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

Fix of wrong MDP underapproximation

tempestpy_adaptions
Alexander Bork 5 years ago
parent
b9c0af6628
commit
bbd3ec7287
2 changed files with 108 additions and 64 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. 131
      src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.cpp
  2. 31
      src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.h

131
src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.cpp
View File

@ -11,6 +11,7 @@
#include "storm/modelchecker/results/ExplicitQualitativeCheckResult.h"
#include "storm/modelchecker/results/ExplicitQuantitativeCheckResult.h"
#include "storm/api/properties.h"
#include "storm/api/export.h"
#include "storm-parsers/api/storm-parsers.h"
namespace storm {
@ -18,7 +19,9 @@ namespace storm {
namespace modelchecker {
template<typename ValueType, typename RewardModelType>
ApproximatePOMDPModelchecker<ValueType, RewardModelType>::ApproximatePOMDPModelchecker() {
cc = storm::utility::ConstantsComparator<ValueType>(storm::utility::convertNumber<ValueType>(0.00000000001), false);
precision = 0.000000001;
cc = storm::utility::ConstantsComparator<ValueType>(storm::utility::convertNumber<ValueType>(precision), false);
useMdp = false;
}
template<typename ValueType, typename RewardModelType>
@ -59,7 +62,8 @@ namespace storm {
//Use caching to avoid multiple computation of the subsimplices and lambdas
std::map<uint64_t, std::vector<std::vector<ValueType>>> subSimplexCache;
std::map<uint64_t, std::vector<ValueType>> lambdaCache;
std::map<uint64_t, uint64_t> chosenActions;
std::map<uint64_t, std::vector<uint64_t>> chosenActions;
std::deque<uint64_t> beliefsToBeExpanded;
@ -101,8 +105,7 @@ namespace storm {
storm::pomdp::Belief<ValueType> gridBelief = {nextId, initialBelief.observation, initSubSimplex[j]};
beliefList.push_back(gridBelief);
beliefGrid.push_back(gridBelief);
beliefIsTarget.push_back(
targetObservations.find(initialBelief.observation) != targetObservations.end());
beliefIsTarget.push_back(targetObservations.find(initialBelief.observation) != targetObservations.end());
beliefsToBeExpanded.push_back(nextId);
++nextId;
}
@ -125,14 +128,12 @@ namespace storm {
result.emplace(std::make_pair(currId, storm::utility::zero<ValueType>()));
result_backup.emplace(std::make_pair(currId, storm::utility::zero<ValueType>()));
uint64_t numChoices = pomdp.getNumberOfChoices(
pomdp.getStatesWithObservation(beliefList[currId].observation).front());
uint64_t numChoices = pomdp.getNumberOfChoices(pomdp.getStatesWithObservation(beliefList[currId].observation).front());
std::vector<std::map<uint32_t, ValueType>> observationProbabilitiesInAction(numChoices);
std::vector<std::map<uint32_t, uint64_t>> nextBelievesInAction(numChoices);
for (uint64_t action = 0; action < numChoices; ++action) {
std::map<uint32_t, ValueType> actionObservationProbabilities = computeObservationProbabilitiesAfterAction(
pomdp, beliefList[currId], action);
std::map<uint32_t, ValueType> actionObservationProbabilities = computeObservationProbabilitiesAfterAction(pomdp, beliefList[currId], action);
std::map<uint32_t, uint64_t> actionObservationBelieves;
for (auto iter = actionObservationProbabilities.begin();
iter != actionObservationProbabilities.end(); ++iter) {
@ -166,8 +167,7 @@ namespace storm {
storm::pomdp::Belief<ValueType> gridBelief = {nextId, observation, subSimplex[j]};
beliefList.push_back(gridBelief);
beliefGrid.push_back(gridBelief);
beliefIsTarget.push_back(
targetObservations.find(observation) != targetObservations.end());
beliefIsTarget.push_back(targetObservations.find(observation) != targetObservations.end());
beliefsToBeExpanded.push_back(nextId);
++nextId;
}
@ -201,9 +201,8 @@ namespace storm {
uint64_t numChoices = pomdp.getNumberOfChoices(
pomdp.getStatesWithObservation(currentBelief.observation).front());
// Initialize the values for the value iteration
ValueType chosenValue = min ? storm::utility::infinity<ValueType>()
: -storm::utility::infinity<ValueType>();
uint64_t chosenActionIndex = std::numeric_limits<uint64_t>::infinity();
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) {
@ -231,8 +230,7 @@ namespace storm {
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_backup.at(
getBeliefIdInVector(beliefGrid, observation, subSimplex[j]));
sum += lambdas[j] * result_backup.at(getBeliefIdInVector(beliefGrid, observation, subSimplex[j]));
}
}
currentValue += iter->second * sum;
@ -243,11 +241,14 @@ namespace storm {
cc.isLess(storm::utility::zero<ValueType>(), currentValue - chosenValue)) ||
cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenValue = currentValue;
chosenActionIndex = action;
if (!(useMdp && cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue))) {
chosenActionIndices.clear();
}
chosenActionIndices.push_back(action);
}
}
result[currentBelief.id] = chosenValue;
chosenActions[currentBelief.id] = chosenActionIndex;
chosenActions[currentBelief.id] = chosenActionIndices;
// Check if the iteration brought an improvement
if (cc.isLess(storm::utility::zero<ValueType>(), result[currentBelief.id] - result_backup[currentBelief.id])) {
improvement = true;
@ -312,7 +313,7 @@ namespace storm {
//Use caching to avoid multiple computation of the subsimplices and lambdas
std::map<uint64_t, std::vector<std::vector<ValueType>>> subSimplexCache;
std::map<uint64_t, std::vector<ValueType>> lambdaCache;
std::map<uint64_t, uint64_t> chosenActions;
std::map<uint64_t, std::vector<uint64_t>> chosenActions;
std::deque<uint64_t> beliefsToBeExpanded;
@ -459,7 +460,7 @@ namespace storm {
// Initialize the values for the value iteration
ValueType chosenValue = min ? storm::utility::infinity<ValueType>()
: -storm::utility::infinity<ValueType>();
uint64_t chosenActionIndex = std::numeric_limits<uint64_t>::infinity();
std::vector<uint64_t> chosenActionIndices;
ValueType currentValue;
for (uint64_t action = 0; action < numChoices; ++action) {
@ -499,15 +500,17 @@ namespace storm {
(!min &&
cc.isLess(storm::utility::zero<ValueType>(), currentValue - chosenValue)) ||
cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenValue = currentValue;
chosenActionIndex = action;
if (!(useMdp && cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue))) {
chosenActionIndices.clear();
}
chosenActionIndices.push_back(action);
}
}
result[currentBelief.id] = chosenValue;
chosenActions[currentBelief.id] = chosenActionIndex;
chosenActions[currentBelief.id] = chosenActionIndices;
// Check if the iteration brought an improvement
if (cc.isLess(storm::utility::zero<ValueType>(), result_backup[currentBelief.id] - result[currentBelief.id]) ||
cc.isLess(storm::utility::zero<ValueType>(), result[currentBelief.id] - result_backup[currentBelief.id])) {
@ -586,7 +589,7 @@ namespace storm {
std::map<uint64_t, ValueType> result;
std::map<uint64_t, ValueType> result_backup;
// Belief ID -> ActionIndex
std::map<uint64_t, uint64_t> chosenActions;
std::map<uint64_t, std::vector<uint64_t>> chosenActions;
// Belief ID -> Observation -> Probability
std::map<uint64_t, std::vector<std::map<uint32_t, ValueType>>> observationProbabilities;
@ -611,11 +614,9 @@ namespace storm {
std::vector<std::map<uint32_t, uint64_t>> nextBelievesInAction(numChoices);
for (uint64_t action = 0; action < numChoices; ++action) {
std::map<uint32_t, ValueType> actionObservationProbabilities = computeObservationProbabilitiesAfterAction(
pomdp, currentBelief, 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) {
for (auto iter = actionObservationProbabilities.begin(); iter != actionObservationProbabilities.end(); ++iter) {
uint32_t observation = iter->first;
// THIS CALL IS SLOW
// TODO speed this up
@ -659,7 +660,7 @@ namespace storm {
// Initialize the values for the value iteration
ValueType chosenValue = min ? storm::utility::infinity<ValueType>()
: -storm::utility::infinity<ValueType>();
uint64_t chosenActionIndex = std::numeric_limits<uint64_t>::infinity();
std::vector<uint64_t> chosenActionIndices;
ValueType currentValue;
for (uint64_t action = 0; action < numChoices; ++action) {
@ -698,11 +699,14 @@ namespace storm {
cc.isLess(storm::utility::zero<ValueType>(), currentValue - chosenValue)) ||
cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenValue = currentValue;
chosenActionIndex = action;
if (!(useMdp && cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue))) {
chosenActionIndices.clear();
}
chosenActionIndices.push_back(action);
}
}
result[currentBelief.id] = chosenValue;
chosenActions[currentBelief.id] = chosenActionIndex;
chosenActions[currentBelief.id] = chosenActionIndices;
// Check if the iteration brought an improvement
if (cc.isLess(storm::utility::zero<ValueType>(), result[currentBelief.id] - result_backup[currentBelief.id])) {
improvement = true;
@ -721,8 +725,7 @@ namespace storm {
STORM_PRINT("Overapproximation took " << iteration << " iterations" << std::endl);
beliefGrid.push_back(initialBelief);
beliefIsTarget.push_back(
targetObservations.find(initialBelief.observation) != targetObservations.end());
beliefIsTarget.push_back(targetObservations.find(initialBelief.observation) != targetObservations.end());
std::pair<std::vector<std::vector<ValueType>>, std::vector<ValueType>> temp = computeSubSimplexAndLambdas(
initialBelief.probabilities, gridResolution);
@ -740,7 +743,6 @@ namespace storm {
overApproxTimer.stop();
// Now onto the under-approximation
bool useMdp = true;
storm::utility::Stopwatch underApproxTimer(true);
ValueType underApprox = useMdp ? computeUnderapproximationWithMDP(pomdp, beliefList, beliefIsTarget, targetObservations, observationProbabilities, nextBelieves,
result, chosenActions, gridResolution, initialBelief.id, min, false) :
@ -795,7 +797,7 @@ namespace storm {
std::map<uint64_t, ValueType> result;
std::map<uint64_t, ValueType> result_backup;
// Belief ID -> ActionIndex
std::map<uint64_t, uint64_t> chosenActions;
std::map<uint64_t, std::vector<uint64_t>> chosenActions;
// Belief ID -> Observation -> Probability
std::map<uint64_t, std::vector<std::map<uint32_t, ValueType>>> observationProbabilities;
@ -874,7 +876,7 @@ namespace storm {
// Initialize the values for the value iteration
ValueType chosenValue = min ? storm::utility::infinity<ValueType>()
: -storm::utility::infinity<ValueType>();
uint64_t chosenActionIndex = std::numeric_limits<uint64_t>::infinity();
std::vector<uint64_t> chosenActionIndices;
ValueType currentValue;
for (uint64_t action = 0; action < numChoices; ++action) {
@ -913,15 +915,17 @@ namespace storm {
(!min &&
cc.isLess(storm::utility::zero<ValueType>(), currentValue - chosenValue)) ||
cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenValue = currentValue;
chosenActionIndex = action;
if (!(useMdp && cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue))) {
chosenActionIndices.clear();
}
chosenActionIndices.push_back(action);
}
}
result[currentBelief.id] = chosenValue;
chosenActions[currentBelief.id] = chosenActionIndex;
chosenActions[currentBelief.id] = chosenActionIndices;
// Check if the iteration brought an improvement
if (cc.isLess(storm::utility::zero<ValueType>(), result_backup[currentBelief.id] - result[currentBelief.id]) ||
cc.isLess(storm::utility::zero<ValueType>(), result[currentBelief.id] - result_backup[currentBelief.id])) {
@ -960,7 +964,6 @@ namespace storm {
overApproxTimer.stop();
// Now onto the under-approximation
bool useMdp = true;
storm::utility::Stopwatch underApproxTimer(true);
ValueType underApprox = useMdp ? computeUnderapproximationWithMDP(pomdp, beliefList, beliefIsTarget, targetObservations, observationProbabilities, nextBelieves,
result, chosenActions, gridResolution, initialBelief.id, min, true) :
@ -989,7 +992,7 @@ namespace storm {
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,
std::map<uint64_t, uint64_t> chosenActions,
std::map<uint64_t, std::vector<uint64_t>> chosenActions,
uint64_t gridResolution, uint64_t initialBeliefId, bool min,
bool computeReward) {
std::set<uint64_t> visitedBelieves;
@ -1017,17 +1020,15 @@ namespace storm {
} else {
if (chosenActions.find(currentBeliefId) == chosenActions.end()) {
// If the current Belief is not part of the grid, we have not computed the action to choose yet
chosenActions[currentBeliefId] = extractBestAction(pomdp, beliefList, beliefIsTarget,
targetObservations,
chosenActions[currentBeliefId] = extractBestAction(pomdp, beliefList, beliefIsTarget, targetObservations,
observationProbabilities,
nextBelieves, result, gridResolution,
currentBeliefId, beliefList.size(), min);
}
for (auto iter = observationProbabilities[currentBeliefId][chosenActions[currentBeliefId]].begin();
iter !=
observationProbabilities[currentBeliefId][chosenActions[currentBeliefId]].end(); ++iter) {
for (auto iter = observationProbabilities[currentBeliefId][chosenActions[currentBeliefId][0]].begin();
iter != observationProbabilities[currentBeliefId][chosenActions[currentBeliefId][0]].end(); ++iter) {
uint32_t observation = iter->first;
uint64_t nextBeliefId = nextBelieves[currentBeliefId][chosenActions[currentBeliefId]][observation];
uint64_t nextBeliefId = nextBelieves[currentBeliefId][chosenActions[currentBeliefId][0]][observation];
if (visitedBelieves.insert(nextBeliefId).second) {
beliefStateMap[nextBeliefId] = stateId;
++stateId;
@ -1049,13 +1050,15 @@ namespace storm {
}
storm::models::sparse::StandardRewardModel<ValueType> rewardModel(std::vector<ValueType>(beliefStateMap.size()));
if (computeReward) {
for (auto const &iter : beliefStateMap) {
auto currentBelief = beliefList[iter.first];
// Add the reward collected by taking the chosen Action in the belief
rewardModel.setStateReward(iter.second, getRewardAfterAction(pomdp, pomdp.getChoiceIndex(
storm::storage::StateActionPair(pomdp.getStatesWithObservation(currentBelief.observation).front(), chosenActions[iter.first])),
storm::storage::StateActionPair(pomdp.getStatesWithObservation(currentBelief.observation).front(), chosenActions[iter.first][0])),
currentBelief));
}
}
std::unordered_map<std::string, RewardModelType> rewardModels = {{"std", rewardModel}};
@ -1092,7 +1095,7 @@ namespace storm {
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,
std::map<uint64_t, uint64_t> chosenActions,
std::map<uint64_t, std::vector<uint64_t>> chosenActions,
uint64_t gridResolution, uint64_t initialBeliefId, bool min,
bool computeRewards) {
std::set<uint64_t> visitedBelieves;
@ -1152,8 +1155,7 @@ namespace storm {
std::map<uint64_t, ValueType> transitionsInStateWithAction;
for (auto iter = observationProbabilities[currentBeliefId][action].begin();
iter !=
observationProbabilities[currentBeliefId][action].end(); ++iter) {
iter != observationProbabilities[currentBeliefId][action].end(); ++iter) {
uint32_t observation = iter->first;
uint64_t nextBeliefId = nextBelieves[currentBeliefId][action][observation];
if (visitedBelieves.insert(nextBeliefId).second) {
@ -1270,7 +1272,7 @@ namespace storm {
}
template<typename ValueType, typename RewardModelType>
uint64_t ApproximatePOMDPModelchecker<ValueType, RewardModelType>::extractBestAction(
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,
@ -1312,7 +1314,7 @@ namespace storm {
// choose the action which results in the value computed by the over-approximation
ValueType chosenValue = min ? storm::utility::infinity<ValueType>()
: -storm::utility::infinity<ValueType>();
uint64_t chosenActionIndex = std::numeric_limits<uint64_t>::infinity();
std::vector<uint64_t> chosenActionIndices;
ValueType currentValue;
for (uint64_t action = 0; action < numChoices; ++action) {
@ -1330,8 +1332,7 @@ namespace storm {
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]));
sum += lambdas[j] * result.at(getBeliefIdInVector(beliefList, observation, subSimplex[j]));
}
}
currentValue += iter->second * sum;
@ -1343,10 +1344,28 @@ namespace storm {
cc.isLess(storm::utility::zero<ValueType>(), currentValue - chosenValue)) ||
cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenValue = currentValue;
chosenActionIndex = action;
if (!cc.isEqual(storm::utility::zero<ValueType>(), chosenValue - currentValue)) {
chosenActionIndices.clear();
}
chosenActionIndices.push_back(action);
}
}
return chosenActionIndices;
}
return chosenActionIndex;
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> &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()};
}

31
src/storm-pomdp/modelchecker/ApproximatePOMDPModelchecker.h
View File

@ -58,7 +58,30 @@ namespace storm {
* @param min
* @return
*/
uint64_t extractBestAction(storm::models::sparse::Pomdp<ValueType, RewardModelType> const &pomdp,
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> &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> &target_observations,
@ -90,7 +113,7 @@ namespace storm {
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,
std::map<uint64_t, uint64_t> chosenActions,
std::map<uint64_t, std::vector<uint64_t>> chosenActions,
uint64_t gridResolution, uint64_t initialBeliefId, bool min, bool computeReward);
ValueType computeUnderapproximationWithMDP(storm::models::sparse::Pomdp<ValueType, RewardModelType> const &pomdp,
@ -100,7 +123,7 @@ namespace storm {
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,
std::map<uint64_t, uint64_t> chosenActions,
std::map<uint64_t, std::vector<uint64_t>> chosenActions,
uint64_t gridResolution, uint64_t initialBeliefId, bool min, bool computeRewards);
/**
@ -200,6 +223,8 @@ namespace storm {
ValueType getRewardAfterAction(storm::models::sparse::Pomdp<ValueType, RewardModelType> const &pomdp, uint64_t action, storm::pomdp::Belief<ValueType> belief);
storm::utility::ConstantsComparator<ValueType> cc;
double precision;
bool useMdp;
};
}

Write Preview
Loading…
Cancel
Save