieves.emplace(std::make_pair(currentBeliefId, nextBelievesInAction));
// choose the action which results in the value computed by the over-approximation
ValueType chosenValue = min ? storm::utility::infinity<ValueType>() : -storm::utility::infinity<ValueType>();
std::vector<uint64_t> chosenActionIndices;
ValueType currentValue;
for (uint64_t action = 0; action < numChoices; ++action) {
currentValue = storm::utility::zero<ValueType>(); // simply change this for rewards?
for (auto iter = observationProbabilities[currentBelief.id][action].begin();
iter != observationProbabilities[currentBelief.id][action].end(); ++iter) {
uint32_t observation = iter->first;
storm::pomdp::Belief<ValueType> nextBelief = beliefList[nextBelieves[currentBelief.id][action][observation]];
// compute subsimplex and lambdas according to the Lovejoy paper to approximate the next belief
auto temp = computeSubSimplexAndLambdas(nextBelief.probabilities, gridResolution, pomdp.getNumberOfStates());
std::vector<std::map<uint64_t, ValueType>> subSimplex = temp.first;
std::vector<ValueType> lambdas = temp.second;
auto sum = storm::utility::zero<ValueType>();
for (size_t j = 0; j < lambdas.size(); ++j) {
if (!cc.isEqual(lambdas[j], storm::utility::zero<ValueType>())) {
sum += lambdas[j] * result.at(getBeliefIdInVector(beliefList, observation, subSimplex[j]));
}
}
currentValue += iter->second * sum;
}
// Update the selected actions
if ((min && cc.isLess(storm::utility::zero<ValueType>(), chosenValue - currentValue)) ||
(!min &&
cc.isLess(storm::utility::zero<ValueType>(), currentValue - chosenValue)) ||
cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenValue = currentValue;
if (!cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenActionIndices.clear();
}
chosenActionIndices.push_back(action);
}
}
return chosenActionIndices;
}
template<typename ValueType, typename RewardModelType>
std::vector<uint64_t> ApproximatePOMDPModelchecker<ValueType, RewardModelType>::extractBestAction(
storm::models::sparse::Pomdp<ValueType, RewardModelType> const &pomdp,
std::vector<storm::pomdp::Belief<ValueType>> &beliefList,
std::vector<bool> &beliefIsTarget,
std::set<uint32_t> const &targetObservations,
std::map<uint64_t, std::vector<std::map<uint32_t, ValueType>>> &observationProbabilities,
std::map<uint64_t, std::vector<std::map<uint32_t, uint64_t>>> &nextBelieves,
std::map<uint64_t, ValueType> &result,
uint64_t gridResolution, uint64_t currentBeliefId, uint64_t nextId, bool min) {
return std::vector<uint64_t>{
extractBestActions(pomdp, beliefList, beliefIsTarget, targetObservations, observationProbabilities, nextBelieves, result, gridResolution, currentBeliefId,
nextId, min).front()};
}
template<typename ValueType, typename RewardModelType>
uint64_t ApproximatePOMDPModelchecker<ValueType, RewardModelType>::getBeliefIdInVector(
std::vector<storm::pomdp::Belief<ValueType>> const &grid, uint32_t observation,
@ -1265,6 +1175,7 @@ namespace storm {
class ApproximatePOMDPModelchecker<double>;
#ifdef STORM_HAVE_CARL
template
class ApproximatePOMDPModelchecker<storm::RationalNumber>;

46
src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.h
View File

@ -179,52 +179,6 @@ namespace storm {
std::set<uint32_t> const &targetObservations, bool min,
uint64_t gridResolution, bool computeRewards);
/**
* TODO
* @param pomdp
* @param beliefList
* @param observationProbabilities
* @param nextBelieves
* @param result
* @param gridResolution
* @param currentBeliefId
* @param nextId
* @param min
* @return
*/
std::vector<uint64_t> extractBestActions(storm::models::sparse::Pomdp<ValueType, RewardModelType> const &pomdp,
std::vector<storm::pomdp::Belief<ValueType>> &beliefList,
std::vector<bool> &beliefIsTarget,
std::set<uint32_t> const &target_observations,
std::map<uint64_t, std::vector<std::map<uint32_t, ValueType>>> &observationProbabilities,
std::map<uint64_t, std::vector<std::map<uint32_t, uint64_t>>> &nextBelieves,
std::map<uint64_t, ValueType> &result,
uint64_t gridResolution, uint64_t currentBeliefId, uint64_t nextId,
bool min);
/**
* TODO
* @param pomdp
* @param beliefList
* @param observationProbabilities
* @param nextBelieves
* @param result
* @param gridResolution
* @param currentBeliefId
* @param nextId
* @param min
* @return
*/
std::vector<uint64_t> extractBestAction(storm::models::sparse::Pomdp<ValueType, RewardModelType> const &pomdp,
std::vector<storm::pomdp::Belief<ValueType>> &beliefList,
std::vector<bool> &beliefIsTarget,
std::set<uint32_t> const &target_observations,
std::map<uint64_t, std::vector<std::map<uint32_t, ValueType>>> &observationProbabilities,
std::map<uint64_t, std::vector<std::map<uint32_t, uint64_t>>> &nextBelieves,
std::map<uint64_t, ValueType> &result,
uint64_t gridResolution, uint64_t currentBeliefId, uint64_t nextId,
bool min);
/**
* Helper to compute an underapproximation of the reachability property.
* The implemented method unrolls the belief support of the given POMDP up to a given number of belief states.

Write Preview
Loading…
Cancel
Save