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

added possibility to lift transient edge destination assignments to the edge by scaling with the probability (only if this preserves the considered properties).

main
TimQu 7 years ago
parent
e2cb68b31f
commit
6aaafea554
6 changed files with 135 additions and 69 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. 16
      src/storm/builder/BuilderOptions.cpp
  2. 12
      src/storm/builder/BuilderOptions.h
  3. 91
      src/storm/generator/JaniNextStateGenerator.cpp
  4. 2
      src/storm/generator/JaniNextStateGenerator.h
  5. 52
      src/storm/storage/jani/OrderedAssignments.cpp
  6. 5
      src/storm/storage/jani/OrderedAssignments.h

16
src/storm/builder/BuilderOptions.cpp
View File

@ -1,6 +1,7 @@
#include "storm/builder/BuilderOptions.h"
#include "storm/logic/Formulas.h"
#include "storm/logic/FragmentSpecification.h"
#include "storm/settings/SettingsManager.h"
#include "storm/settings/modules/BuildSettings.h"
@ -36,7 +37,7 @@ namespace storm {
return boost::get<storm::expressions::Expression>(labelOrExpression);
}
BuilderOptions::BuilderOptions(bool buildAllRewardModels, bool buildAllLabels) : buildAllRewardModels(buildAllRewardModels), buildAllLabels(buildAllLabels), buildChoiceLabels(false), buildStateValuations(false), buildChoiceOrigins(false), explorationChecks(false), addOverlappingGuardsLabel(false), addOutOfBoundsState(false), showProgress(false), showProgressDelay(0) {
BuilderOptions::BuilderOptions(bool buildAllRewardModels, bool buildAllLabels) : buildAllRewardModels(buildAllRewardModels), buildAllLabels(buildAllLabels), buildChoiceLabels(false), buildStateValuations(false), buildChoiceOrigins(false), scaleAndLiftTransitionRewards(true), explorationChecks(false), addOverlappingGuardsLabel(false), addOutOfBoundsState(false), showProgress(false), showProgressDelay(0) {
// Intentionally left empty.
}
@ -85,6 +86,10 @@ namespace storm {
for (auto const& formula : atomicExpressionFormulas) {
addLabel(formula->getExpression());
}
storm::logic::FragmentSpecification transitionRewardScalingFragment = storm::logic::csl().setRewardOperatorsAllowed(true).setReachabilityRewardFormulasAllowed(true).setLongRunAverageOperatorsAllowed(true).setMultiObjectiveFormulasAllowed(true).setTotalRewardFormulasAllowed(true).setStepBoundedCumulativeRewardFormulasAllowed(true).setTimeBoundedCumulativeRewardFormulasAllowed(true);
scaleAndLiftTransitionRewards = scaleAndLiftTransitionRewards && formula.isInFragment(transitionRewardScalingFragment);
}
void BuilderOptions::setTerminalStatesFromFormula(storm::logic::Formula const& formula) {
@ -160,6 +165,10 @@ namespace storm {
return buildAllLabels;
}
bool BuilderOptions::isScaleAndLiftTransitionRewardsSet() const {
return scaleAndLiftTransitionRewards;
}
bool BuilderOptions::isAddOutOfBoundsStateSet() const {
return addOutOfBoundsState;
}
@ -237,6 +246,11 @@ namespace storm {
return *this;
}
BuilderOptions& BuilderOptions::setScaleAndLiftTransitionRewards(bool newValue) {
scaleAndLiftTransitionRewards = newValue;
return *this;
}
BuilderOptions& BuilderOptions::setAddOutOfBoundsState(bool newValue) {
addOutOfBoundsState = newValue;
return *this;

12
src/storm/builder/BuilderOptions.h
View File

@ -107,6 +107,7 @@ namespace storm {
bool isBuildAllLabelsSet() const;
bool isExplorationChecksSet() const;
bool isShowProgressSet() const;
bool isScaleAndLiftTransitionRewardsSet() const;
bool isAddOutOfBoundsStateSet() const;
bool isAddOverlappingGuardLabelSet() const;
uint64_t getShowProgressDelay() const;
@ -158,6 +159,13 @@ namespace storm {
*/
BuilderOptions& setExplorationChecks(bool newValue = true);
/**
* Should extra checks be performed during exploration
* @param newValue The new value (default true)
* @return this
*/
BuilderOptions& setScaleAndLiftTransitionRewards(bool newValue = true);
/**
* Should a state for out of bounds be constructed
* @param newValue The new value (default true)
@ -202,6 +210,9 @@ namespace storm {
// each choice originates.
bool buildChoiceOrigins;
/// A flag that stores whether potentially occurring transition rewards should be scaled and lifted to the edge
bool scaleAndLiftTransitionRewards;
/// A flag that stores whether exploration checks are to be performed.
bool explorationChecks;
@ -216,6 +227,7 @@ namespace storm {
/// The delay for printing progress information.
uint64_t showProgressDelay;
};
}

91
src/storm/generator/JaniNextStateGenerator.cpp
View File

@ -42,6 +42,7 @@ namespace storm {
STORM_LOG_THROW(!model.hasNonGlobalTransientVariable(), storm::exceptions::InvalidSettingsException, "The explicit next-state generator currently does not support automata-local transient variables.");
STORM_LOG_THROW(!this->options.isBuildChoiceLabelsSet(), storm::exceptions::InvalidSettingsException, "JANI next-state generator cannot generate choice labels.");
// Eliminate arrays if necessary.
if (this->model.containsArrayVariables()) {
arrayEliminatorData = this->model.eliminateArrays(true);
}
@ -50,8 +51,16 @@ namespace storm {
uint64_t lowestAssignmentLevel = storm::jani::AssignmentLevelFinder().getLowestAssignmentLevel(this->model);
if (this->model.hasTransientEdgeDestinationAssignments()) {
this->model.liftTransientEdgeDestinationAssignments(lowestAssignmentLevel);
if (this->model.hasTransientEdgeDestinationAssignments()) {
STORM_LOG_THROW(options.isScaleAndLiftTransitionRewardsSet(), storm::exceptions::InvalidSettingsException, "The explicit next-state generator currently does not support transient edge destination assignments.");
} else {
// There are no edge destination assignments so we turn the lifting to edges off.
this->options.setScaleAndLiftTransitionRewards(false);
}
} else {
// There are no edge destination assignments so we turn the lifting to edges off.
this->options.setScaleAndLiftTransitionRewards(false);
}
STORM_LOG_THROW(!this->model.hasTransientEdgeDestinationAssignments(), storm::exceptions::InvalidSettingsException, "The explicit next-state generator currently does not support transient edge destination assignments.");
// Create all synchronization-related information, e.g. the automata that are put in parallel.
this->createSynchronizationInformation();
@ -358,7 +367,7 @@ namespace storm {
uint64_t currentLocationIndex = locations[automatonIndex];
storm::jani::Location const& location = automaton.getLocation(currentLocationIndex);
auto valueIt = stateRewards.begin();
performTransientAssignments(location.getAssignments().getTransientAssignments(), [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
performTransientAssignments(location.getAssignments().getTransientAssignments(), *this->evaluator, [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
++automatonIndex;
}
result.addStateRewards(std::move(stateRewards));
@ -446,6 +455,7 @@ namespace storm {
}
Choice<ValueType> choice(edge.getActionIndex(), static_cast<bool>(exitRate));
std::vector<ValueType> stateActionRewards(rewardVariables.size(), storm::utility::zero<ValueType>());
// Iterate over all updates of the current command.
ValueType probabilitySum = storm::utility::zero<ValueType>();
@ -457,13 +467,26 @@ namespace storm {
// seen, we also add it to the set of states that have yet to be explored.
uint64_t assignmentLevel = edge.getLowestAssignmentLevel(); // Might be the largest possible integer, if there is no assignment
uint64_t const& highestLevel = edge.getHighestAssignmentLevel();
bool hasTransientRewardAssignments = destination.hasTransientAssignment();
CompressedState newState = applyUpdate(state, destination, this->variableInformation.locationVariables[automatonIndex], assignmentLevel, *this->evaluator);
if (hasTransientRewardAssignments) {
STORM_LOG_ASSERT(this->options.isScaleAndLiftTransitionRewardsSet(), "Transition rewards are not supported and scaling to action rewards is disabled.");
// Create the rewards for this destination
auto valueIt = stateActionRewards.begin();
performTransientAssignments(destination.getOrderedAssignments().getTransientAssignments(assignmentLevel), *this->evaluator, [&] (ValueType const& value) { *valueIt += (value * probability); ++valueIt; } );
}
while (assignmentLevel < highestLevel) {
++assignmentLevel;
auto nextLevelEvaluator = storm::expressions::ExpressionEvaluator<ValueType>(model.getManager());
unpackStateIntoEvaluator(newState, this->variableInformation, nextLevelEvaluator);
newState = applyUpdate(newState, destination, this->variableInformation.locationVariables[automatonIndex], assignmentLevel, nextLevelEvaluator);
auto currentLevelEvaluator = storm::expressions::ExpressionEvaluator<ValueType>(model.getManager());
unpackStateIntoEvaluator(newState, this->variableInformation, currentLevelEvaluator);
newState = applyUpdate(newState, destination, this->variableInformation.locationVariables[automatonIndex], assignmentLevel, currentLevelEvaluator);
if (hasTransientRewardAssignments) {
// update the rewards for this destination
auto valueIt = stateActionRewards.begin();
performTransientAssignments(destination.getOrderedAssignments().getTransientAssignments(assignmentLevel), currentLevelEvaluator, [&] (ValueType const& value) { *valueIt += (value * probability); ++valueIt; } );
}
}
StateType stateIndex = stateToIdCallback(newState);
// Update the choice by adding the probability/target state to it.
@ -476,8 +499,10 @@ namespace storm {
}
}
// Create the state-action reward for the newly created choice.
performTransientAssignments(edge.getAssignments().getTransientAssignments(), [&choice] (ValueType const& value) { choice.addReward(value); } );
// Create the state-action reward for the newly created choice. Note that edge assignments are all transient and we assume that no transient variables occur on the rhs of transient assignments, i.e., the assignment level does not matter here
auto valueIt = stateActionRewards.begin();
performTransientAssignments(edge.getAssignments().getTransientAssignments(), *this->evaluator, [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
choice.addRewards(std::move(stateActionRewards));
if (this->options.isExplorationChecksSet()) {
// Check that the resulting distribution is in fact a distribution.
@ -531,13 +556,21 @@ namespace storm {
storm::jani::Edge const& edge = *indexAndEdge.second;
for (auto const& destination : edge.getDestinations()) {
ValueType destinationProbability = this->evaluator->asRational(destination.getProbability());
if (!storm::utility::isZero(destinationProbability)) {
if (destination.hasTransientAssignment()) {
STORM_LOG_ASSERT(this->options.isScaleAndLiftTransitionRewardsSet(), "Transition rewards are not supported and scaling to action rewards is disabled.");
// add the rewards for this destination
auto valueIt = stateActionRewards.begin();
performTransientAssignments(edge.getAssignments().getTransientAssignments(), *this->evaluator, [&] (ValueType const& value) { *valueIt += (value * destinationProbability); ++valueIt; } );
}
for (auto const& stateProbability : currentDistribution) {
// Compute the new state under the current update and add it to the set of new target states.
CompressedState newTargetState = applyUpdate(stateProbability.getState(), destination, this->variableInformation.locationVariables[edgeCombination[i].first], assignmentLevel, *this->evaluator);
// If the new state was already found as a successor state, update the probability
// and otherwise insert it.
ValueType probability = stateProbability.getValue() * this->evaluator->asRational(destination.getProbability());
// and insert it.
ValueType probability = destinationProbability * stateProbability.getValue();
if (edge.hasRate()) {
probability *= this->evaluator->asRational(edge.getRate());
}
@ -546,10 +579,7 @@ namespace storm {
}
}
}
// Create the state-action reward for the newly created choice.
auto valueIt = stateActionRewards.begin();
performTransientAssignments(edge.getAssignments().getTransientAssignments(), [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
}
nextDistribution.compress();
@ -562,13 +592,18 @@ namespace storm {
// If there are different assignment levels, we need to expand the possible destinations which causes an exponential blowup...
uint64_t destinationId = 0;
bool lastDestinationId = false;
do {
// First assignment level
std::vector<ValueType> destinationRewards;
std::vector<storm::jani::EdgeDestination const*> destinations;
std::vector<LocationVariableInformation const*> locationVars;
destinations.reserve(iteratorList.size());
locationVars.reserve(iteratorList.size());
do {
// First assignment level
destinations.clear();
locationVars.clear();
destinationRewards.assign(destinationRewards.size(), storm::utility::zero<ValueType>());
CompressedState successorState = state;
ValueType successorProbability = storm::utility::one<ValueType>();
@ -591,10 +626,12 @@ namespace storm {
if (storm::utility::isZero(successorProbability)) {
break;
}
// Create the state-action reward for the newly created choice.
auto valueIt = stateActionRewards.begin();
performTransientAssignments(edge.getAssignments().getTransientAssignments(), [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
successorState = applyUpdate(successorState, *destinations.back(), *locationVars.back(), assignmentLevel, *this->evaluator);
// add the reward for this destination to the destination rewards
auto valueIt = destinationRewards.begin();
performTransientAssignments(destinations.back()->getOrderedAssignments().getTransientAssignments(assignmentLevel), *this->evaluator, [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
}
if (!storm::utility::isZero(successorProbability)) {
@ -606,17 +643,27 @@ namespace storm {
auto locationVarIt = locationVars.begin();
for (auto const& destPtr : destinations) {
successorState = applyUpdate(successorState, *destPtr, **locationVarIt, assignmentLevel, currentLevelEvaluator);
// add the reward for this destination to the destination rewards
auto valueIt = destinationRewards.begin();
performTransientAssignments(destinations.back()->getOrderedAssignments().getTransientAssignments(assignmentLevel), currentLevelEvaluator, [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
++locationVarIt;
}
}
nextDistribution.add(successorState, successorProbability);
storm::utility::vector::addScaledVector(stateActionRewards, destinationRewards, successorProbability);
}
++destinationId;
} while (!lastDestinationId);
std::cout << std::endl;
}
for (uint_fast64_t i = 0; i < iteratorList.size(); ++i) {
storm::jani::Edge const& edge = *iteratorList[i]->second;
// Add the state-action reward for the newly created choice. Note that edge assignments are all transient and we assume that no transient variables occur on the rhs of transient assignments, i.e., the assignment level does not matter here
auto valueIt = stateActionRewards.begin();
performTransientAssignments(edge.getAssignments().getTransientAssignments(), *this->evaluator, [&valueIt] (ValueType const& value) { *valueIt += value; ++valueIt; } );
}
// At this point, we applied all commands of the current command combination and newTargetStates
// contains all target states and their respective probabilities. That means we are now ready to
// add the choice to the list of transitions.
@ -792,7 +839,7 @@ namespace storm {
}
template<typename ValueType, typename StateType>
void JaniNextStateGenerator<ValueType, StateType>::performTransientAssignments(storm::jani::detail::ConstAssignments const& transientAssignments, std::function<void (ValueType const&)> const& callback) {
void JaniNextStateGenerator<ValueType, StateType>::performTransientAssignments(storm::jani::detail::ConstAssignments const& transientAssignments, storm::expressions::ExpressionEvaluator<ValueType> const& expressionEvaluator, std::function<void (ValueType const&)> const& callback) {
// If there are no reward variables, there is no need to iterate at all.
if (rewardVariables.empty()) {
return;
@ -810,7 +857,7 @@ namespace storm {
if (rewardVariableIt == rewardVariableIte) {
break;
} else if (*rewardVariableIt == assignment.getExpressionVariable()) {
callback(ValueType(this->evaluator->asRational(assignment.getAssignedExpression())));
callback(ValueType(expressionEvaluator.asRational(assignment.getAssignedExpression())));
++rewardVariableIt;
}
}

2
src/storm/generator/JaniNextStateGenerator.h
View File

@ -105,7 +105,7 @@ namespace storm {
* Treats the given transient assignments by calling the callback function whenever a transient assignment
* to one of the reward variables of this generator is performed.
*/
void performTransientAssignments(storm::jani::detail::ConstAssignments const& transientAssignments, std::function<void (ValueType const&)> const& callback);
void performTransientAssignments(storm::jani::detail::ConstAssignments const& transientAssignments, storm::expressions::ExpressionEvaluator<ValueType> const& expressionEvaluator, std::function<void (ValueType const&)> const& callback);
/*!
* Builds the information structs for the reward models.

52
src/storm/storage/jani/OrderedAssignments.cpp
View File

@ -81,18 +81,10 @@ namespace storm {
}
bool OrderedAssignments::hasMultipleLevels(bool onlyTransient) const {
if (allAssignments.empty()) {
if ((onlyTransient ? transientAssignments : allAssignments).empty()) {
return false;
}
if (onlyTransient) {
for (auto const& a : allAssignments) {
if (a->isTransient()) {
return a->getLevel() != 0 || getHighestLevel(true) != 0;
}
}
return false; // no transient assignments
}
return getLowestLevel(false) != 0 || getHighestLevel(false) != 0;
return getLowestLevel(onlyTransient) != 0 || getHighestLevel(onlyTransient) != 0;
}
bool OrderedAssignments::empty() const {
@ -110,31 +102,15 @@ namespace storm {
}
int_fast64_t OrderedAssignments::getLowestLevel(bool onlyTransient) const {
assert(!allAssignments.empty());
if (onlyTransient) {
for (auto const& a : allAssignments) {
if (a->getLValue().isTransient()) {
return a->getLevel();
}
}
// assert that at least one transient variable is contained.
assert(false);
}
return allAssignments.front()->getLevel();
auto const& as = onlyTransient ? transientAssignments : allAssignments;
assert(!as.empty());
return as.front()->getLevel();
}
int_fast64_t OrderedAssignments::getHighestLevel(bool onlyTransient) const {
assert(!allAssignments.empty());
if (onlyTransient) {
for (auto aIt = allAssignments.rbegin(); aIt != allAssignments.rend(); ++aIt) {
if ((*aIt)->getLValue().isTransient()) {
return (*aIt)->getLevel();
}
}
// assert that at least one transient variable is contained.
assert(false);
}
return allAssignments.back()->getLevel();
auto const& as = onlyTransient ? transientAssignments : allAssignments;
assert(!as.empty());
return as.back()->getLevel();
}
bool OrderedAssignments::contains(Assignment const& assignment) const {
@ -222,6 +198,18 @@ namespace storm {
return detail::ConstAssignments(nonTransientAssignments.begin(), nonTransientAssignments.end());
}
detail::ConstAssignments OrderedAssignments::getTransientAssignments(int_fast64_t assignmentLevel) const {
auto begin = transientAssignments.begin();
while (begin != transientAssignments.end() && (*begin)->getLevel() < assignmentLevel) {
++begin;
}
auto end = begin;
while (end != transientAssignments.end() && (*begin)->getLevel() == assignmentLevel) {
++end;
}
return detail::ConstAssignments(begin, end);
}
bool OrderedAssignments::hasTransientAssignment() const {
return !transientAssignments.empty();
}

5
src/storm/storage/jani/OrderedAssignments.h
View File

@ -96,6 +96,11 @@ namespace storm {
*/
detail::ConstAssignments getTransientAssignments() const;
/*!
* Returns all transient assignments in this set of assignments.
*/
detail::ConstAssignments getTransientAssignments(int_fast64_t assignmentLevel) const;
/*!
* Returns all non-transient assignments in this set of assignments.
*/

Write Preview
Loading…
Cancel
Save