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

better modularity for multi-objective preprocessing

main
TimQu 8 years ago
parent
e4561a70e5
commit
4c0bda2664
11 changed files with 1063 additions and 958 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. 611
      src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessor.cpp
  2. 105
      src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessor.h
  3. 105
      src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessorResult.h
  4. 137
      src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessorTask.h
  5. 103
      src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveMemoryIncorporation.cpp
  6. 33
      src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveMemoryIncorporation.h
  7. 424
      src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessor.cpp
  8. 105
      src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessor.h
  9. 95
      src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessorResult.h
  10. 230
      src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveRewardAnalysis.cpp
  11. 73
      src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveRewardAnalysis.h

611
src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessor.cpp
View File

@ -1,611 +0,0 @@
#include "storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessor.h"
#include <algorithm>
#include <set>
#include "storm/models/sparse/Mdp.h"
#include "storm/models/sparse/MarkovAutomaton.h"
#include "storm/models/sparse/StandardRewardModel.h"
#include "storm/modelchecker/propositional/SparsePropositionalModelChecker.h"
#include "storm/modelchecker/prctl/helper/BaierUpperRewardBoundsComputer.h"
#include "storm/modelchecker/results/ExplicitQualitativeCheckResult.h"
#include "storm/storage/MaximalEndComponentDecomposition.h"
#include "storm/storage/memorystructure/MemoryStructureBuilder.h"
#include "storm/storage/memorystructure/SparseModelMemoryProduct.h"
#include "storm/storage/expressions/ExpressionManager.h"
#include "storm/transformer/EndComponentEliminator.h"
#include "storm/utility/macros.h"
#include "storm/utility/vector.h"
#include "storm/utility/graph.h"
#include "storm/settings/SettingsManager.h"
#include "storm/settings/modules/GeneralSettings.h"
#include "storm/exceptions/InvalidPropertyException.h"
#include "storm/exceptions/UnexpectedException.h"
#include "storm/exceptions/NotImplementedException.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
template<typename SparseModelType>
typename SparseMultiObjectivePreprocessor<SparseModelType>::ReturnType SparseMultiObjectivePreprocessor<SparseModelType>::preprocess(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula) {
PreprocessorData data(originalModel);
//Invoke preprocessing on the individual objectives
for (auto const& subFormula : originalFormula.getSubformulas()) {
STORM_LOG_INFO("Preprocessing objective " << *subFormula<< ".");
data.objectives.push_back(std::make_shared<Objective<ValueType>>());
data.objectives.back()->originalFormula = subFormula;
data.finiteRewardCheckObjectives.resize(data.objectives.size(), false);
data.upperResultBoundObjectives.resize(data.objectives.size(), false);
if (data.objectives.back()->originalFormula->isOperatorFormula()) {
preprocessOperatorFormula(data.objectives.back()->originalFormula->asOperatorFormula(), data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Could not preprocess the subformula " << *subFormula << " of " << originalFormula << " because it is not supported");
}
}
// Incorporate the required memory into the state space
storm::storage::SparseModelMemoryProduct<ValueType> product = data.memory->product(originalModel);
std::shared_ptr<SparseModelType> preprocessedModel = std::dynamic_pointer_cast<SparseModelType>(product.build());
auto backwardTransitions = preprocessedModel->getBackwardTransitions();
// compute the end components of the model (if required)
bool endComponentAnalysisRequired = false;
for (auto& task : data.tasks) {
endComponentAnalysisRequired = endComponentAnalysisRequired || task->requiresEndComponentAnalysis();
}
if (endComponentAnalysisRequired) {
// TODO
STORM_LOG_THROW(false, storm::exceptions::NotImplementedException, "End component analysis required but currently not implemented.");
}
for (auto& task : data.tasks) {
task->perform(*preprocessedModel);
}
// Remove reward models that are not needed anymore
std::set<std::string> relevantRewardModels;
for (auto const& obj : data.objectives) {
obj->formula->gatherReferencedRewardModels(relevantRewardModels);
}
preprocessedModel->restrictRewardModels(relevantRewardModels);
// Build the actual result
return buildResult(originalModel, originalFormula, data, preprocessedModel, backwardTransitions);
}
template <typename SparseModelType>
SparseMultiObjectivePreprocessor<SparseModelType>::PreprocessorData::PreprocessorData(SparseModelType const& model) : originalModel(model) {
memory = std::make_shared<storm::storage::MemoryStructure>(storm::storage::MemoryStructureBuilder<ValueType, RewardModelType>::buildTrivialMemoryStructure(model));
// The memoryLabelPrefix should not be a prefix of a state label of the given model to ensure uniqueness of label names
memoryLabelPrefix = "mem";
while (true) {
bool prefixIsUnique = true;
for (auto const& label : originalModel.getStateLabeling().getLabels()) {
if (memoryLabelPrefix.size() <= label.size()) {
if (std::mismatch(memoryLabelPrefix.begin(), memoryLabelPrefix.end(), label.begin()).first == memoryLabelPrefix.end()) {
prefixIsUnique = false;
memoryLabelPrefix = "_" + memoryLabelPrefix;
break;
}
}
}
if (prefixIsUnique) {
break;
}
}
// The rewardModelNamePrefix should not be a prefix of a reward model name of the given model to ensure uniqueness of reward model names
rewardModelNamePrefix = "obj";
while (true) {
bool prefixIsUnique = true;
for (auto const& label : originalModel.getStateLabeling().getLabels()) {
if (memoryLabelPrefix.size() <= label.size()) {
if (std::mismatch(memoryLabelPrefix.begin(), memoryLabelPrefix.end(), label.begin()).first == memoryLabelPrefix.end()) {
prefixIsUnique = false;
memoryLabelPrefix = "_" + memoryLabelPrefix;
break;
}
}
}
if (prefixIsUnique) {
break;
}
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessOperatorFormula(storm::logic::OperatorFormula const& formula, PreprocessorData& data) {
Objective<ValueType>& objective = *data.objectives.back();
// Check whether the complementary event is considered
objective.considersComplementaryEvent = formula.isProbabilityOperatorFormula() && formula.getSubformula().isGloballyFormula();
storm::logic::OperatorInformation opInfo;
if (formula.hasBound()) {
opInfo.bound = formula.getBound();
// Invert the bound (if necessary)
if (objective.considersComplementaryEvent) {
opInfo.bound->threshold = opInfo.bound->threshold.getManager().rational(storm::utility::one<storm::RationalNumber>()) - opInfo.bound->threshold;
switch (opInfo.bound->comparisonType) {
case storm::logic::ComparisonType::Greater:
opInfo.bound->comparisonType = storm::logic::ComparisonType::Less;
break;
case storm::logic::ComparisonType::GreaterEqual:
opInfo.bound->comparisonType = storm::logic::ComparisonType::LessEqual;
break;
case storm::logic::ComparisonType::Less:
opInfo.bound->comparisonType = storm::logic::ComparisonType::Greater;
break;
case storm::logic::ComparisonType::LessEqual:
opInfo.bound->comparisonType = storm::logic::ComparisonType::GreaterEqual;
break;
default:
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Current objective " << formula << " has unexpected comparison type");
}
}
if (storm::logic::isLowerBound(opInfo.bound->comparisonType)) {
opInfo.optimalityType = storm::solver::OptimizationDirection::Maximize;
} else {
opInfo.optimalityType = storm::solver::OptimizationDirection::Minimize;
}
STORM_LOG_WARN_COND(!formula.hasOptimalityType(), "Optimization direction of formula " << formula << " ignored as the formula also specifies a threshold.");
} else if (formula.hasOptimalityType()){
opInfo.optimalityType = formula.getOptimalityType();
// Invert the optimality type (if necessary)
if (objective.considersComplementaryEvent) {
opInfo.optimalityType = storm::solver::invert(opInfo.optimalityType.get());
}
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Objective " << formula << " does not specify whether to minimize or maximize");
}
if (formula.isProbabilityOperatorFormula()){
preprocessProbabilityOperatorFormula(formula.asProbabilityOperatorFormula(), opInfo, data);
} else if (formula.isRewardOperatorFormula()){
preprocessRewardOperatorFormula(formula.asRewardOperatorFormula(), opInfo, data);
} else if (formula.isTimeOperatorFormula()){
preprocessTimeOperatorFormula(formula.asTimeOperatorFormula(), opInfo, data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Could not preprocess the objective " << formula << " because it is not supported");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessProbabilityOperatorFormula(storm::logic::ProbabilityOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// Probabilities are between zero and one
data.objectives.back()->lowerResultBound = storm::utility::zero<ValueType>();
data.objectives.back()->upperResultBound = storm::utility::one<ValueType>();
if (formula.getSubformula().isUntilFormula()){
preprocessUntilFormula(formula.getSubformula().asUntilFormula(), opInfo, data);
} else if (formula.getSubformula().isBoundedUntilFormula()){
preprocessBoundedUntilFormula(formula.getSubformula().asBoundedUntilFormula(), opInfo, data);
} else if (formula.getSubformula().isGloballyFormula()){
preprocessGloballyFormula(formula.getSubformula().asGloballyFormula(), opInfo, data);
} else if (formula.getSubformula().isEventuallyFormula()){
preprocessEventuallyFormula(formula.getSubformula().asEventuallyFormula(), opInfo, data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The subformula of " << formula << " is not supported.");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessRewardOperatorFormula(storm::logic::RewardOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
STORM_LOG_THROW((formula.hasRewardModelName() && data.originalModel.hasRewardModel(formula.getRewardModelName()))
|| (!formula.hasRewardModelName() && data.originalModel.hasUniqueRewardModel()), storm::exceptions::InvalidPropertyException, "The reward model is not unique or the formula " << formula << " does not specify an existing reward model.");
std::string rewardModelName;
if (formula.hasRewardModelName()) {
rewardModelName = formula.getRewardModelName();
STORM_LOG_THROW(data.originalModel.hasRewardModel(rewardModelName), storm::exceptions::InvalidPropertyException, "The reward model specified by formula " << formula << " does not exist in the model");
} else {
STORM_LOG_THROW(data.originalModel.hasUniqueRewardModel(), storm::exceptions::InvalidOperationException, "The formula " << formula << " does not specify a reward model name and the reward model is not unique.");
rewardModelName = data.originalModel.getRewardModels().begin()->first;
}
data.objectives.back()->lowerResultBound = storm::utility::zero<ValueType>();
if (formula.getSubformula().isEventuallyFormula()){
preprocessEventuallyFormula(formula.getSubformula().asEventuallyFormula(), opInfo, data, rewardModelName);
} else if (formula.getSubformula().isCumulativeRewardFormula()) {
preprocessCumulativeRewardFormula(formula.getSubformula().asCumulativeRewardFormula(), opInfo, data, rewardModelName);
} else if (formula.getSubformula().isTotalRewardFormula()) {
preprocessTotalRewardFormula(opInfo, data, rewardModelName);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The subformula of " << formula << " is not supported.");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessTimeOperatorFormula(storm::logic::TimeOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// Time formulas are only supported for Markov automata
STORM_LOG_THROW(data.originalModel.isOfType(storm::models::ModelType::MarkovAutomaton), storm::exceptions::InvalidPropertyException, "Time operator formulas are only supported for Markov automata.");
data.objectives.back()->lowerResultBound = storm::utility::zero<ValueType>();
if (formula.getSubformula().isEventuallyFormula()){
preprocessEventuallyFormula(formula.getSubformula().asEventuallyFormula(), opInfo, data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The subformula of " << formula << " is not supported.");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessUntilFormula(storm::logic::UntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, std::shared_ptr<storm::logic::Formula const> subformula) {
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(data.originalModel);
storm::storage::BitVector rightSubformulaResult = mc.check(formula.getRightSubformula())->asExplicitQualitativeCheckResult().getTruthValuesVector();
storm::storage::BitVector leftSubformulaResult = mc.check(formula.getLeftSubformula())->asExplicitQualitativeCheckResult().getTruthValuesVector();
// Check if the formula is already satisfied in the initial state because then the transformation to expected rewards will fail.
// TODO: Handle this case more properly
STORM_LOG_THROW((data.originalModel.getInitialStates() & rightSubformulaResult).empty(), storm::exceptions::NotImplementedException, "The Probability for the objective " << *data.objectives.back()->originalFormula << " is always one as the rhs of the until formula is true in the initial state. This (trivial) case is currently not implemented.");
// Create a memory structure that stores whether a non-PhiState or a PsiState has already been reached
storm::storage::MemoryStructureBuilder<ValueType, RewardModelType> builder(2, data.originalModel);
std::string relevantStatesLabel = data.memoryLabelPrefix + "_obj" + std::to_string(data.objectives.size()) + "_relevant";
builder.setLabel(0, relevantStatesLabel);
storm::storage::BitVector nonRelevantStates = ~leftSubformulaResult | rightSubformulaResult;
builder.setTransition(0, 0, ~nonRelevantStates);
builder.setTransition(0, 1, nonRelevantStates);
builder.setTransition(1, 1, storm::storage::BitVector(data.originalModel.getNumberOfStates(), true));
for (auto const& initState : data.originalModel.getInitialStates()) {
builder.setInitialMemoryState(initState, nonRelevantStates.get(initState) ? 1 : 0);
}
storm::storage::MemoryStructure objectiveMemory = builder.build();
data.memory = std::make_shared<storm::storage::MemoryStructure>(data.memory->product(objectiveMemory));
std::string rewardModelName = data.rewardModelNamePrefix + std::to_string(data.objectives.size());
if (subformula == nullptr) {
subformula = std::make_shared<storm::logic::TotalRewardFormula>();
}
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(subformula, rewardModelName, opInfo);
auto relevantStatesFormula = std::make_shared<storm::logic::AtomicLabelFormula>(relevantStatesLabel);
data.tasks.push_back(std::make_shared<SparseMultiObjectivePreprocessorReachProbToTotalRewTask<SparseModelType>>(data.objectives.back(), relevantStatesFormula, formula.getRightSubformula().asSharedPointer()));
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessBoundedUntilFormula(storm::logic::BoundedUntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// Check how to handle this query
if (formula.isMultiDimensional() || formula.getTimeBoundReference().isRewardBound()) {
data.objectives.back()->formula = std::make_shared<storm::logic::ProbabilityOperatorFormula>(formula.asSharedPointer(), opInfo);
} else if (!formula.hasLowerBound() || (!formula.isLowerBoundStrict() && storm::utility::isZero(formula.template getLowerBound<storm::RationalNumber>()))) {
std::shared_ptr<storm::logic::Formula const> subformula;
if (!formula.hasUpperBound()) {
// The formula is actually unbounded
subformula = std::make_shared<storm::logic::TotalRewardFormula>();
} else {
STORM_LOG_THROW(!data.originalModel.isOfType(storm::models::ModelType::MarkovAutomaton) || formula.getTimeBoundReference().isTimeBound(), storm::exceptions::InvalidPropertyException, "Bounded until formulas for Markov Automata are only allowed when time bounds are considered.");
storm::logic::TimeBound bound(formula.isUpperBoundStrict(), formula.getUpperBound());
subformula = std::make_shared<storm::logic::CumulativeRewardFormula>(bound, formula.getTimeBoundReference());
}
preprocessUntilFormula(storm::logic::UntilFormula(formula.getLeftSubformula().asSharedPointer(), formula.getRightSubformula().asSharedPointer()), opInfo, data, subformula);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Property " << formula << "is not supported");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessGloballyFormula(storm::logic::GloballyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// The formula is transformed to an until formula for the complementary event.
auto negatedSubformula = std::make_shared<storm::logic::UnaryBooleanStateFormula>(storm::logic::UnaryBooleanStateFormula::OperatorType::Not, formula.getSubformula().asSharedPointer());
preprocessUntilFormula(storm::logic::UntilFormula(storm::logic::Formula::getTrueFormula(), negatedSubformula), opInfo, data);
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessEventuallyFormula(storm::logic::EventuallyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName) {
if (formula.isReachabilityProbabilityFormula()){
preprocessUntilFormula(storm::logic::UntilFormula(storm::logic::Formula::getTrueFormula(), formula.getSubformula().asSharedPointer()), opInfo, data);
return;
}
// Analyze the subformula
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(data.originalModel);
storm::storage::BitVector subFormulaResult = mc.check(formula.getSubformula())->asExplicitQualitativeCheckResult().getTruthValuesVector();
// Create a memory structure that stores whether a target state has already been reached
storm::storage::MemoryStructureBuilder<ValueType, RewardModelType> builder(2, data.originalModel);
// Get a unique label that is not already present in the model
std::string relevantStatesLabel = data.memoryLabelPrefix + "_obj" + std::to_string(data.objectives.size()) + "_relevant";
builder.setLabel(0, relevantStatesLabel);
builder.setTransition(0, 0, ~subFormulaResult);
builder.setTransition(0, 1, subFormulaResult);
builder.setTransition(1, 1, storm::storage::BitVector(data.originalModel.getNumberOfStates(), true));
for (auto const& initState : data.originalModel.getInitialStates()) {
builder.setInitialMemoryState(initState, subFormulaResult.get(initState) ? 1 : 0);
}
storm::storage::MemoryStructure objectiveMemory = builder.build();
data.memory = std::make_shared<storm::storage::MemoryStructure>(data.memory->product(objectiveMemory));
auto relevantStatesFormula = std::make_shared<storm::logic::AtomicLabelFormula>(relevantStatesLabel);
std::string auxRewardModelName = data.rewardModelNamePrefix + std::to_string(data.objectives.size());
auto totalRewardFormula = std::make_shared<storm::logic::TotalRewardFormula>();
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(totalRewardFormula, auxRewardModelName, opInfo);
data.finiteRewardCheckObjectives.set(data.objectives.size() - 1, true);
data.upperResultBoundObjectives.set(data.objectives.size() - 1, true);
if (formula.isReachabilityRewardFormula()) {
assert(optionalRewardModelName.is_initialized());
data.tasks.push_back(std::make_shared<SparseMultiObjectivePreprocessorReachRewToTotalRewTask<SparseModelType>>(data.objectives.back(), relevantStatesFormula, optionalRewardModelName.get()));
} else if (formula.isReachabilityTimeFormula() && data.originalModel.isOfType(storm::models::ModelType::MarkovAutomaton)) {
data.tasks.push_back(std::make_shared<SparseMultiObjectivePreprocessorReachTimeToTotalRewTask<SparseModelType>>(data.objectives.back(), relevantStatesFormula));
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The formula " << formula << " neither considers reachability probabilities nor reachability rewards " << (data.originalModel.isOfType(storm::models::ModelType::MarkovAutomaton) ? "nor reachability time" : "") << ". This is not supported.");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessCumulativeRewardFormula(storm::logic::CumulativeRewardFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName) {
STORM_LOG_THROW(data.originalModel.isOfType(storm::models::ModelType::Mdp), storm::exceptions::InvalidPropertyException, "Cumulative reward formulas are not supported for the given model type.");
auto cumulativeRewardFormula = std::make_shared<storm::logic::CumulativeRewardFormula>(formula);
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(cumulativeRewardFormula, *optionalRewardModelName, opInfo);
bool onlyRewardBounds = true;
for (uint64_t i = 0; i < cumulativeRewardFormula->getDimension(); ++i) {
if (!cumulativeRewardFormula->getTimeBoundReference(i).isRewardBound()) {
onlyRewardBounds = false;
break;
}
}
if (onlyRewardBounds) {
data.finiteRewardCheckObjectives.set(data.objectives.size() - 1, true);
data.upperResultBoundObjectives.set(data.objectives.size() - 1, true);
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessTotalRewardFormula(storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName) {
auto totalRewardFormula = std::make_shared<storm::logic::TotalRewardFormula>();
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(totalRewardFormula, *optionalRewardModelName, opInfo);
data.finiteRewardCheckObjectives.set(data.objectives.size() - 1, true);
data.upperResultBoundObjectives.set(data.objectives.size() - 1, true);
}
template<typename SparseModelType>
typename SparseMultiObjectivePreprocessor<SparseModelType>::ReturnType SparseMultiObjectivePreprocessor<SparseModelType>::buildResult(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula, PreprocessorData& data, std::shared_ptr<SparseModelType> const& preprocessedModel, storm::storage::SparseMatrix<ValueType> const& backwardTransitions) {
ReturnType result(originalFormula, originalModel);
result.preprocessedModel = preprocessedModel;
for (auto& obj : data.objectives) {
result.objectives.push_back(std::move(*obj));
}
result.queryType = getQueryType(result.objectives);
setReward0States(result, backwardTransitions);
checkRewardFiniteness(result, data.finiteRewardCheckObjectives, backwardTransitions);
// We compute upper result bounds if the 'sound' option has been enabled
if (storm::settings::getModule<storm::settings::modules::GeneralSettings>().isSoundSet()) {
for (auto const& objIndex : data.upperResultBoundObjectives) {
result.objectives[objIndex].upperResultBound = computeUpperResultBound(result, objIndex, backwardTransitions);
}
}
return result;
}
template<typename SparseModelType>
typename SparseMultiObjectivePreprocessor<SparseModelType>::ReturnType::QueryType SparseMultiObjectivePreprocessor<SparseModelType>::getQueryType(std::vector<Objective<ValueType>> const& objectives) {
uint_fast64_t numOfObjectivesWithThreshold = 0;
for (auto& obj : objectives) {
if (obj.formula->hasBound()) {
++numOfObjectivesWithThreshold;
}
}
if (numOfObjectivesWithThreshold == objectives.size()) {
return ReturnType::QueryType::Achievability;
} else if (numOfObjectivesWithThreshold + 1 == objectives.size()) {
// Note: We do not want to consider a Pareto query when the total number of objectives is one.
return ReturnType::QueryType::Quantitative;
} else if (numOfObjectivesWithThreshold == 0) {
return ReturnType::QueryType::Pareto;
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Invalid Multi-objective query: The numer of qualitative objectives should be either 0 (Pareto query), 1 (quantitative query), or #objectives (achievability query).");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::setReward0States(ReturnType& result, storm::storage::SparseMatrix<ValueType> const& backwardTransitions) {
uint_fast64_t stateCount = result.preprocessedModel->getNumberOfStates();
auto const& transitions = result.preprocessedModel->getTransitionMatrix();
std::vector<uint_fast64_t> const& groupIndices = transitions.getRowGroupIndices();
storm::storage::BitVector allStates(stateCount, true);
// Get the choices that yield non-zero reward
storm::storage::BitVector zeroRewardChoices(result.preprocessedModel->getNumberOfChoices(), true);
for (auto const& obj : result.objectives) {
if (obj.formula->isRewardOperatorFormula()) {
auto const& rewModel = result.preprocessedModel->getRewardModel(obj.formula->asRewardOperatorFormula().getRewardModelName());
zeroRewardChoices &= rewModel.getChoicesWithZeroReward(transitions);
}
}
// Get the states that have reward for at least one (or for all) choices assigned to it.
storm::storage::BitVector statesWithRewardForOneChoice = storm::storage::BitVector(stateCount, false);
storm::storage::BitVector statesWithRewardForAllChoices = storm::storage::BitVector(stateCount, true);
for (uint_fast64_t state = 0; state < stateCount; ++state) {
bool stateHasChoiceWithReward = false;
bool stateHasChoiceWithoutReward = false;
uint_fast64_t const& groupEnd = groupIndices[state + 1];
for (uint_fast64_t choice = groupIndices[state]; choice < groupEnd; ++choice) {
if (zeroRewardChoices.get(choice)) {
stateHasChoiceWithoutReward = true;
} else {
stateHasChoiceWithReward = true;
}
}
if (stateHasChoiceWithReward) {
statesWithRewardForOneChoice.set(state, true);
}
if (stateHasChoiceWithoutReward) {
statesWithRewardForAllChoices.set(state, false);
}
}
// get the states for which there is a scheduler yielding reward zero
result.reward0EStates = storm::utility::graph::performProbGreater0A(transitions, groupIndices, backwardTransitions, allStates, statesWithRewardForAllChoices, false, 0, zeroRewardChoices);
result.reward0EStates.complement();
result.reward0AStates = storm::utility::graph::performProb0A(backwardTransitions, allStates, statesWithRewardForOneChoice);
assert(result.reward0AStates.isSubsetOf(result.reward0EStates));
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::checkRewardFiniteness(ReturnType& result, storm::storage::BitVector const& finiteRewardCheckObjectives, storm::storage::SparseMatrix<ValueType> const& backwardTransitions) {
result.rewardFinitenessType = ReturnType::RewardFinitenessType::AllFinite;
auto const& transitions = result.preprocessedModel->getTransitionMatrix();
std::vector<uint_fast64_t> const& groupIndices = transitions.getRowGroupIndices();
storm::storage::BitVector maxRewardsToCheck(result.preprocessedModel->getNumberOfChoices(), true);
storm::storage::BitVector minRewardsToCheck(result.preprocessedModel->getNumberOfChoices(), true);
for (auto const& objIndex : finiteRewardCheckObjectives) {
STORM_LOG_ASSERT(result.objectives[objIndex].formula->isRewardOperatorFormula(), "Objective needs to be checked for finite reward but has no reward operator.");
auto const& rewModel = result.preprocessedModel->getRewardModel(result.objectives[objIndex].formula->asRewardOperatorFormula().getRewardModelName());
auto unrelevantChoices = rewModel.getChoicesWithZeroReward(transitions);
// For (upper) reward bounded cumulative reward formulas, we do not need to consider the choices where boundReward is collected.
if (result.objectives[objIndex].formula->getSubformula().isCumulativeRewardFormula()) {
auto const& timeBoundReference = result.objectives[objIndex].formula->getSubformula().asCumulativeRewardFormula().getTimeBoundReference();
// Only reward bounded formulas need a finiteness check
assert(timeBoundReference.isRewardBound());
auto const& rewModelOfBound = result.preprocessedModel->getRewardModel(timeBoundReference.getRewardName());
unrelevantChoices |= ~rewModelOfBound.getChoicesWithZeroReward(transitions);
}
if (storm::solver::minimize(result.objectives[objIndex].formula->getOptimalityType())) {
minRewardsToCheck &= unrelevantChoices;
} else {
maxRewardsToCheck &= unrelevantChoices;
}
}
maxRewardsToCheck.complement();
minRewardsToCheck.complement();
// Check reward finiteness under all schedulers
storm::storage::BitVector allStates(result.preprocessedModel->getNumberOfStates(), true);
if (storm::utility::graph::checkIfECWithChoiceExists(transitions, backwardTransitions, allStates, maxRewardsToCheck | minRewardsToCheck)) {
// Check whether there is a scheduler yielding infinite reward for a maximizing objective
if (storm::utility::graph::checkIfECWithChoiceExists(transitions, backwardTransitions, allStates, maxRewardsToCheck)) {
result.rewardFinitenessType = ReturnType::RewardFinitenessType::Infinite;
} else {
// Check whether there is a scheduler under which all rewards are finite.
result.rewardLessInfinityEStates = storm::utility::graph::performProb1E(transitions, groupIndices, backwardTransitions, allStates, result.reward0EStates);
if ((result.rewardLessInfinityEStates.get() & result.preprocessedModel->getInitialStates()).empty()) {
// There is no scheduler that induces finite reward for the initial state
result.rewardFinitenessType = ReturnType::RewardFinitenessType::Infinite;
} else {
result.rewardFinitenessType = ReturnType::RewardFinitenessType::ExistsParetoFinite;
}
}
} else {
result.rewardLessInfinityEStates = allStates;
}
}
template<typename SparseModelType>
boost::optional<typename SparseModelType::ValueType> SparseMultiObjectivePreprocessor<SparseModelType>::computeUpperResultBound(ReturnType const& result, uint64_t objIndex, storm::storage::SparseMatrix<ValueType> const& backwardTransitions) {
boost::optional<ValueType> upperBound;
if (!result.originalModel.isOfType(storm::models::ModelType::Mdp)) {
return upperBound;
}
auto const& transitions = result.preprocessedModel->getTransitionMatrix();
if (result.objectives[objIndex].formula->isRewardOperatorFormula()) {
auto const& rewModel = result.preprocessedModel->getRewardModel(result.objectives[objIndex].formula->asRewardOperatorFormula().getRewardModelName());
auto actionRewards = rewModel.getTotalRewardVector(transitions);
if (result.objectives[objIndex].formula->getSubformula().isTotalRewardFormula() || result.objectives[objIndex].formula->getSubformula().isCumulativeRewardFormula()) {
// We have to eliminate ECs here to treat zero-reward ECs
storm::storage::BitVector allStates(result.preprocessedModel->getNumberOfStates(), true);
// Get the set of states from which no reward is reachable
auto nonZeroRewardStates = rewModel.getStatesWithZeroReward(transitions);
nonZeroRewardStates.complement();
auto expRewGreater0EStates = storm::utility::graph::performProbGreater0E(backwardTransitions, allStates, nonZeroRewardStates);
auto zeroRewardChoices = rewModel.getChoicesWithZeroReward(transitions);
auto ecElimRes = storm::transformer::EndComponentEliminator<ValueType>::transform(transitions, expRewGreater0EStates, zeroRewardChoices, ~allStates);
allStates.resize(ecElimRes.matrix.getRowGroupCount());
storm::storage::BitVector outStates(allStates.size(), false);
std::vector<ValueType> rew0StateProbs;
rew0StateProbs.reserve(ecElimRes.matrix.getRowCount());
for (uint64_t state = 0; state < allStates.size(); ++ state) {
for (uint64_t choice = ecElimRes.matrix.getRowGroupIndices()[state]; choice < ecElimRes.matrix.getRowGroupIndices()[state + 1]; ++choice) {
// Check whether the choice lead to a state with expRew 0 in the original model
bool isOutChoice = false;
uint64_t originalModelChoice = ecElimRes.newToOldRowMapping[choice];
for (auto const& entry : transitions.getRow(originalModelChoice)) {
if (!expRewGreater0EStates.get(entry.getColumn())) {
isOutChoice = true;
outStates.set(state, true);
rew0StateProbs.push_back(storm::utility::one<ValueType>() - ecElimRes.matrix.getRowSum(choice));
assert (!storm::utility::isZero(rew0StateProbs.back()));
break;
}
}
if (!isOutChoice) {
rew0StateProbs.push_back(storm::utility::zero<ValueType>());
}
}
}
// An upper reward bound can only be computed if it is below infinity
if (storm::utility::graph::performProb1A(ecElimRes.matrix, ecElimRes.matrix.getRowGroupIndices(), ecElimRes.matrix.transpose(true), allStates, outStates).full()) {
std::vector<ValueType> rewards;
rewards.reserve(ecElimRes.matrix.getRowCount());
for (auto row : ecElimRes.newToOldRowMapping) {
rewards.push_back(actionRewards[row]);
}
storm::modelchecker::helper::BaierUpperRewardBoundsComputer<ValueType> baier(ecElimRes.matrix, rewards, rew0StateProbs);
if (upperBound) {
upperBound = std::min(upperBound.get(), baier.computeUpperBound());
} else {
upperBound = baier.computeUpperBound();
}
}
}
}
if (upperBound) {
STORM_LOG_INFO("Computed upper result bound " << upperBound.get() << " for objective " << *result.objectives[objIndex].formula << ".");
} else {
STORM_LOG_WARN("Could not compute upper result bound for objective " << *result.objectives[objIndex].formula);
}
return upperBound;
}
template class SparseMultiObjectivePreprocessor<storm::models::sparse::Mdp<double>>;
template class SparseMultiObjectivePreprocessor<storm::models::sparse::MarkovAutomaton<double>>;
template class SparseMultiObjectivePreprocessor<storm::models::sparse::Mdp<storm::RationalNumber>>;
template class SparseMultiObjectivePreprocessor<storm::models::sparse::MarkovAutomaton<storm::RationalNumber>>;
}
}
}

105
src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessor.h
View File

@ -1,105 +0,0 @@
#pragma once
#include <memory>
#include <string>
#include "storm/logic/Formulas.h"
#include "storm/storage/BitVector.h"
#include "storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessorResult.h"
#include "storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessorTask.h"
#include "storm/storage/memorystructure/MemoryStructure.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
/*
* This class invokes the necessary preprocessing for the constraint based multi-objective model checking algorithm
*/
template <class SparseModelType>
class SparseMultiObjectivePreprocessor {
public:
typedef typename SparseModelType::ValueType ValueType;
typedef typename SparseModelType::RewardModelType RewardModelType;
typedef SparseMultiObjectivePreprocessorResult<SparseModelType> ReturnType;
/*!
* Preprocesses the given model w.r.t. the given formulas
* @param originalModel The considered model
* @param originalFormula the considered formula. The subformulas should only contain one OperatorFormula at top level.
*/
static ReturnType preprocess(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula);
private:
struct PreprocessorData {
SparseModelType const& originalModel;
std::vector<std::shared_ptr<Objective<ValueType>>> objectives;
std::vector<std::shared_ptr<SparseMultiObjectivePreprocessorTask<SparseModelType>>> tasks;
std::shared_ptr<storm::storage::MemoryStructure> memory;
// Indices of the objectives that require a check for finite reward
storm::storage::BitVector finiteRewardCheckObjectives;
// Indices of the objectives for which we need to compute an upper bound for the result
storm::storage::BitVector upperResultBoundObjectives;
std::string memoryLabelPrefix;
std::string rewardModelNamePrefix;
PreprocessorData(SparseModelType const& model);
};
/*!
* Apply the neccessary preprocessing for the given formula.
* @param formula the current (sub)formula
* @param opInfo the information of the resulting operator formula
* @param data the current data. The currently processed objective is located at data.objectives.back()
* @param optionalRewardModelName the reward model name that is considered for the formula (if available)
*/
static void preprocessOperatorFormula(storm::logic::OperatorFormula const& formula, PreprocessorData& data);
static void preprocessProbabilityOperatorFormula(storm::logic::ProbabilityOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessRewardOperatorFormula(storm::logic::RewardOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessTimeOperatorFormula(storm::logic::TimeOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessUntilFormula(storm::logic::UntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, std::shared_ptr<storm::logic::Formula const> subformula = nullptr);
static void preprocessBoundedUntilFormula(storm::logic::BoundedUntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessGloballyFormula(storm::logic::GloballyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessEventuallyFormula(storm::logic::EventuallyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName = boost::none);
static void preprocessCumulativeRewardFormula(storm::logic::CumulativeRewardFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName = boost::none);
static void preprocessTotalRewardFormula(storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName = boost::none); // The total reward formula itself does not need to be provided as it is unique.
/*!
* Builds the result from preprocessing
*/
static ReturnType buildResult(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula, PreprocessorData& data, std::shared_ptr<SparseModelType> const& preprocessedModel, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
/*!
* Returns the query type
*/
static typename ReturnType::QueryType getQueryType(std::vector<Objective<ValueType>> const& objectives);
/*!
* Computes the set of states that have zero expected reward w.r.t. all expected reward objectives
*/
static void setReward0States(ReturnType& result, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
/*!
* Checks whether the occurring expected rewards are finite and sets the RewardFinitenessType accordingly
* Returns the set of states for which a scheduler exists that induces finite reward for all objectives
*/
static void checkRewardFiniteness(ReturnType& result, storm::storage::BitVector const& finiteRewardCheckObjectives, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
/*!
* Finds an upper bound for the expected reward of the objective with the given index (assuming it considers an expected reward objective)
*/
static boost::optional<ValueType> computeUpperResultBound(ReturnType const& result, uint64_t objIndex, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
};
}
}
}

105
src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessorResult.h
View File

@ -1,105 +0,0 @@
#pragma once
#include <vector>
#include <memory>
#include <boost/optional.hpp>
#include "storm/logic/Formulas.h"
#include "storm/modelchecker/multiobjective/Objective.h"
#include "storm/storage/BitVector.h"
#include "storm/exceptions/UnexpectedException.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
template <class SparseModelType>
struct SparseMultiObjectivePreprocessorResult {
enum class QueryType { Achievability, Quantitative, Pareto };
enum class RewardFinitenessType {
AllFinite, // The expected reward is finite for all objectives and all schedulers
ExistsParetoFinite, // There is a Pareto optimal scheduler yielding finite rewards for all objectives
Infinite // All Pareto optimal schedulers yield infinite reward for at least one objective
};
// Original data
storm::logic::MultiObjectiveFormula const& originalFormula;
SparseModelType const& originalModel;
// The preprocessed model and objectives
std::shared_ptr<SparseModelType> preprocessedModel;
std::vector<Objective<typename SparseModelType::ValueType>> objectives;
// Data about the query
QueryType queryType;
RewardFinitenessType rewardFinitenessType;
// The states of the preprocessed model for which...
storm::storage::BitVector reward0EStates; // ... there is a scheduler such that all expected reward objectives have value zero
storm::storage::BitVector reward0AStates; // ... all schedulers induce value 0 for all expected reward objectives
boost::optional<storm::storage::BitVector> rewardLessInfinityEStates; // ... there is a scheduler yielding finite reward for all expected reward objectives
// Note that other types of objectives (e.g., reward bounded reachability objectives) are not considered.
// Encodes a superset of the set of choices of preprocessedModel that are part of an end component (if given).
//boost::optional<storm::storage::BitVector> ecChoicesHint;
SparseMultiObjectivePreprocessorResult(storm::logic::MultiObjectiveFormula const& originalFormula, SparseModelType const& originalModel) : originalFormula(originalFormula), originalModel(originalModel) {
// Intentionally left empty
}
bool containsOnlyTrivialObjectives() const {
// Trivial objectives are either total reward formulas or single-dimensional step or time bounded cumulative reward formulas
for (auto const& obj : objectives) {
if (obj.formula->isRewardOperatorFormula() && obj.formula->getSubformula().isTotalRewardFormula()) {
continue;
}
if (obj.formula->isRewardOperatorFormula() && obj.formula->getSubformula().isCumulativeRewardFormula()) {
auto const& subf = obj.formula->getSubformula().asCumulativeRewardFormula();
if (!subf.isMultiDimensional() && (subf.getTimeBoundReference().isTimeBound() || subf.getTimeBoundReference().isStepBound())) {
continue;
}
}
// Reaching this point means that the objective is considered as non-trivial
return false;
}
return true;
}
void printToStream(std::ostream& out) const {
out << std::endl;
out << "---------------------------------------------------------------------------------------------------------------------------------------" << std::endl;
out << " Multi-objective Query " << std::endl;
out << "---------------------------------------------------------------------------------------------------------------------------------------" << std::endl;
out << std::endl;
out << "Original Formula: " << std::endl;
out << "--------------------------------------------------------------" << std::endl;
out << "\t" << originalFormula << std::endl;
out << std::endl;
out << "Objectives:" << std::endl;
out << "--------------------------------------------------------------" << std::endl;
for (auto const& obj : objectives) {
obj.printToStream(out);
}
out << "--------------------------------------------------------------" << std::endl;
out << std::endl;
out << "Original Model Information:" << std::endl;
originalModel.printModelInformationToStream(out);
out << std::endl;
out << "Preprocessed Model Information:" << std::endl;
preprocessedModel->printModelInformationToStream(out);
out << std::endl;
out << "---------------------------------------------------------------------------------------------------------------------------------------" << std::endl;
}
friend std::ostream& operator<<(std::ostream& out, SparseMultiObjectivePreprocessorResult<SparseModelType> const& ret) {
ret.printToStream(out);
return out;
}
};
}
}
}

137
src/storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessorTask.h
View File

@ -1,137 +0,0 @@
#pragma once
#include <boost/optional.hpp>
#include <memory>
#include "storm/logic/Formula.h"
#include "storm/logic/Bound.h"
#include "storm/solver/OptimizationDirection.h"
#include "storm/modelchecker/multiobjective/Objective.h"
#include "storm/modelchecker/propositional/SparsePropositionalModelChecker.h"
#include "storm/modelchecker/results/ExplicitQualitativeCheckResult.h"
#include "storm/models/sparse/Mdp.h"
#include "storm/models/sparse/MarkovAutomaton.h"
#include "storm/models/sparse/StandardRewardModel.h"
#include "storm/utility/vector.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
template <typename SparseModelType>
class SparseMultiObjectivePreprocessorTask {
public:
SparseMultiObjectivePreprocessorTask(std::shared_ptr<Objective<typename SparseModelType::ValueType>> const& objective) : objective(objective) {
// intentionally left empty
}
virtual ~SparseMultiObjectivePreprocessorTask() = default;
virtual void perform(SparseModelType& preprocessedModel) const = 0;
virtual bool requiresEndComponentAnalysis() const {
return false;
}
protected:
std::shared_ptr<Objective<typename SparseModelType::ValueType>> objective;
};
// Transforms reachability probabilities to total expected rewards by adding a rewardModel
// such that one reward is given whenever a goal state is reached from a relevant state
template <typename SparseModelType>
class SparseMultiObjectivePreprocessorReachProbToTotalRewTask : public SparseMultiObjectivePreprocessorTask<SparseModelType> {
public:
SparseMultiObjectivePreprocessorReachProbToTotalRewTask(std::shared_ptr<Objective<typename SparseModelType::ValueType>> const& objective, std::shared_ptr<storm::logic::Formula const> const& relevantStateFormula, std::shared_ptr<storm::logic::Formula const> const& goalStateFormula) : SparseMultiObjectivePreprocessorTask<SparseModelType>(objective), relevantStateFormula(relevantStateFormula), goalStateFormula(goalStateFormula) {
// Intentionally left empty
}
virtual void perform(SparseModelType& preprocessedModel) const override {
// build stateAction reward vector that gives (one*transitionProbability) reward whenever a transition leads from a relevantState to a goalState
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(preprocessedModel);
storm::storage::BitVector relevantStates = mc.check(*relevantStateFormula)->asExplicitQualitativeCheckResult().getTruthValuesVector();
storm::storage::BitVector goalStates = mc.check(*goalStateFormula)->asExplicitQualitativeCheckResult().getTruthValuesVector();
std::vector<typename SparseModelType::ValueType> objectiveRewards(preprocessedModel.getTransitionMatrix().getRowCount(), storm::utility::zero<typename SparseModelType::ValueType>());
for (auto const& state : relevantStates) {
for (uint_fast64_t row = preprocessedModel.getTransitionMatrix().getRowGroupIndices()[state]; row < preprocessedModel.getTransitionMatrix().getRowGroupIndices()[state + 1]; ++row) {
objectiveRewards[row] = preprocessedModel.getTransitionMatrix().getConstrainedRowSum(row, goalStates);
}
}
STORM_LOG_ASSERT(this->objective->formula->isRewardOperatorFormula(), "No reward operator formula.");
STORM_LOG_ASSERT(this->objective->formula->asRewardOperatorFormula().hasRewardModelName(), "No reward model name has been specified");
preprocessedModel.addRewardModel(this->objective->formula->asRewardOperatorFormula().getRewardModelName(), typename SparseModelType::RewardModelType(boost::none, std::move(objectiveRewards)));
}
private:
std::shared_ptr<storm::logic::Formula const> relevantStateFormula;
std::shared_ptr<storm::logic::Formula const> goalStateFormula;
};
// Transforms expected reachability rewards to total expected rewards by adding a rewardModel
// such that non-relevant states get reward zero
template <typename SparseModelType>
class SparseMultiObjectivePreprocessorReachRewToTotalRewTask : public SparseMultiObjectivePreprocessorTask<SparseModelType> {
public:
SparseMultiObjectivePreprocessorReachRewToTotalRewTask(std::shared_ptr<Objective<typename SparseModelType::ValueType>> const& objective, std::shared_ptr<storm::logic::Formula const> const& relevantStateFormula, std::string const& originalRewardModelName) : SparseMultiObjectivePreprocessorTask<SparseModelType>(objective), relevantStateFormula(relevantStateFormula), originalRewardModelName(originalRewardModelName) {
// Intentionally left empty
}
virtual void perform(SparseModelType& preprocessedModel) const override {
// build stateAction reward vector that only gives reward for relevant states
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(preprocessedModel);
storm::storage::BitVector nonRelevantStates = ~mc.check(*relevantStateFormula)->asExplicitQualitativeCheckResult().getTruthValuesVector();
typename SparseModelType::RewardModelType objectiveRewards = preprocessedModel.getRewardModel(originalRewardModelName);
objectiveRewards.reduceToStateBasedRewards(preprocessedModel.getTransitionMatrix(), false);
if (objectiveRewards.hasStateRewards()) {
storm::utility::vector::setVectorValues(objectiveRewards.getStateRewardVector(), nonRelevantStates, storm::utility::zero<typename SparseModelType::ValueType>());
}
if (objectiveRewards.hasStateActionRewards()) {
for (auto state : nonRelevantStates) {
std::fill_n(objectiveRewards.getStateActionRewardVector().begin() + preprocessedModel.getTransitionMatrix().getRowGroupIndices()[state], preprocessedModel.getTransitionMatrix().getRowGroupSize(state), storm::utility::zero<typename SparseModelType::ValueType>());
}
}
STORM_LOG_ASSERT(this->objective->formula->isRewardOperatorFormula(), "No reward operator formula.");
STORM_LOG_ASSERT(this->objective->formula->asRewardOperatorFormula().hasRewardModelName(), "No reward model name has been specified");
preprocessedModel.addRewardModel(this->objective->formula->asRewardOperatorFormula().getRewardModelName(), std::move(objectiveRewards));
}
private:
std::shared_ptr<storm::logic::Formula const> relevantStateFormula;
std::string originalRewardModelName;
};
// Transforms expected reachability time to total expected rewards by adding a rewardModel
// such that every time step done from a relevant state yields one reward
template <typename SparseModelType>
class SparseMultiObjectivePreprocessorReachTimeToTotalRewTask : public SparseMultiObjectivePreprocessorTask<SparseModelType> {
public:
SparseMultiObjectivePreprocessorReachTimeToTotalRewTask(std::shared_ptr<Objective<typename SparseModelType::ValueType>> const& objective, std::shared_ptr<storm::logic::Formula const> const& relevantStateFormula) : SparseMultiObjectivePreprocessorTask<SparseModelType>(objective), relevantStateFormula(relevantStateFormula) {
// Intentionally left empty
}
virtual void perform(SparseModelType& preprocessedModel) const override {
// build stateAction reward vector that only gives reward for relevant states
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(preprocessedModel);
storm::storage::BitVector relevantStates = mc.check(*relevantStateFormula)->asExplicitQualitativeCheckResult().getTruthValuesVector();
std::vector<typename SparseModelType::ValueType> timeRewards(preprocessedModel.getNumberOfStates(), storm::utility::zero<typename SparseModelType::ValueType>());
storm::utility::vector::setVectorValues(timeRewards, dynamic_cast<storm::models::sparse::MarkovAutomaton<typename SparseModelType::ValueType> const&>(preprocessedModel).getMarkovianStates() & relevantStates, storm::utility::one<typename SparseModelType::ValueType>());
STORM_LOG_ASSERT(this->objective->formula->isRewardOperatorFormula(), "No reward operator formula.");
STORM_LOG_ASSERT(this->objective->formula->asRewardOperatorFormula().hasRewardModelName(), "No reward model name has been specified");
preprocessedModel.addRewardModel(this->objective->formula->asRewardOperatorFormula().getRewardModelName(), typename SparseModelType::RewardModelType(std::move(timeRewards)));
}
private:
std::shared_ptr<storm::logic::Formula const> relevantStateFormula;
};
}
}
}

103
src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveMemoryIncorporation.cpp
View File

@ -0,0 +1,103 @@
#include "storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveMemoryIncorporation.h"
#include "storm/models/sparse/Mdp.h"
#include "storm/models/sparse/MarkovAutomaton.h"
#include "storm/models/sparse/StandardRewardModel.h"
#include "storm/storage/memorystructure/MemoryStructureBuilder.h"
#include "storm/storage/memorystructure/SparseModelMemoryProduct.h"
#include "storm/logic/Formulas.h"
#include "storm/logic/FragmentSpecification.h"
#include "storm/modelchecker/propositional/SparsePropositionalModelChecker.h"
#include "storm/modelchecker/results/ExplicitQualitativeCheckResult.h"
#include "storm/utility/macros.h"
#include "storm/exceptions/NotImplementedException.h"
#include "storm/exceptions/NotSupportedException.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
namespace preprocessing {
template <class SparseModelType>
storm::storage::MemoryStructure getGoalMemory(SparseModelType const& model, storm::logic::Formula const& propositionalGoalStateFormula) {
STORM_LOG_THROW(propositionalGoalStateFormula.isInFragment(storm::logic::propositional()), storm::exceptions::NotSupportedException, "The subformula " << propositionalGoalStateFormula << " should be propositional.");
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(model);
storm::storage::BitVector goalStates = mc.check(propositionalGoalStateFormula)->asExplicitQualitativeCheckResult().getTruthValuesVector();
// Check if the formula is already satisfied for all initial states. In such a case the trivial memory structure suffices.
if (model.getInitialStates().isSubsetOf(goalStates)) {
STORM_LOG_INFO("One objective is already satisfied for all initial states.");
return storm::storage::MemoryStructureBuilder<typename SparseModelType::ValueType, typename SparseModelType::RewardModelType>::buildTrivialMemoryStructure(model);
}
// Create a memory structure that stores whether a goal state has already been reached
storm::storage::MemoryStructureBuilder<typename SparseModelType::ValueType, typename SparseModelType::RewardModelType> builder(2, model);
builder.setTransition(0, 0, ~goalStates);
builder.setTransition(0, 1, goalStates);
builder.setTransition(1, 1, storm::storage::BitVector(model.getNumberOfStates(), true));
for (auto const& initState : model.getInitialStates()) {
builder.setInitialMemoryState(initState, goalStates.get(initState) ? 1 : 0);
}
return builder.build();
}
template <class SparseModelType>
storm::storage::MemoryStructure getUntilFormulaMemory(SparseModelType const& model, storm::logic::Formula const& leftSubFormula, storm::logic::Formula const& rightSubFormula) {
auto notLeftOrRight = std::make_shared<storm::logic::BinaryBooleanStateFormula>(storm::logic::BinaryBooleanStateFormula::OperatorType::Or,
std::make_shared<storm::logic::UnaryBooleanStateFormula>(storm::logic::UnaryBooleanStateFormula::OperatorType::Not, leftSubFormula.asSharedPointer()),
rightSubFormula.asSharedPointer());
return getGoalMemory<SparseModelType>(model, *notLeftOrRight);
}
template <class SparseModelType>
std::shared_ptr<SparseModelType> SparseMultiObjectiveMemoryIncorporation<SparseModelType>::incorporateGoalMemory(SparseModelType const& model, storm::logic::MultiObjectiveFormula const& formula) {
storm::storage::MemoryStructure memory = storm::storage::MemoryStructureBuilder<ValueType, RewardModelType>::buildTrivialMemoryStructure(model);
for (auto const& subFormula : formula.getSubformulas()) {
STORM_LOG_THROW(subFormula->isOperatorFormula(), storm::exceptions::NotSupportedException, "The given Formula " << *subFormula << " is not supported.");
auto const& subsubFormula = subFormula->asOperatorFormula().getSubformula();
if (subsubFormula.isEventuallyFormula()) {
memory = memory.product(getGoalMemory(model, subsubFormula.asEventuallyFormula().getSubformula()));
} else if (subsubFormula.isUntilFormula()) {
memory = memory.product(getUntilFormulaMemory(model, subsubFormula.asUntilFormula().getLeftSubformula(), subsubFormula.asUntilFormula().getRightSubformula()));
} else if (subsubFormula.isBoundedUntilFormula()) {
// For bounded formulas it is only reasonable to add the goal memory if it considers a single upper step/time bound.
auto const& buf = subsubFormula.asBoundedUntilFormula();
if (!buf.isMultiDimensional() && !buf.getTimeBoundReference().isRewardBound() && (!buf.hasLowerBound() || (!buf.isLowerBoundStrict() && storm::utility::isZero(buf.template getLowerBound<storm::RationalNumber>())))) {
memory = memory.product(getUntilFormulaMemory(model, buf.getLeftSubformula(), buf.getRightSubformula()));
}
} else if (subsubFormula.isGloballyFormula()) {
auto notPhi = std::make_shared<storm::logic::UnaryBooleanStateFormula>(storm::logic::UnaryBooleanStateFormula::OperatorType::Not, subsubFormula.asGloballyFormula().getSubformula().asSharedPointer());
memory = memory.product(getGoalMemory(model, *notPhi));
} else {
STORM_LOG_THROW(subFormula->isOperatorFormula(), storm::exceptions::NotSupportedException, "The given Formula " << subsubFormula << " is not supported.");
}
}
storm::storage::SparseModelMemoryProduct<ValueType> product = memory.product(model);
return std::dynamic_pointer_cast<SparseModelType>(product.build());
}
template <class SparseModelType>
std::shared_ptr<SparseModelType> SparseMultiObjectiveMemoryIncorporation<SparseModelType>::incorporateFullMemory(SparseModelType const& model, uint64_t memoryStates) {
STORM_LOG_THROW(false, storm::exceptions::NotImplementedException, "not implemented");
return nullptr;
}
template class SparseMultiObjectiveMemoryIncorporation<storm::models::sparse::Mdp<double>>;
template class SparseMultiObjectiveMemoryIncorporation<storm::models::sparse::MarkovAutomaton<double>>;
template class SparseMultiObjectiveMemoryIncorporation<storm::models::sparse::Mdp<storm::RationalNumber>>;
template class SparseMultiObjectiveMemoryIncorporation<storm::models::sparse::MarkovAutomaton<storm::RationalNumber>>;
}
}
}
}

33
src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveMemoryIncorporation.h
View File

@ -0,0 +1,33 @@
#pragma once
#include <memory>
#include "storm/storage/memorystructure/MemoryStructure.h"
#include "storm/logic/MultiObjectiveFormula.h"
#include "storm/logic/Formula.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
namespace preprocessing {
template <class SparseModelType>
class SparseMultiObjectiveMemoryIncorporation {
typedef typename SparseModelType::ValueType ValueType;
typedef typename SparseModelType::RewardModelType RewardModelType;
public:
static std::shared_ptr<SparseModelType> incorporateGoalMemory(SparseModelType const& model, storm::logic::MultiObjectiveFormula const& formula);
static std::shared_ptr<SparseModelType> incorporateFullMemory(SparseModelType const& model, uint64_t memoryStates);
};
}
}
}
}

424
src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessor.cpp
View File

@ -0,0 +1,424 @@
#include "storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessor.h"
#include <algorithm>
#include <set>
#include "storm/models/sparse/Mdp.h"
#include "storm/models/sparse/MarkovAutomaton.h"
#include "storm/models/sparse/StandardRewardModel.h"
#include "storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveMemoryIncorporation.h"
#include "storm/modelchecker/propositional/SparsePropositionalModelChecker.h"
#include "storm/modelchecker/prctl/helper/BaierUpperRewardBoundsComputer.h"
#include "storm/modelchecker/results/ExplicitQualitativeCheckResult.h"
#include "storm/storage/MaximalEndComponentDecomposition.h"
#include "storm/storage/expressions/ExpressionManager.h"
#include "storm/transformer/EndComponentEliminator.h"
#include "storm/utility/macros.h"
#include "storm/utility/vector.h"
#include "storm/utility/graph.h"
#include "storm/settings/SettingsManager.h"
#include "storm/settings/modules/GeneralSettings.h"
#include "storm/exceptions/InvalidPropertyException.h"
#include "storm/exceptions/UnexpectedException.h"
#include "storm/exceptions/NotImplementedException.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
namespace preprocessing {
template<typename SparseModelType>
typename SparseMultiObjectivePreprocessor<SparseModelType>::ReturnType SparseMultiObjectivePreprocessor<SparseModelType>::preprocess(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula) {
auto model = SparseMultiObjectiveMemoryIncorporation<SparseModelType>::incorporateGoalMemory(originalModel, originalFormula);
PreprocessorData data(model);
//Invoke preprocessing on the individual objectives
for (auto const& subFormula : originalFormula.getSubformulas()) {
STORM_LOG_INFO("Preprocessing objective " << *subFormula<< ".");
data.objectives.push_back(std::make_shared<Objective<ValueType>>());
data.objectives.back()->originalFormula = subFormula;
data.finiteRewardCheckObjectives.resize(data.objectives.size(), false);
data.upperResultBoundObjectives.resize(data.objectives.size(), false);
if (data.objectives.back()->originalFormula->isOperatorFormula()) {
preprocessOperatorFormula(data.objectives.back()->originalFormula->asOperatorFormula(), data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Could not preprocess the subformula " << *subFormula << " of " << originalFormula << " because it is not supported");
}
}
// Remove reward models that are not needed anymore
std::set<std::string> relevantRewardModels;
for (auto const& obj : data.objectives) {
obj->formula->gatherReferencedRewardModels(relevantRewardModels);
}
data.model->restrictRewardModels(relevantRewardModels);
// Build the actual result
return buildResult(originalModel, originalFormula, data);
}
template <typename SparseModelType>
SparseMultiObjectivePreprocessor<SparseModelType>::PreprocessorData::PreprocessorData(std::shared_ptr<SparseModelType> model) : model(model) {
// The rewardModelNamePrefix should not be a prefix of a reward model name of the given model to ensure uniqueness of new reward model names
rewardModelNamePrefix = "obj";
while (true) {
bool prefixIsUnique = true;
for (auto const& rewardModels : model->getRewardModels()) {
if (rewardModelNamePrefix.size() <= rewardModels.first.size()) {
if (std::mismatch(rewardModelNamePrefix.begin(), rewardModelNamePrefix.end(), rewardModels.first.begin()).first == rewardModelNamePrefix.end()) {
prefixIsUnique = false;
rewardModelNamePrefix = "_" + rewardModelNamePrefix;
break;
}
}
}
if (prefixIsUnique) {
break;
}
}
}
storm::logic::OperatorInformation getOperatorInformation(storm::logic::OperatorFormula const& formula, bool considerComplementaryEvent) {
storm::logic::OperatorInformation opInfo;
if (formula.hasBound()) {
opInfo.bound = formula.getBound();
// Invert the bound (if necessary)
if (considerComplementaryEvent) {
opInfo.bound->threshold = opInfo.bound->threshold.getManager().rational(storm::utility::one<storm::RationalNumber>()) - opInfo.bound->threshold;
switch (opInfo.bound->comparisonType) {
case storm::logic::ComparisonType::Greater:
opInfo.bound->comparisonType = storm::logic::ComparisonType::Less;
break;
case storm::logic::ComparisonType::GreaterEqual:
opInfo.bound->comparisonType = storm::logic::ComparisonType::LessEqual;
break;
case storm::logic::ComparisonType::Less:
opInfo.bound->comparisonType = storm::logic::ComparisonType::Greater;
break;
case storm::logic::ComparisonType::LessEqual:
opInfo.bound->comparisonType = storm::logic::ComparisonType::GreaterEqual;
break;
default:
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Current objective " << formula << " has unexpected comparison type");
}
}
if (storm::logic::isLowerBound(opInfo.bound->comparisonType)) {
opInfo.optimalityType = storm::solver::OptimizationDirection::Maximize;
} else {
opInfo.optimalityType = storm::solver::OptimizationDirection::Minimize;
}
STORM_LOG_WARN_COND(!formula.hasOptimalityType(), "Optimization direction of formula " << formula << " ignored as the formula also specifies a threshold.");
} else if (formula.hasOptimalityType()){
opInfo.optimalityType = formula.getOptimalityType();
// Invert the optimality type (if necessary)
if (considerComplementaryEvent) {
opInfo.optimalityType = storm::solver::invert(opInfo.optimalityType.get());
}
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Objective " << formula << " does not specify whether to minimize or maximize");
}
return opInfo;
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessOperatorFormula(storm::logic::OperatorFormula const& formula, PreprocessorData& data) {
Objective<ValueType>& objective = *data.objectives.back();
// Check whether the complementary event is considered
objective.considersComplementaryEvent = formula.isProbabilityOperatorFormula() && formula.getSubformula().isGloballyFormula();
// Extract the operator information from the formula and potentially invert it for the complementary event
storm::logic::OperatorInformation opInfo = getOperatorInformation(formula, objective.considersComplementaryEvent);
if (formula.isProbabilityOperatorFormula()){
preprocessProbabilityOperatorFormula(formula.asProbabilityOperatorFormula(), opInfo, data);
} else if (formula.isRewardOperatorFormula()){
preprocessRewardOperatorFormula(formula.asRewardOperatorFormula(), opInfo, data);
} else if (formula.isTimeOperatorFormula()){
preprocessTimeOperatorFormula(formula.asTimeOperatorFormula(), opInfo, data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Could not preprocess the objective " << formula << " because it is not supported");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessProbabilityOperatorFormula(storm::logic::ProbabilityOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// Probabilities are between zero and one
data.objectives.back()->lowerResultBound = storm::utility::zero<ValueType>();
data.objectives.back()->upperResultBound = storm::utility::one<ValueType>();
if (formula.getSubformula().isUntilFormula()){
preprocessUntilFormula(formula.getSubformula().asUntilFormula(), opInfo, data);
} else if (formula.getSubformula().isBoundedUntilFormula()){
preprocessBoundedUntilFormula(formula.getSubformula().asBoundedUntilFormula(), opInfo, data);
} else if (formula.getSubformula().isGloballyFormula()){
preprocessGloballyFormula(formula.getSubformula().asGloballyFormula(), opInfo, data);
} else if (formula.getSubformula().isEventuallyFormula()){
preprocessEventuallyFormula(formula.getSubformula().asEventuallyFormula(), opInfo, data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The subformula of " << formula << " is not supported.");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessRewardOperatorFormula(storm::logic::RewardOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
STORM_LOG_THROW((formula.hasRewardModelName() && data.model->hasRewardModel(formula.getRewardModelName()))
|| (!formula.hasRewardModelName() && data.model->hasUniqueRewardModel()), storm::exceptions::InvalidPropertyException, "The reward model is not unique or the formula " << formula << " does not specify an existing reward model.");
std::string rewardModelName;
if (formula.hasRewardModelName()) {
rewardModelName = formula.getRewardModelName();
STORM_LOG_THROW(data.model->hasRewardModel(rewardModelName), storm::exceptions::InvalidPropertyException, "The reward model specified by formula " << formula << " does not exist in the model");
} else {
STORM_LOG_THROW(data.model->hasUniqueRewardModel(), storm::exceptions::InvalidOperationException, "The formula " << formula << " does not specify a reward model name and the reward model is not unique.");
rewardModelName = data.model->getRewardModels().begin()->first;
}
data.objectives.back()->lowerResultBound = storm::utility::zero<ValueType>();
if (formula.getSubformula().isEventuallyFormula()){
preprocessEventuallyFormula(formula.getSubformula().asEventuallyFormula(), opInfo, data, rewardModelName);
} else if (formula.getSubformula().isCumulativeRewardFormula()) {
preprocessCumulativeRewardFormula(formula.getSubformula().asCumulativeRewardFormula(), opInfo, data, rewardModelName);
} else if (formula.getSubformula().isTotalRewardFormula()) {
preprocessTotalRewardFormula(opInfo, data, rewardModelName);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The subformula of " << formula << " is not supported.");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessTimeOperatorFormula(storm::logic::TimeOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// Time formulas are only supported for Markov automata
STORM_LOG_THROW(data.model->isOfType(storm::models::ModelType::MarkovAutomaton), storm::exceptions::InvalidPropertyException, "Time operator formulas are only supported for Markov automata.");
data.objectives.back()->lowerResultBound = storm::utility::zero<ValueType>();
if (formula.getSubformula().isEventuallyFormula()){
preprocessEventuallyFormula(formula.getSubformula().asEventuallyFormula(), opInfo, data);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The subformula of " << formula << " is not supported.");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessUntilFormula(storm::logic::UntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, std::shared_ptr<storm::logic::Formula const> subformula) {
// Try to transform the formula to expected total (or cumulative) rewards
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(*data.model);
storm::storage::BitVector rightSubformulaResult = mc.check(formula.getRightSubformula())->asExplicitQualitativeCheckResult().getTruthValuesVector();
// Check if the formula is already satisfied in the initial state because then the transformation to expected rewards will fail.
// TODO: Handle this case more properly
STORM_LOG_THROW((data.model->getInitialStates() & rightSubformulaResult).empty(), storm::exceptions::NotImplementedException, "The Probability for the objective " << *data.objectives.back()->originalFormula << " is always one as the rhs of the until formula is true in the initial state. This (trivial) case is currently not implemented.");
// Whenever a state that violates the left subformula or satisfies the right subformula is reached, the objective is 'decided', i.e., no more reward should be collected from there
storm::storage::BitVector notLeftOrRight = mc.check(formula.getLeftSubformula())->asExplicitQualitativeCheckResult().getTruthValuesVector();
notLeftOrRight.complement();
notLeftOrRight |=rightSubformulaResult;
// Get the states that are reachable from a notLeftOrRight state
storm::storage::BitVector allStates(data.model->getNumberOfStates(), true), noStates(data.model->getNumberOfStates(), false);
storm::storage::BitVector reachableFromGoal = storm::utility::graph::getReachableStates(data.model->getTransitionMatrix(), notLeftOrRight, allStates, noStates);
// Get the states that are reachable from an initial state, stopping at the states reachable from goal
storm::storage::BitVector reachableFromInit = storm::utility::graph::getReachableStates(data.model->getTransitionMatrix(), data.model->getInitialStates(), ~notLeftOrRight, reachableFromGoal);
// Exclude the actual notLeftOrRight states from the states that are reachable from init
reachableFromInit &= ~notLeftOrRight;
// If we can reach a state that is reachable from goal, but which is not a goal state, it means that the transformation to expected rewards is not possible.
if ((reachableFromInit & reachableFromGoal).empty()) {
STORM_LOG_INFO("Objective " << data.objectives.back()->originalFormula << " is transformed to an expected total/cumulative reward property.");
// Transform to expected total rewards:
// build stateAction reward vector that gives (one*transitionProbability) reward whenever a transition leads from a reachableFromInit state to a goalState
std::vector<typename SparseModelType::ValueType> objectiveRewards(data.model->getTransitionMatrix().getRowCount(), storm::utility::zero<typename SparseModelType::ValueType>());
for (auto const& state : reachableFromInit) {
for (uint_fast64_t row = data.model->getTransitionMatrix().getRowGroupIndices()[state]; row < data.model->getTransitionMatrix().getRowGroupIndices()[state + 1]; ++row) {
objectiveRewards[row] = data.model->getTransitionMatrix().getConstrainedRowSum(row, rightSubformulaResult);
}
}
std::string rewardModelName = data.rewardModelNamePrefix + std::to_string(data.objectives.size());
data.model->addRewardModel(rewardModelName, typename SparseModelType::RewardModelType(boost::none, std::move(objectiveRewards)));
if (subformula == nullptr) {
subformula = std::make_shared<storm::logic::TotalRewardFormula>();
}
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(subformula, rewardModelName, opInfo);
} else {
STORM_LOG_INFO("Objective " << data.objectives.back()->originalFormula << " can not be transformed to an expected total/cumulative reward property.");
data.objectives.back()->formula = std::make_shared<storm::logic::ProbabilityOperatorFormula>(formula.asSharedPointer(), opInfo);
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessBoundedUntilFormula(storm::logic::BoundedUntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// Check how to handle this query
if (formula.isMultiDimensional() || formula.getTimeBoundReference().isRewardBound()) {
STORM_LOG_INFO("Objective " << data.objectives.back()->originalFormula << " is not transformed to an expected cumulative reward property.");
data.objectives.back()->formula = std::make_shared<storm::logic::ProbabilityOperatorFormula>(formula.asSharedPointer(), opInfo);
} else if (!formula.hasLowerBound() || (!formula.isLowerBoundStrict() && storm::utility::isZero(formula.template getLowerBound<storm::RationalNumber>()))) {
std::shared_ptr<storm::logic::Formula const> subformula;
if (!formula.hasUpperBound()) {
// The formula is actually unbounded
subformula = std::make_shared<storm::logic::TotalRewardFormula>();
} else {
STORM_LOG_THROW(!data.model->isOfType(storm::models::ModelType::MarkovAutomaton) || formula.getTimeBoundReference().isTimeBound(), storm::exceptions::InvalidPropertyException, "Bounded until formulas for Markov Automata are only allowed when time bounds are considered.");
storm::logic::TimeBound bound(formula.isUpperBoundStrict(), formula.getUpperBound());
subformula = std::make_shared<storm::logic::CumulativeRewardFormula>(bound, formula.getTimeBoundReference());
}
preprocessUntilFormula(storm::logic::UntilFormula(formula.getLeftSubformula().asSharedPointer(), formula.getRightSubformula().asSharedPointer()), opInfo, data, subformula);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Property " << formula << "is not supported");
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessGloballyFormula(storm::logic::GloballyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data) {
// The formula is transformed to an until formula for the complementary event.
auto negatedSubformula = std::make_shared<storm::logic::UnaryBooleanStateFormula>(storm::logic::UnaryBooleanStateFormula::OperatorType::Not, formula.getSubformula().asSharedPointer());
preprocessUntilFormula(storm::logic::UntilFormula(storm::logic::Formula::getTrueFormula(), negatedSubformula), opInfo, data);
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessEventuallyFormula(storm::logic::EventuallyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName) {
if (formula.isReachabilityProbabilityFormula()){
preprocessUntilFormula(storm::logic::UntilFormula(storm::logic::Formula::getTrueFormula(), formula.getSubformula().asSharedPointer()), opInfo, data);
return;
}
// Analyze the subformula
storm::modelchecker::SparsePropositionalModelChecker<SparseModelType> mc(*data.model);
storm::storage::BitVector subFormulaResult = mc.check(formula.getSubformula())->asExplicitQualitativeCheckResult().getTruthValuesVector();
// Get the states that are reachable from a goal state
storm::storage::BitVector allStates(data.model->getNumberOfStates(), true), noStates(data.model->getNumberOfStates(), false);
storm::storage::BitVector reachableFromGoal = storm::utility::graph::getReachableStates(data.model->getTransitionMatrix(), subFormulaResult, allStates, noStates);
// Get the states that are reachable from an initial state, stopping at the states reachable from goal
storm::storage::BitVector reachableFromInit = storm::utility::graph::getReachableStates(data.model->getTransitionMatrix(), data.model->getInitialStates(), allStates, reachableFromGoal);
// Exclude the actual goal states from the states that are reachable from an initial state
reachableFromInit &= ~subFormulaResult;
// If we can reach a state that is reachable from goal but which is not a goal state, it means that the transformation to expected total rewards is not possible.
if ((reachableFromInit & reachableFromGoal).empty()) {
STORM_LOG_INFO("Objective " << data.objectives.back()->originalFormula << " is transformed to an expected total reward property.");
// Transform to expected total rewards:
std::string rewardModelName = data.rewardModelNamePrefix + std::to_string(data.objectives.size());
auto totalRewardFormula = std::make_shared<storm::logic::TotalRewardFormula>();
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(totalRewardFormula, rewardModelName, opInfo);
if (formula.isReachabilityRewardFormula()) {
// build stateAction reward vector that only gives reward for states that are reachable from init
assert(optionalRewardModelName.is_initialized());
typename SparseModelType::RewardModelType objectiveRewards = data.model->getRewardModel(optionalRewardModelName.get());
objectiveRewards.reduceToStateBasedRewards(data.model->getTransitionMatrix(), false);
if (objectiveRewards.hasStateRewards()) {
storm::utility::vector::setVectorValues(objectiveRewards.getStateRewardVector(), reachableFromGoal, storm::utility::zero<typename SparseModelType::ValueType>());
}
if (objectiveRewards.hasStateActionRewards()) {
for (auto state : reachableFromGoal) {
std::fill_n(objectiveRewards.getStateActionRewardVector().begin() + data.model->getTransitionMatrix().getRowGroupIndices()[state], data.model->getTransitionMatrix().getRowGroupSize(state), storm::utility::zero<typename SparseModelType::ValueType>());
}
}
data.model->addRewardModel(rewardModelName, std::move(objectiveRewards));
} else if (formula.isReachabilityTimeFormula() && data.model->isOfType(storm::models::ModelType::MarkovAutomaton)) {
// build stateAction reward vector that only gives reward for relevant states
std::vector<typename SparseModelType::ValueType> timeRewards(data.model->getNumberOfStates(), storm::utility::zero<typename SparseModelType::ValueType>());
storm::utility::vector::setVectorValues(timeRewards, dynamic_cast<storm::models::sparse::MarkovAutomaton<typename SparseModelType::ValueType> const&>(*data.model).getMarkovianStates() & reachableFromInit, storm::utility::one<typename SparseModelType::ValueType>());
data.model->addRewardModel(rewardModelName, typename SparseModelType::RewardModelType(std::move(timeRewards)));
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The formula " << formula << " neither considers reachability probabilities nor reachability rewards " << (data.model->isOfType(storm::models::ModelType::MarkovAutomaton) ? "nor reachability time" : "") << ". This is not supported.");
}
} else {
STORM_LOG_INFO("Objective " << data.objectives.back()->originalFormula << " can not be transformed to an expected total/cumulative reward property.");
if (formula.isReachabilityRewardFormula()) {
assert(optionalRewardModelName.is_initialized());
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(formula.asSharedPointer(), optionalRewardModelName.get(), opInfo);
} else if (formula.isReachabilityTimeFormula() && data.model->isOfType(storm::models::ModelType::MarkovAutomaton)) {
data.objectives.back()->formula = std::make_shared<storm::logic::TimeOperatorFormula>(formula.asSharedPointer(), opInfo);
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "The formula " << formula << " neither considers reachability probabilities nor reachability rewards " << (data.model->isOfType(storm::models::ModelType::MarkovAutomaton) ? "nor reachability time" : "") << ". This is not supported.");
}
}
data.finiteRewardCheckObjectives.set(data.objectives.size() - 1, true);
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessCumulativeRewardFormula(storm::logic::CumulativeRewardFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName) {
STORM_LOG_THROW(data.model->isOfType(storm::models::ModelType::Mdp), storm::exceptions::InvalidPropertyException, "Cumulative reward formulas are not supported for the given model type.");
auto cumulativeRewardFormula = std::make_shared<storm::logic::CumulativeRewardFormula>(formula);
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(cumulativeRewardFormula, *optionalRewardModelName, opInfo);
bool onlyRewardBounds = true;
for (uint64_t i = 0; i < cumulativeRewardFormula->getDimension(); ++i) {
if (!cumulativeRewardFormula->getTimeBoundReference(i).isRewardBound()) {
onlyRewardBounds = false;
break;
}
}
if (onlyRewardBounds) {
data.finiteRewardCheckObjectives.set(data.objectives.size() - 1, true);
}
}
template<typename SparseModelType>
void SparseMultiObjectivePreprocessor<SparseModelType>::preprocessTotalRewardFormula(storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName) {
auto totalRewardFormula = std::make_shared<storm::logic::TotalRewardFormula>();
data.objectives.back()->formula = std::make_shared<storm::logic::RewardOperatorFormula>(totalRewardFormula, *optionalRewardModelName, opInfo);
data.finiteRewardCheckObjectives.set(data.objectives.size() - 1, true);
}
template<typename SparseModelType>
typename SparseMultiObjectivePreprocessor<SparseModelType>::ReturnType SparseMultiObjectivePreprocessor<SparseModelType>::buildResult(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula, PreprocessorData& data) {
ReturnType result(originalFormula, originalModel);
auto backwardTransitions = data.model->getBackwardTransitions();
result.preprocessedModel = data.model;
for (auto& obj : data.objectives) {
result.objectives.push_back(std::move(*obj));
}
result.queryType = getQueryType(result.objectives);
result.maybeInfiniteRewardObjectives = std::move(data.finiteRewardCheckObjectives);
return result;
}
template<typename SparseModelType>
typename SparseMultiObjectivePreprocessor<SparseModelType>::ReturnType::QueryType SparseMultiObjectivePreprocessor<SparseModelType>::getQueryType(std::vector<Objective<ValueType>> const& objectives) {
uint_fast64_t numOfObjectivesWithThreshold = 0;
for (auto& obj : objectives) {
if (obj.formula->hasBound()) {
++numOfObjectivesWithThreshold;
}
}
if (numOfObjectivesWithThreshold == objectives.size()) {
return ReturnType::QueryType::Achievability;
} else if (numOfObjectivesWithThreshold + 1 == objectives.size()) {
// Note: We do not want to consider a Pareto query when the total number of objectives is one.
return ReturnType::QueryType::Quantitative;
} else if (numOfObjectivesWithThreshold == 0) {
return ReturnType::QueryType::Pareto;
} else {
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Invalid Multi-objective query: The numer of qualitative objectives should be either 0 (Pareto query), 1 (quantitative query), or #objectives (achievability query).");
}
}
template class SparseMultiObjectivePreprocessor<storm::models::sparse::Mdp<double>>;
template class SparseMultiObjectivePreprocessor<storm::models::sparse::MarkovAutomaton<double>>;
template class SparseMultiObjectivePreprocessor<storm::models::sparse::Mdp<storm::RationalNumber>>;
template class SparseMultiObjectivePreprocessor<storm::models::sparse::MarkovAutomaton<storm::RationalNumber>>;
}
}
}
}

105
src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessor.h
View File

@ -0,0 +1,105 @@
#pragma once
#include <memory>
#include <string>
#include "storm/logic/Formulas.h"
#include "storm/storage/BitVector.h"
#include "storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessorResult.h"
#include "storm/storage/memorystructure/MemoryStructure.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
namespace preprocessing {
/*
* This class invokes the necessary preprocessing for the constraint based multi-objective model checking algorithm
*/
template <class SparseModelType>
class SparseMultiObjectivePreprocessor {
public:
typedef typename SparseModelType::ValueType ValueType;
typedef typename SparseModelType::RewardModelType RewardModelType;
typedef SparseMultiObjectivePreprocessorResult<SparseModelType> ReturnType;
/*!
* Preprocesses the given model w.r.t. the given formulas
* @param originalModel The considered model
* @param originalFormula the considered formula. The subformulas should only contain one OperatorFormula at top level.
*/
static ReturnType preprocess(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula);
private:
struct PreprocessorData {
std::shared_ptr<SparseModelType> model;
std::vector<std::shared_ptr<Objective<ValueType>>> objectives;
// Indices of the objectives that require a check for finite reward
storm::storage::BitVector finiteRewardCheckObjectives;
// Indices of the objectives for which we need to compute an upper bound for the result
storm::storage::BitVector upperResultBoundObjectives;
std::string rewardModelNamePrefix;
PreprocessorData(std::shared_ptr<SparseModelType> model);
};
/*!
* Apply the neccessary preprocessing for the given formula.
* @param formula the current (sub)formula
* @param opInfo the information of the resulting operator formula
* @param data the current data. The currently processed objective is located at data.objectives.back()
* @param optionalRewardModelName the reward model name that is considered for the formula (if available)
*/
static void preprocessOperatorFormula(storm::logic::OperatorFormula const& formula, PreprocessorData& data);
static void preprocessProbabilityOperatorFormula(storm::logic::ProbabilityOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessRewardOperatorFormula(storm::logic::RewardOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessTimeOperatorFormula(storm::logic::TimeOperatorFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessUntilFormula(storm::logic::UntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, std::shared_ptr<storm::logic::Formula const> subformula = nullptr);
static void preprocessBoundedUntilFormula(storm::logic::BoundedUntilFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessGloballyFormula(storm::logic::GloballyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data);
static void preprocessEventuallyFormula(storm::logic::EventuallyFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName = boost::none);
static void preprocessCumulativeRewardFormula(storm::logic::CumulativeRewardFormula const& formula, storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName = boost::none);
static void preprocessTotalRewardFormula(storm::logic::OperatorInformation const& opInfo, PreprocessorData& data, boost::optional<std::string> const& optionalRewardModelName = boost::none); // The total reward formula itself does not need to be provided as it is unique.
/*!
* Builds the result from preprocessing
*/
static ReturnType buildResult(SparseModelType const& originalModel, storm::logic::MultiObjectiveFormula const& originalFormula, PreprocessorData& data);
/*!
* Returns the query type
*/
static typename ReturnType::QueryType getQueryType(std::vector<Objective<ValueType>> const& objectives);
/*!
* Computes the set of states that have zero expected reward w.r.t. all expected reward objectives
*/
static void setReward0States(ReturnType& result, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
/*!
* Checks whether the occurring expected rewards are finite and sets the RewardFinitenessType accordingly
* Returns the set of states for which a scheduler exists that induces finite reward for all objectives
*/
static void checkRewardFiniteness(ReturnType& result, storm::storage::BitVector const& finiteRewardCheckObjectives, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
/*!
* Finds an upper bound for the expected reward of the objective with the given index (assuming it considers an expected reward objective)
*/
static boost::optional<ValueType> computeUpperResultBound(ReturnType const& result, uint64_t objIndex, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
};
}
}
}
}

95
src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessorResult.h
View File

@ -0,0 +1,95 @@
#pragma once
#include <vector>
#include <memory>
#include <boost/optional.hpp>
#include "storm/logic/Formulas.h"
#include "storm/modelchecker/multiobjective/Objective.h"
#include "storm/storage/BitVector.h"
#include "storm/exceptions/UnexpectedException.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
namespace preprocessing {
template <class SparseModelType>
struct SparseMultiObjectivePreprocessorResult {
enum class QueryType { Achievability, Quantitative, Pareto };
// Original data
storm::logic::MultiObjectiveFormula const& originalFormula;
SparseModelType const& originalModel;
// The preprocessed model and objectives
std::shared_ptr<SparseModelType> preprocessedModel;
std::vector<Objective<typename SparseModelType::ValueType>> objectives;
// Data about the query
QueryType queryType;
// Indices of the objectives that can potentially yield infinite reward
storm::storage::BitVector maybeInfiniteRewardObjectives;
SparseMultiObjectivePreprocessorResult(storm::logic::MultiObjectiveFormula const& originalFormula, SparseModelType const& originalModel) : originalFormula(originalFormula), originalModel(originalModel) {
// Intentionally left empty
}
bool containsOnlyTrivialObjectives() const {
// Trivial objectives are either total reward formulas or single-dimensional step or time bounded cumulative reward formulas
for (auto const& obj : objectives) {
if (obj.formula->isRewardOperatorFormula() && obj.formula->getSubformula().isTotalRewardFormula()) {
continue;
}
if (obj.formula->isRewardOperatorFormula() && obj.formula->getSubformula().isCumulativeRewardFormula()) {
auto const& subf = obj.formula->getSubformula().asCumulativeRewardFormula();
if (!subf.isMultiDimensional() && (subf.getTimeBoundReference().isTimeBound() || subf.getTimeBoundReference().isStepBound())) {
continue;
}
}
// Reaching this point means that the objective is considered as non-trivial
return false;
}
return true;
}
void printToStream(std::ostream& out) const {
out << std::endl;
out << "---------------------------------------------------------------------------------------------------------------------------------------" << std::endl;
out << " Multi-objective Query " << std::endl;
out << "---------------------------------------------------------------------------------------------------------------------------------------" << std::endl;
out << std::endl;
out << "Original Formula: " << std::endl;
out << "--------------------------------------------------------------" << std::endl;
out << "\t" << originalFormula << std::endl;
out << std::endl;
out << "Objectives:" << std::endl;
out << "--------------------------------------------------------------" << std::endl;
for (auto const& obj : objectives) {
obj.printToStream(out);
}
out << "--------------------------------------------------------------" << std::endl;
out << std::endl;
out << "Original Model Information:" << std::endl;
originalModel.printModelInformationToStream(out);
out << std::endl;
out << "Preprocessed Model Information:" << std::endl;
preprocessedModel->printModelInformationToStream(out);
out << std::endl;
out << "---------------------------------------------------------------------------------------------------------------------------------------" << std::endl;
}
friend std::ostream& operator<<(std::ostream& out, SparseMultiObjectivePreprocessorResult<SparseModelType> const& ret) {
ret.printToStream(out);
return out;
}
};
}
}
}
}

230
src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveRewardAnalysis.cpp
View File

@ -0,0 +1,230 @@
#include "storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveRewardAnalysis.h"
#include <algorithm>
#include <set>
#include <storm/modelchecker/multiobjective/SparseMultiObjectivePreprocessorResult.h>
#include "storm/models/sparse/Mdp.h"
#include "storm/models/sparse/MarkovAutomaton.h"
#include "storm/models/sparse/StandardRewardModel.h"
#include "storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveMemoryIncorporation.h"
#include "storm/modelchecker/propositional/SparsePropositionalModelChecker.h"
#include "storm/modelchecker/prctl/helper/BaierUpperRewardBoundsComputer.h"
#include "storm/modelchecker/results/ExplicitQualitativeCheckResult.h"
#include "storm/storage/MaximalEndComponentDecomposition.h"
#include "storm/storage/expressions/ExpressionManager.h"
#include "storm/transformer/EndComponentEliminator.h"
#include "storm/utility/macros.h"
#include "storm/utility/vector.h"
#include "storm/utility/graph.h"
#include "storm/settings/SettingsManager.h"
#include "storm/settings/modules/GeneralSettings.h"
#include "storm/exceptions/InvalidPropertyException.h"
#include "storm/exceptions/UnexpectedException.h"
#include "storm/exceptions/NotImplementedException.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
namespace preprocessing {
template<typename SparseModelType>
typename SparseMultiObjectiveRewardAnalysis<SparseModelType>::ReturnType SparseMultiObjectiveRewardAnalysis<SparseModelType>::analyze(storm::modelchecker::multiobjective::preprocessing::SparseMultiObjectivePreprocessorResult<SparseModelType> const& preprocessorResult) {
ReturnType result;
auto backwardTransitions = preprocessorResult.preprocessedModel->getBackwardTransitions();
setReward0States(result, preprocessorResult, backwardTransitions);
checkRewardFiniteness(result, preprocessorResult, backwardTransitions);
return result;
}
template<typename SparseModelType>
void SparseMultiObjectiveRewardAnalysis<SparseModelType>::setReward0States(ReturnType& result, storm::modelchecker::multiobjective::preprocessing::SparseMultiObjectivePreprocessorResult<SparseModelType> const& preprocessorResult, storm::storage::SparseMatrix<ValueType> const& backwardTransitions) {
uint_fast64_t stateCount = preprocessorResult.preprocessedModel->getNumberOfStates();
auto const& transitions = preprocessorResult.preprocessedModel->getTransitionMatrix();
std::vector<uint_fast64_t> const& groupIndices = transitions.getRowGroupIndices();
storm::storage::BitVector allStates(stateCount, true);
// Get the choices that yield non-zero reward
storm::storage::BitVector zeroRewardChoices(preprocessorResult.preprocessedModel->getNumberOfChoices(), true);
for (auto const& obj : preprocessorResult.objectives) {
if (obj.formula->isRewardOperatorFormula()) {
STORM_LOG_WARN_COND(obj.formula->getSubformula().isTotalRewardFormula() || obj.formula->getSubformula().isCumulativeRewardFormula(), "Analyzing reachability reward formulas is not supported properly.");
auto const& rewModel = preprocessorResult.preprocessedModel->getRewardModel(obj.formula->asRewardOperatorFormula().getRewardModelName());
zeroRewardChoices &= rewModel.getChoicesWithZeroReward(transitions);
}
}
// Get the states that have reward for at least one (or for all) choices assigned to it.
storm::storage::BitVector statesWithRewardForOneChoice = storm::storage::BitVector(stateCount, false);
storm::storage::BitVector statesWithRewardForAllChoices = storm::storage::BitVector(stateCount, true);
for (uint_fast64_t state = 0; state < stateCount; ++state) {
bool stateHasChoiceWithReward = false;
bool stateHasChoiceWithoutReward = false;
uint_fast64_t const& groupEnd = groupIndices[state + 1];
for (uint_fast64_t choice = groupIndices[state]; choice < groupEnd; ++choice) {
if (zeroRewardChoices.get(choice)) {
stateHasChoiceWithoutReward = true;
} else {
stateHasChoiceWithReward = true;
}
}
if (stateHasChoiceWithReward) {
statesWithRewardForOneChoice.set(state, true);
}
if (stateHasChoiceWithoutReward) {
statesWithRewardForAllChoices.set(state, false);
}
}
// get the states for which there is a scheduler yielding reward zero
result.reward0EStates = storm::utility::graph::performProbGreater0A(transitions, groupIndices, backwardTransitions, allStates, statesWithRewardForAllChoices, false, 0, zeroRewardChoices);
result.reward0EStates.complement();
result.reward0AStates = storm::utility::graph::performProb0A(backwardTransitions, allStates, statesWithRewardForOneChoice);
assert(result.reward0AStates.isSubsetOf(result.reward0EStates));
}
template<typename SparseModelType>
void SparseMultiObjectiveRewardAnalysis<SparseModelType>::checkRewardFiniteness(ReturnType& result, storm::modelchecker::multiobjective::preprocessing::SparseMultiObjectivePreprocessorResult<SparseModelType> const& preprocessorResult, storm::storage::SparseMatrix<ValueType> const& backwardTransitions) {
result.rewardFinitenessType = RewardFinitenessType::AllFinite;
auto const& transitions = preprocessorResult.preprocessedModel->getTransitionMatrix();
std::vector<uint_fast64_t> const& groupIndices = transitions.getRowGroupIndices();
storm::storage::BitVector maxRewardsToCheck(preprocessorResult.preprocessedModel->getNumberOfChoices(), true);
storm::storage::BitVector minRewardsToCheck(preprocessorResult.preprocessedModel->getNumberOfChoices(), true);
for (auto const& objIndex : preprocessorResult.maybeInfiniteRewardObjectives) {
STORM_LOG_ASSERT(preprocessorResult.objectives[objIndex].formula->isRewardOperatorFormula(), "Objective needs to be checked for finite reward but has no reward operator.");
auto const& rewModel = preprocessorResult.preprocessedModel->getRewardModel(preprocessorResult.objectives[objIndex].formula->asRewardOperatorFormula().getRewardModelName());
auto unrelevantChoices = rewModel.getChoicesWithZeroReward(transitions);
// For (upper) reward bounded cumulative reward formulas, we do not need to consider the choices where boundReward is collected.
if (preprocessorResult.objectives[objIndex].formula->getSubformula().isCumulativeRewardFormula()) {
auto const& timeBoundReference = preprocessorResult.objectives[objIndex].formula->getSubformula().asCumulativeRewardFormula().getTimeBoundReference();
// Only reward bounded formulas need a finiteness check
assert(timeBoundReference.isRewardBound());
auto const& rewModelOfBound = preprocessorResult.preprocessedModel->getRewardModel(timeBoundReference.getRewardName());
unrelevantChoices |= ~rewModelOfBound.getChoicesWithZeroReward(transitions);
}
if (storm::solver::minimize(preprocessorResult.objectives[objIndex].formula->getOptimalityType())) {
minRewardsToCheck &= unrelevantChoices;
} else {
maxRewardsToCheck &= unrelevantChoices;
}
}
maxRewardsToCheck.complement();
minRewardsToCheck.complement();
// Check reward finiteness under all schedulers
storm::storage::BitVector allStates(preprocessorResult.preprocessedModel->getNumberOfStates(), true);
if (storm::utility::graph::checkIfECWithChoiceExists(transitions, backwardTransitions, allStates, maxRewardsToCheck | minRewardsToCheck)) {
// Check whether there is a scheduler yielding infinite reward for a maximizing objective
if (storm::utility::graph::checkIfECWithChoiceExists(transitions, backwardTransitions, allStates, maxRewardsToCheck)) {
result.rewardFinitenessType = RewardFinitenessType::Infinite;
} else {
// Check whether there is a scheduler under which all rewards are finite.
result.rewardLessInfinityEStates = storm::utility::graph::performProb1E(transitions, groupIndices, backwardTransitions, allStates, result.reward0EStates);
if ((result.rewardLessInfinityEStates.get() & preprocessorResult.preprocessedModel->getInitialStates()).empty()) {
// There is no scheduler that induces finite reward for the initial state
result.rewardFinitenessType = RewardFinitenessType::Infinite;
} else {
result.rewardFinitenessType = RewardFinitenessType::ExistsParetoFinite;
}
}
} else {
result.rewardLessInfinityEStates = allStates;
}
}
template<typename SparseModelType>
void SparseMultiObjectiveRewardAnalysis<SparseModelType>::computeUpperResultBound(SparseModelType const& model, storm::modelchecker::multiobjective::Objective<ValueType>& objective, storm::storage::SparseMatrix<ValueType> const& backwardTransitions) {
STORM_LOG_INFO_COND(!objective.upperResultBound.is_initialized(), "Tried to find an upper result bound for an objective, but a result bound is already there.");
if (model.isOfType(storm::models::ModelType::Mdp)) {
auto const& transitions = model.getTransitionMatrix();
if (objective.formula->isRewardOperatorFormula()) {
auto const& rewModel = model.getRewardModel(objective.formula->asRewardOperatorFormula().getRewardModelName());
auto actionRewards = rewModel.getTotalRewardVector(transitions);
if (objective.formula->getSubformula().isTotalRewardFormula() || objective.formula->getSubformula().isCumulativeRewardFormula()) {
// We have to eliminate ECs here to treat zero-reward ECs
storm::storage::BitVector allStates(model.getNumberOfStates(), true);
// Get the set of states from which no reward is reachable
auto nonZeroRewardStates = rewModel.getStatesWithZeroReward(transitions);
nonZeroRewardStates.complement();
auto expRewGreater0EStates = storm::utility::graph::performProbGreater0E(backwardTransitions, allStates, nonZeroRewardStates);
auto zeroRewardChoices = rewModel.getChoicesWithZeroReward(transitions);
auto ecElimRes = storm::transformer::EndComponentEliminator<ValueType>::transform(transitions, expRewGreater0EStates, zeroRewardChoices, ~allStates);
allStates.resize(ecElimRes.matrix.getRowGroupCount());
storm::storage::BitVector outStates(allStates.size(), false);
std::vector<ValueType> rew0StateProbs;
rew0StateProbs.reserve(ecElimRes.matrix.getRowCount());
for (uint64_t state = 0; state < allStates.size(); ++ state) {
for (uint64_t choice = ecElimRes.matrix.getRowGroupIndices()[state]; choice < ecElimRes.matrix.getRowGroupIndices()[state + 1]; ++choice) {
// Check whether the choice lead to a state with expRew 0 in the original model
bool isOutChoice = false;
uint64_t originalModelChoice = ecElimRes.newToOldRowMapping[choice];
for (auto const& entry : transitions.getRow(originalModelChoice)) {
if (!expRewGreater0EStates.get(entry.getColumn())) {
isOutChoice = true;
outStates.set(state, true);
rew0StateProbs.push_back(storm::utility::one<ValueType>() - ecElimRes.matrix.getRowSum(choice));
assert (!storm::utility::isZero(rew0StateProbs.back()));
break;
}
}
if (!isOutChoice) {
rew0StateProbs.push_back(storm::utility::zero<ValueType>());
}
}
}
// An upper reward bound can only be computed if it is below infinity
if (storm::utility::graph::performProb1A(ecElimRes.matrix, ecElimRes.matrix.getRowGroupIndices(), ecElimRes.matrix.transpose(true), allStates, outStates).full()) {
std::vector<ValueType> rewards;
rewards.reserve(ecElimRes.matrix.getRowCount());
for (auto row : ecElimRes.newToOldRowMapping) {
rewards.push_back(actionRewards[row]);
}
storm::modelchecker::helper::BaierUpperRewardBoundsComputer<ValueType> baier(ecElimRes.matrix, rewards, rew0StateProbs);
if (objective.upperResultBound) {
objective.upperResultBound = std::min(objective.upperResultBound.get(), baier.computeUpperBound());
} else {
objective.upperResultBound = baier.computeUpperBound();
}
}
}
}
if (objective.upperResultBound) {
STORM_LOG_INFO("Computed upper result bound " << objective.upperResultBound.get() << " for objective " << *objective.formula << ".");
} else {
STORM_LOG_WARN("Could not compute upper result bound for objective " << *objective.formula);
}
}
}
template class SparseMultiObjectiveRewardAnalysis<storm::models::sparse::Mdp<double>>;
template class SparseMultiObjectiveRewardAnalysis<storm::models::sparse::MarkovAutomaton<double>>;
template class SparseMultiObjectiveRewardAnalysis<storm::models::sparse::Mdp<storm::RationalNumber>>;
template class SparseMultiObjectiveRewardAnalysis<storm::models::sparse::MarkovAutomaton<storm::RationalNumber>>;
}
}
}
}

73
src/storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectiveRewardAnalysis.h
View File

@ -0,0 +1,73 @@
#pragma once
#include <memory>
#include <string>
#include "storm/logic/Formulas.h"
#include "storm/storage/BitVector.h"
#include "storm/storage/SparseMatrix.h"
#include "storm/modelchecker/multiobjective/preprocessing/SparseMultiObjectivePreprocessorResult.h"
namespace storm {
namespace modelchecker {
namespace multiobjective {
namespace preprocessing {
enum class RewardFinitenessType {
AllFinite, // The expected reward is finite for all objectives and all schedulers
ExistsParetoFinite, // There is a Pareto optimal scheduler yielding finite rewards for all objectives
Infinite // All Pareto optimal schedulers yield infinite reward for at least one objective
};
/*
* This class performs some analysis task regarding the occurring expected reward objectives
*/
template <class SparseModelType>
class SparseMultiObjectiveRewardAnalysis {
public:
typedef typename SparseModelType::ValueType ValueType;
typedef typename SparseModelType::RewardModelType RewardModelType;
struct ReturnType {
RewardFinitenessType rewardFinitenessType;
// The states of the preprocessed model for which...
storm::storage::BitVector reward0EStates; // ... there is a scheduler such that all expected reward objectives have value zero
storm::storage::BitVector reward0AStates; // ... all schedulers induce value 0 for all expected reward objectives
boost::optional<storm::storage::BitVector> rewardLessInfinityEStates; // ... there is a scheduler yielding finite reward for all expected reward objectives
};
/*!
* Analyzes the reward objectives of the multi objective query
* @param preprocessorResult The result from preprocessing. Must only contain expected total reward or cumulative reward objectives.
*/
static ReturnType analyze(storm::modelchecker::multiobjective::preprocessing::SparseMultiObjectivePreprocessorResult<SparseModelType> const& preprocessorResult);
/*!
* Tries to finds an upper bound for the expected reward of the objective (assuming it considers an expected total reward objective)
*/
static void computeUpperResultBound(SparseModelType const& model, storm::modelchecker::multiobjective::Objective<ValueType>& objective, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
private:
/*!
* Computes the set of states that have zero expected reward w.r.t. all expected total/cumulative reward objectives
*/
static void setReward0States(ReturnType& result, storm::modelchecker::multiobjective::preprocessing::SparseMultiObjectivePreprocessorResult<SparseModelType> const& preprocessorResult, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
/*!
* Checks whether the occurring expected rewards are finite and sets the RewardFinitenessType accordingly
* Returns the set of states for which a scheduler exists that induces finite reward for all objectives
*/
static void checkRewardFiniteness(ReturnType& result, storm::modelchecker::multiobjective::preprocessing::SparseMultiObjectivePreprocessorResult<SparseModelType> const& preprocessorResult, storm::storage::SparseMatrix<ValueType> const& backwardTransitions);
};
}
}
}
}
Write Preview
Loading…
Cancel
Save