#include "storm/modelchecker/abstraction/GameBasedMdpModelChecker.h"
#include "storm/modelchecker/results/ExplicitQuantitativeCheckResult.h"
#include "storm/modelchecker/results/ExplicitQualitativeCheckResult.h"
#include "storm/models/symbolic/StandardRewardModel.h"
#include "storm/models/symbolic/Dtmc.h"
#include "storm/models/symbolic/Mdp.h"
#include "storm/storage/expressions/ExpressionManager.h"
#include "storm/storage/expressions/VariableSetPredicateSplitter.h"
#include "storm/storage/jani/Edge.h"
#include "storm/storage/jani/EdgeDestination.h"
#include "storm/storage/jani/Model.h"
#include "storm/storage/jani/Automaton.h"
#include "storm/storage/jani/Location.h"
#include "storm/storage/jani/AutomatonComposition.h"
#include "storm/storage/jani/ParallelComposition.h"
#include "storm/storage/jani/CompositionInformationVisitor.h"
#include "storm/storage/dd/DdManager.h"
#include "storm/abstraction/prism/PrismMenuGameAbstractor.h"
#include "storm/abstraction/jani/JaniMenuGameAbstractor.h"
#include "storm/abstraction/MenuGameRefiner.h"
#include "storm/logic/FragmentSpecification.h"
#include "storm/solver/SymbolicGameSolver.h"
#include "storm/settings/SettingsManager.h"
#include "storm/settings/modules/CoreSettings.h"
#include "storm/settings/modules/AbstractionSettings.h"
#include "storm/utility/prism.h"
#include "storm/utility/macros.h"
#include "storm/exceptions/NotSupportedException.h"
#include "storm/exceptions/InvalidPropertyException.h"
#include "storm/exceptions/InvalidModelException.h"
#include "storm/modelchecker/results/CheckResult.h"
namespace storm {
namespace modelchecker {
using storm::abstraction::QuantitativeGameResult;
using storm::abstraction::QuantitativeGameResultMinMax;
template<storm::dd::DdType Type, typename ModelType>
GameBasedMdpModelChecker<Type, ModelType>::GameBasedMdpModelChecker(storm::storage::SymbolicModelDescription const& model, std::shared_ptr<storm::utility::solver::SmtSolverFactory> const& smtSolverFactory) : smtSolverFactory(smtSolverFactory), comparator(storm::settings::getModule<storm::settings::modules::AbstractionSettings>().getPrecision()), reuseQualitativeResults(false), reuseQuantitativeResults(false) {
model.requireNoUndefinedConstants();
if (model.isPrismProgram()) {
storm::prism::Program const& originalProgram = model.asPrismProgram();
STORM_LOG_THROW(originalProgram.getModelType() == storm::prism::Program::ModelType::DTMC || originalProgram.getModelType() == storm::prism::Program::ModelType::MDP, storm::exceptions::NotSupportedException, "Currently only DTMCs/MDPs are supported by the game-based model checker.");
// Flatten the modules if there is more than one.
if (originalProgram.getNumberOfModules() > 1) {
preprocessedModel = originalProgram.substituteFormulas().flattenModules(this->smtSolverFactory);
} else {
preprocessedModel = originalProgram;
}
STORM_LOG_TRACE("Game-based model checker got program " << preprocessedModel.asPrismProgram());
} else {
storm::jani::Model const& originalModel = model.asJaniModel();
STORM_LOG_THROW(originalModel.getModelType() == storm::jani::ModelType::DTMC || originalModel.getModelType() == storm::jani::ModelType::MDP, storm::exceptions::NotSupportedException, "Currently only DTMCs/MDPs are supported by the game-based model checker.");
// Flatten the parallel composition.
preprocessedModel = model.asJaniModel().flattenComposition();
}
storm::settings::modules::AbstractionSettings::ReuseMode reuseMode = storm::settings::getModule<storm::settings::modules::AbstractionSettings>().getReuseMode();
reuseQualitativeResults = reuseMode == storm::settings::modules::AbstractionSettings::ReuseMode::All || reuseMode == storm::settings::modules::AbstractionSettings::ReuseMode::Qualitative;
reuseQuantitativeResults = reuseMode == storm::settings::modules::AbstractionSettings::ReuseMode::All || reuseMode == storm::settings::modules::AbstractionSettings::ReuseMode::Quantitative;
}
template<storm::dd::DdType Type, typename ModelType>
bool GameBasedMdpModelChecker<Type, ModelType>::canHandle(CheckTask<storm::logic::Formula> const& checkTask) const {
storm::logic::Formula const& formula = checkTask.getFormula();
storm::logic::FragmentSpecification fragment = storm::logic::reachability();
return formula.isInFragment(fragment) && checkTask.isOnlyInitialStatesRelevantSet();
}
template<storm::dd::DdType Type, typename ModelType>
std::unique_ptr<CheckResult> GameBasedMdpModelChecker<Type, ModelType>::computeUntilProbabilities(Environment const& env, CheckTask<storm::logic::UntilFormula> const& checkTask) {
storm::logic::UntilFormula const& pathFormula = checkTask.getFormula();
std::map<std::string, storm::expressions::Expression> labelToExpressionMapping;
if (preprocessedModel.isPrismProgram()) {
labelToExpressionMapping = preprocessedModel.asPrismProgram().getLabelToExpressionMapping();
} else {
storm::jani::Model const& janiModel = preprocessedModel.asJaniModel();
for (auto const& variable : janiModel.getGlobalVariables().getBooleanVariables()) {
if (variable.isTransient()) {
labelToExpressionMapping[variable.getName()] = janiModel.getLabelExpression(variable.asBooleanVariable());
}
}
}
storm::expressions::Expression constraintExpression = pathFormula.getLeftSubformula().toExpression(preprocessedModel.getManager(), labelToExpressionMapping);
storm::expressions::Expression targetStateExpression = pathFormula.getRightSubformula().toExpression(preprocessedModel.getManager(), labelToExpressionMapping);
return performGameBasedAbstractionRefinement(env, checkTask.substituteFormula<storm::logic::Formula>(pathFormula), constraintExpression, targetStateExpression);
}
template<storm::dd::DdType Type, typename ModelType>
std::unique_ptr<CheckResult> GameBasedMdpModelChecker<Type, ModelType>::computeReachabilityProbabilities(Environment const& env, CheckTask<storm::logic::EventuallyFormula> const& checkTask) {
storm::logic::EventuallyFormula const& pathFormula = checkTask.getFormula();
std::map<std::string, storm::expressions::Expression> labelToExpressionMapping;
if (preprocessedModel.isPrismProgram()) {
labelToExpressionMapping = preprocessedModel.asPrismProgram().getLabelToExpressionMapping();
} else {
storm::jani::Model const& janiModel = preprocessedModel.asJaniModel();
for (auto const& variable : janiModel.getGlobalVariables().getBooleanVariables()) {
if (variable.isTransient()) {
labelToExpressionMapping[variable.getName()] = janiModel.getLabelExpression(variable.asBooleanVariable());
}
}
}
storm::expressions::Expression constraintExpression = preprocessedModel.getManager().boolean(true);
storm::expressions::Expression targetStateExpression = pathFormula.getSubformula().toExpression(preprocessedModel.getManager(), labelToExpressionMapping);
return performGameBasedAbstractionRefinement(env, checkTask.substituteFormula<storm::logic::Formula>(pathFormula), constraintExpression, targetStateExpression);
}
template<storm::dd::DdType Type, typename ValueType>
std::unique_ptr<CheckResult> checkForResultAfterQualitativeCheck(CheckTask<storm::logic::Formula> const& checkTask, storm::OptimizationDirection player2Direction, storm::dd::Bdd<Type> const& initialStates, storm::dd::Bdd<Type> const& prob0, storm::dd::Bdd<Type> const& prob1) {
std::unique_ptr<CheckResult> result;
if (checkTask.isBoundSet()) {
// Despite having a bound, we create a quantitative result so that the next layer can perform the comparison.
if (player2Direction == storm::OptimizationDirection::Minimize) {
if (storm::logic::isLowerBound(checkTask.getBoundComparisonType())) {
if ((prob1 && initialStates) == initialStates) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), storm::utility::one<ValueType>());
}
} else {
if (!(prob1 && initialStates).isZero()) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), storm::utility::one<ValueType>());
}
}
} else if (player2Direction == storm::OptimizationDirection::Maximize) {
if (!storm::logic::isLowerBound(checkTask.getBoundComparisonType())) {
if ((prob0 && initialStates) == initialStates) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), storm::utility::zero<ValueType>());
}
} else {
if (!(prob0 && initialStates).isZero()) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), storm::utility::zero<ValueType>());
}
}
}
}
return result;
}
template<storm::dd::DdType Type, typename ValueType>
std::unique_ptr<CheckResult> checkForResultAfterQualitativeCheck(CheckTask<storm::logic::Formula> const& checkTask, storm::dd::Bdd<Type> const& initialStates, QualitativeGameResultMinMax<Type> const& qualitativeResult) {
// Check whether we can already give the answer based on the current information.
std::unique_ptr<CheckResult> result = checkForResultAfterQualitativeCheck<Type, ValueType>(checkTask, storm::OptimizationDirection::Minimize, initialStates, qualitativeResult.prob0Min.getPlayer1States(), qualitativeResult.prob1Min.getPlayer1States());
if (result) {
return result;
}
result = checkForResultAfterQualitativeCheck<Type, ValueType>(checkTask, storm::OptimizationDirection::Maximize, initialStates, qualitativeResult.prob0Max.getPlayer1States(), qualitativeResult.prob1Max.getPlayer1States());
if (result) {
return result;
}
return result;
}
template<typename ValueType>
std::unique_ptr<CheckResult> checkForResultAfterQuantitativeCheck(CheckTask<storm::logic::Formula> const& checkTask, storm::OptimizationDirection const& player2Direction, std::pair<ValueType, ValueType> const& initialValueRange) {
std::unique_ptr<CheckResult> result;
// If the minimum value exceeds an upper threshold or the maximum value is below a lower threshold, we can
// return the value because the property will definitely hold. Vice versa, if the minimum value exceeds an
// upper bound or the maximum value is below a lower bound, the property will definitely not hold and we can
// return the value.
if (!checkTask.isBoundSet()) {
return result;
}
ValueType const& lowerValue = initialValueRange.first;
ValueType const& upperValue = initialValueRange.second;
storm::logic::ComparisonType comparisonType = checkTask.getBoundComparisonType();
ValueType threshold = checkTask.getBoundThreshold();
if (storm::logic::isLowerBound(comparisonType)) {
if (player2Direction == storm::OptimizationDirection::Minimize) {
if ((storm::logic::isStrict(comparisonType) && lowerValue > threshold)
|| (!storm::logic::isStrict(comparisonType) && lowerValue >= threshold)) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), lowerValue);
}
} else {
if ((storm::logic::isStrict(comparisonType) && upperValue <= threshold)
|| (!storm::logic::isStrict(comparisonType) && upperValue < threshold)) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), upperValue);
}
}
} else {
if (player2Direction == storm::OptimizationDirection::Maximize) {
if ((storm::logic::isStrict(comparisonType) && upperValue < threshold) ||
(!storm::logic::isStrict(comparisonType) && upperValue <= threshold)) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), upperValue);
}
} else {
if ((storm::logic::isStrict(comparisonType) && lowerValue >= threshold) ||
(!storm::logic::isStrict(comparisonType) && lowerValue > threshold)) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), lowerValue);
}
}
}
return result;
}
template<typename ValueType>
std::unique_ptr<CheckResult> checkForResultAfterQuantitativeCheck(ValueType const& minValue, ValueType const& maxValue, storm::utility::ConstantsComparator<ValueType> const& comparator) {
std::unique_ptr<CheckResult> result;
// If the lower and upper bounds are close enough, we can return the result.
if (comparator.isEqual(minValue, maxValue)) {
result = std::make_unique<storm::modelchecker::ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), (minValue + maxValue) / ValueType(2));
}
return result;
}
template<storm::dd::DdType Type, typename ValueType>
QuantitativeGameResult<Type, ValueType> solveMaybeStates(Environment const& env, storm::OptimizationDirection const& player1Direction, storm::OptimizationDirection const& player2Direction, storm::abstraction::MenuGame<Type, ValueType> const& game, storm::dd::Bdd<Type> const& maybeStates, storm::dd::Bdd<Type> const& prob1States, boost::optional<QuantitativeGameResult<Type, ValueType>> const& startInfo = boost::none) {
STORM_LOG_TRACE("Performing quantative solution step. Player 1: " << player1Direction << ", player 2: " << player2Direction << ".");
// Compute the ingredients of the equation system.
storm::dd::Add<Type, ValueType> maybeStatesAdd = maybeStates.template toAdd<ValueType>();
storm::dd::Add<Type, ValueType> submatrix = maybeStatesAdd * game.getTransitionMatrix();
storm::dd::Add<Type, ValueType> prob1StatesAsColumn = prob1States.template toAdd<ValueType>().swapVariables(game.getRowColumnMetaVariablePairs());
storm::dd::Add<Type, ValueType> subvector = submatrix * prob1StatesAsColumn;
subvector = subvector.sumAbstract(game.getColumnVariables());
// Cut away all columns targeting non-maybe states.
submatrix *= maybeStatesAdd.swapVariables(game.getRowColumnMetaVariablePairs());
// Cut the starting vector to the maybe states of this query.
storm::dd::Add<Type, ValueType> startVector;
if (startInfo) {
startVector = startInfo.get().values * maybeStatesAdd;
} else {
startVector = game.getManager().template getAddZero<ValueType>();
}
// Create the solver and solve the equation system.
storm::solver::SymbolicGameSolverFactory<Type, ValueType> solverFactory;
std::unique_ptr<storm::solver::SymbolicGameSolver<Type, ValueType>> solver = solverFactory.create(submatrix, maybeStates, game.getIllegalPlayer1Mask(), game.getIllegalPlayer2Mask(), game.getRowVariables(), game.getColumnVariables(), game.getRowColumnMetaVariablePairs(), game.getPlayer1Variables(), game.getPlayer2Variables());
solver->setGeneratePlayersStrategies(true);
auto values = solver->solveGame(env, player1Direction, player2Direction, startVector, subvector, startInfo ? boost::make_optional(startInfo.get().getPlayer1Strategy()) : boost::none, startInfo ? boost::make_optional(startInfo.get().getPlayer2Strategy()) : boost::none);
return QuantitativeGameResult<Type, ValueType>(std::make_pair(storm::utility::zero<ValueType>(), storm::utility::one<ValueType>()), values, solver->getPlayer1Strategy(), solver->getPlayer2Strategy());
}
template<storm::dd::DdType Type, typename ValueType>
QuantitativeGameResult<Type, ValueType> computeQuantitativeResult(Environment const& env, storm::OptimizationDirection player1Direction, storm::OptimizationDirection player2Direction, storm::abstraction::MenuGame<Type, ValueType> const& game, QualitativeGameResultMinMax<Type> const& qualitativeResult, storm::dd::Add<Type, ValueType> const& initialStatesAdd, storm::dd::Bdd<Type> const& maybeStates, boost::optional<QuantitativeGameResult<Type, ValueType>> const& startInfo = boost::none) {
bool min = player2Direction == storm::OptimizationDirection::Minimize;
QuantitativeGameResult<Type, ValueType> result;
// We fix the strategies. That is, we take the decisions of the strategies obtained in the qualitiative
// preprocessing if possible.
storm::dd::Bdd<Type> combinedPlayer1QualitativeStrategies;
storm::dd::Bdd<Type> combinedPlayer2QualitativeStrategies;
if (min) {
combinedPlayer1QualitativeStrategies = (qualitativeResult.prob0Min.getPlayer1Strategy() || qualitativeResult.prob1Min.getPlayer1Strategy());
combinedPlayer2QualitativeStrategies = (qualitativeResult.prob0Min.getPlayer2Strategy() || qualitativeResult.prob1Min.getPlayer2Strategy());
} else {
combinedPlayer1QualitativeStrategies = (qualitativeResult.prob0Max.getPlayer1Strategy() || qualitativeResult.prob1Max.getPlayer1Strategy());
combinedPlayer2QualitativeStrategies = (qualitativeResult.prob0Max.getPlayer2Strategy() || qualitativeResult.prob1Max.getPlayer2Strategy());
}
result.player1Strategy = combinedPlayer1QualitativeStrategies;
result.player2Strategy = combinedPlayer2QualitativeStrategies;
result.values = game.getManager().template getAddZero<ValueType>();
auto start = std::chrono::high_resolution_clock::now();
if (!maybeStates.isZero()) {
STORM_LOG_TRACE("Solving " << maybeStates.getNonZeroCount() << " maybe states.");
// Solve the quantitative values of maybe states.
result = solveMaybeStates(env, player1Direction, player2Direction, game, maybeStates, min ? qualitativeResult.prob1Min.getPlayer1States() : qualitativeResult.prob1Max.getPlayer1States(), startInfo);
// Cut the obtained strategies to the reachable states of the game.
result.getPlayer1Strategy() &= game.getReachableStates();
result.getPlayer2Strategy() &= game.getReachableStates();
// Extend the values of the maybe states by the qualitative values.
result.values += min ? qualitativeResult.prob1Min.getPlayer1States().template toAdd<ValueType>() : qualitativeResult.prob1Max.getPlayer1States().template toAdd<ValueType>();
} else {
STORM_LOG_TRACE("No maybe states.");
// Extend the values of the maybe states by the qualitative values.
result.values += min ? qualitativeResult.prob1Min.getPlayer1States().template toAdd<ValueType>() : qualitativeResult.prob1Max.getPlayer1States().template toAdd<ValueType>();
}
// Construct an ADD holding the initial values of initial states and extract the bound on the initial states.
storm::dd::Add<Type, ValueType> initialStateValueAdd = initialStatesAdd * result.values;
ValueType maxValueOverInitialStates = initialStateValueAdd.getMax();
initialStateValueAdd += (!game.getInitialStates()).template toAdd<ValueType>();
ValueType minValueOverInitialStates = initialStateValueAdd.getMin();
result.initialStatesRange = std::make_pair(minValueOverInitialStates, maxValueOverInitialStates);
result.player1Strategy = combinedPlayer1QualitativeStrategies.existsAbstract(game.getPlayer1Variables()).ite(combinedPlayer1QualitativeStrategies, result.getPlayer1Strategy());
result.player2Strategy = combinedPlayer2QualitativeStrategies.existsAbstract(game.getPlayer2Variables()).ite(combinedPlayer2QualitativeStrategies, result.getPlayer2Strategy());
auto end = std::chrono::high_resolution_clock::now();
STORM_LOG_TRACE("Obtained quantitative " << (min ? "lower" : "upper") << " bound " << (min ? result.getInitialStatesRange().first : result.getInitialStatesRange().second) << " in " << std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count() << "ms.");
return result;
}
template<storm::dd::DdType Type, typename ModelType>
std::unique_ptr<CheckResult> GameBasedMdpModelChecker<Type, ModelType>::performGameBasedAbstractionRefinement(Environment const& env, CheckTask<storm::logic::Formula> const& checkTask, storm::expressions::Expression const& constraintExpression, storm::expressions::Expression const& targetStateExpression) {
STORM_LOG_THROW(checkTask.isOnlyInitialStatesRelevantSet(), storm::exceptions::InvalidPropertyException, "The game-based abstraction refinement model checker can only compute the result for the initial states.");
// Optimization: do not compute both bounds if not necessary (e.g. if bound given and exceeded, etc.)
// Set up initial predicates.
std::vector<storm::expressions::Expression> initialPredicates = getInitialPredicates(constraintExpression, targetStateExpression);
// Derive the optimization direction for player 1 (assuming menu-game abstraction).
storm::OptimizationDirection player1Direction = getPlayer1Direction(checkTask);
// Create the abstractor.
std::shared_ptr<storm::abstraction::MenuGameAbstractor<Type, ValueType>> abstractor;
if (preprocessedModel.isPrismProgram()) {
abstractor = std::make_shared<storm::abstraction::prism::PrismMenuGameAbstractor<Type, ValueType>>(preprocessedModel.asPrismProgram(), smtSolverFactory);
} else {
abstractor = std::make_shared<storm::abstraction::jani::JaniMenuGameAbstractor<Type, ValueType>>(preprocessedModel.asJaniModel(), smtSolverFactory);
}
if (!constraintExpression.isTrue()) {
abstractor->addTerminalStates(!constraintExpression);
}
abstractor->addTerminalStates(targetStateExpression);
abstractor->setTargetStates(targetStateExpression);
// Create a refiner that can be used to refine the abstraction when needed.
storm::abstraction::MenuGameRefiner<Type, ValueType> refiner(*abstractor, smtSolverFactory->create(preprocessedModel.getManager()));
refiner.refine(initialPredicates);
storm::dd::Bdd<Type> globalConstraintStates = abstractor->getStates(constraintExpression);
storm::dd::Bdd<Type> globalTargetStates = abstractor->getStates(targetStateExpression);
// Enter the main-loop of abstraction refinement.
boost::optional<QualitativeGameResultMinMax<Type>> previousQualitativeResult = boost::none;
boost::optional<QuantitativeGameResult<Type, ValueType>> previousMinQuantitativeResult = boost::none;
for (uint_fast64_t iterations = 0; iterations < 10000; ++iterations) {
auto iterationStart = std::chrono::high_resolution_clock::now();
STORM_LOG_TRACE("Starting iteration " << iterations << ".");
// (1) build the abstraction.
auto abstractionStart = std::chrono::high_resolution_clock::now();
storm::abstraction::MenuGame<Type, ValueType> game = abstractor->abstract();
auto abstractionEnd = std::chrono::high_resolution_clock::now();
STORM_LOG_DEBUG("Abstraction in iteration " << iterations << " has " << game.getNumberOfStates() << " (player 1) states, " << game.getNumberOfTransitions() << " transitions, " << game.getBottomStates().getNonZeroCount() << " bottom states (computed in " << std::chrono::duration_cast<std::chrono::milliseconds>(abstractionEnd - abstractionStart).count() << "ms).");
// (2) Prepare transition matrix BDD and target state BDD for later use.
storm::dd::Bdd<Type> transitionMatrixBdd = game.getTransitionMatrix().toBdd();
storm::dd::Bdd<Type> initialStates = game.getInitialStates();
STORM_LOG_THROW(initialStates.getNonZeroCount() == 1 || checkTask.isBoundSet(), storm::exceptions::InvalidPropertyException, "Game-based abstraction refinement requires a bound on the formula for model with " << initialStates.getNonZeroCount() << " initial states.");
storm::dd::Bdd<Type> constraintStates = globalConstraintStates && game.getReachableStates();
storm::dd::Bdd<Type> targetStates = globalTargetStates && game.getReachableStates();
if (player1Direction == storm::OptimizationDirection::Minimize) {
targetStates |= game.getBottomStates();
}
// #ifdef LOCAL_DEBUG
// targetStates.template toAdd<ValueType>().exportToDot("target.dot");
// abstractor->exportToDot("game" + std::to_string(iterations) + ".dot", targetStates, game.getManager().getBddOne());
// game.getReachableStates().template toAdd<ValueType>().exportToDot("reach.dot");
// #endif
// (3) compute all states with probability 0/1 wrt. to the two different player 2 goals (min/max).
auto qualitativeStart = std::chrono::high_resolution_clock::now();
QualitativeGameResultMinMax<Type> qualitativeResult = computeProb01States(previousQualitativeResult, game, player1Direction, transitionMatrixBdd, constraintStates, targetStates);
std::unique_ptr<CheckResult> result = checkForResultAfterQualitativeCheck<Type, ValueType>(checkTask, initialStates, qualitativeResult);
if (result) {
printStatistics(*abstractor, game);
return result;
}
previousQualitativeResult = qualitativeResult;
auto qualitativeEnd = std::chrono::high_resolution_clock::now();
STORM_LOG_DEBUG("Qualitative computation completed in " << std::chrono::duration_cast<std::chrono::milliseconds>(qualitativeEnd - qualitativeStart).count() << "ms.");
// (4) compute the states for which we have to determine quantitative information.
storm::dd::Bdd<Type> maybeMin = !(qualitativeResult.prob0Min.getPlayer1States() || qualitativeResult.prob1Min.getPlayer1States()) && game.getReachableStates();
storm::dd::Bdd<Type> maybeMax = !(qualitativeResult.prob0Max.getPlayer1States() || qualitativeResult.prob1Max.getPlayer1States()) && game.getReachableStates();
// (5) if the initial states are not maybe states, then we can refine at this point.
storm::dd::Bdd<Type> initialMaybeStates = (initialStates && maybeMin) || (initialStates && maybeMax);
bool qualitativeRefinement = false;
if (initialMaybeStates.isZero()) {
// In this case, we know the result for the initial states for both player 2 minimizing and maximizing.
STORM_LOG_TRACE("No initial state is a 'maybe' state.");
STORM_LOG_DEBUG("Obtained qualitative bounds [0, 1] on the actual value for the initial states. Refining abstraction based on qualitative check.");
// If we get here, the initial states were all identified as prob0/1 states, but the value (0 or 1)
// depends on whether player 2 is minimizing or maximizing. Therefore, we need to find a place to refine.
auto qualitativeRefinementStart = std::chrono::high_resolution_clock::now();
qualitativeRefinement = refiner.refine(game, transitionMatrixBdd, qualitativeResult);
auto qualitativeRefinementEnd = std::chrono::high_resolution_clock::now();
STORM_LOG_DEBUG("Qualitative refinement completed in " << std::chrono::duration_cast<std::chrono::milliseconds>(qualitativeRefinementEnd - qualitativeRefinementStart).count() << "ms.");
} else if (initialMaybeStates == initialStates && checkTask.isQualitativeSet()) {
// If all initial states are 'maybe' states and the property we needed to check is a qualitative one,
// we can return the result here.
return std::make_unique<ExplicitQuantitativeCheckResult<ValueType>>(storm::storage::sparse::state_type(0), ValueType(0.5));
}
// (6) if we arrived at this point and no refinement was made, we need to compute the quantitative solution.
if (!qualitativeRefinement) {
// At this point, we know that we cannot answer the query without further numeric computation.
storm::dd::Add<Type, ValueType> initialStatesAdd = initialStates.template toAdd<ValueType>();
STORM_LOG_TRACE("Starting numerical solution step.");
auto quantitativeStart = std::chrono::high_resolution_clock::now();
QuantitativeGameResultMinMax<Type, ValueType> quantitativeResult;
// (7) Solve the min values and check whether we can give the answer already.
quantitativeResult.min = computeQuantitativeResult(env, player1Direction, storm::OptimizationDirection::Minimize, game, qualitativeResult, initialStatesAdd, maybeMin, reuseQuantitativeResults ? previousMinQuantitativeResult : boost::none);
previousMinQuantitativeResult = quantitativeResult.min;
result = checkForResultAfterQuantitativeCheck<ValueType>(checkTask, storm::OptimizationDirection::Minimize, quantitativeResult.min.getInitialStatesRange());
if (result) {
printStatistics(*abstractor, game);
return result;
}
// (8) Solve the max values and check whether we can give the answer already.
quantitativeResult.max = computeQuantitativeResult(env, player1Direction, storm::OptimizationDirection::Maximize, game, qualitativeResult, initialStatesAdd, maybeMax, boost::make_optional(quantitativeResult.min));
result = checkForResultAfterQuantitativeCheck<ValueType>(checkTask, storm::OptimizationDirection::Maximize, quantitativeResult.max.getInitialStatesRange());
if (result) {
printStatistics(*abstractor, game);
return result;
}
auto quantitativeEnd = std::chrono::high_resolution_clock::now();
STORM_LOG_DEBUG("Obtained quantitative bounds [" << quantitativeResult.min.getInitialStatesRange().first << ", " << quantitativeResult.max.getInitialStatesRange().second << "] on the actual value for the initial states in " << std::chrono::duration_cast<std::chrono::milliseconds>(quantitativeEnd - quantitativeStart).count() << "ms.");
// (9) Check whether the lower and upper bounds are close enough to terminate with an answer.
result = checkForResultAfterQuantitativeCheck<ValueType>(quantitativeResult.min.getInitialStatesRange().first, quantitativeResult.max.getInitialStatesRange().second, comparator);
if (result) {
printStatistics(*abstractor, game);
return result;
}
// Make sure that all strategies are still valid strategies.
STORM_LOG_ASSERT(quantitativeResult.min.getPlayer1Strategy().isZero() || quantitativeResult.min.getPlayer1Strategy().template toAdd<ValueType>().sumAbstract(game.getPlayer1Variables()).getMax() <= 1, "Player 1 strategy for min is illegal.");
STORM_LOG_ASSERT(quantitativeResult.max.getPlayer1Strategy().isZero() || quantitativeResult.max.getPlayer1Strategy().template toAdd<ValueType>().sumAbstract(game.getPlayer1Variables()).getMax() <= 1, "Player 1 strategy for max is illegal.");
STORM_LOG_ASSERT(quantitativeResult.min.getPlayer2Strategy().isZero() || quantitativeResult.min.getPlayer2Strategy().template toAdd<ValueType>().sumAbstract(game.getPlayer2Variables()).getMax() <= 1, "Player 2 strategy for min is illegal.");
STORM_LOG_ASSERT(quantitativeResult.max.getPlayer2Strategy().isZero() || quantitativeResult.max.getPlayer2Strategy().template toAdd<ValueType>().sumAbstract(game.getPlayer2Variables()).getMax() <= 1, "Player 2 strategy for max is illegal.");
auto quantitativeRefinementStart = std::chrono::high_resolution_clock::now();
// (10) If we arrived at this point, it means that we have all qualitative and quantitative
// information about the game, but we could not yet answer the query. In this case, we need to refine.
refiner.refine(game, transitionMatrixBdd, quantitativeResult);
auto quantitativeRefinementEnd = std::chrono::high_resolution_clock::now();
STORM_LOG_DEBUG("Quantitative refinement completed in " << std::chrono::duration_cast<std::chrono::milliseconds>(quantitativeRefinementEnd - quantitativeRefinementStart).count() << "ms.");
}
auto iterationEnd = std::chrono::high_resolution_clock::now();
STORM_LOG_DEBUG("Iteration " << iterations << " took " << std::chrono::duration_cast<std::chrono::milliseconds>(iterationEnd - iterationStart).count() << "ms.");
}
STORM_LOG_ASSERT(false, "This point must not be reached.");
return nullptr;
}
template<storm::dd::DdType Type, typename ModelType>
std::vector<storm::expressions::Expression> GameBasedMdpModelChecker<Type, ModelType>::getInitialPredicates(storm::expressions::Expression const& constraintExpression, storm::expressions::Expression const& targetStateExpression) {
std::vector<storm::expressions::Expression> initialPredicates;
if (preprocessedModel.isJaniModel()) {
storm::expressions::VariableSetPredicateSplitter splitter(preprocessedModel.asJaniModel().getAllLocationExpressionVariables());
std::vector<storm::expressions::Expression> splitExpressions = splitter.split(targetStateExpression);
initialPredicates.insert(initialPredicates.end(), splitExpressions.begin(), splitExpressions.end());
splitExpressions = splitter.split(constraintExpression);
initialPredicates.insert(initialPredicates.end(), splitExpressions.begin(), splitExpressions.end());
} else {
if (!targetStateExpression.isTrue() && !targetStateExpression.isFalse()) {
initialPredicates.push_back(targetStateExpression);
}
if (!constraintExpression.isTrue() && !constraintExpression.isFalse()) {
initialPredicates.push_back(constraintExpression);
}
}
return initialPredicates;
}
template<storm::dd::DdType Type, typename ModelType>
storm::OptimizationDirection GameBasedMdpModelChecker<Type, ModelType>::getPlayer1Direction(CheckTask<storm::logic::Formula> const& checkTask) {
if (preprocessedModel.getModelType() == storm::storage::SymbolicModelDescription::ModelType::DTMC) {
return storm::OptimizationDirection::Maximize;
} else if (checkTask.isOptimizationDirectionSet()) {
return checkTask.getOptimizationDirection();
} else if (checkTask.isBoundSet() && preprocessedModel.getModelType() != storm::storage::SymbolicModelDescription::ModelType::DTMC) {
return storm::logic::isLowerBound(checkTask.getBoundComparisonType()) ? storm::OptimizationDirection::Minimize : storm::OptimizationDirection::Maximize;
}
STORM_LOG_THROW(false, storm::exceptions::InvalidPropertyException, "Could not derive player 1 optimization direction.");
return storm::OptimizationDirection::Maximize;
}
template<storm::dd::DdType Type>
bool checkQualitativeStrategies(bool prob0, QualitativeGameResult<Type> const& result, storm::dd::Bdd<Type> const& targetStates) {
if (prob0) {
STORM_LOG_ASSERT(result.hasPlayer1Strategy() && (result.getPlayer1States().isZero() || !result.getPlayer1Strategy().isZero()), "Unable to proceed without strategy.");
} else {
STORM_LOG_ASSERT(result.hasPlayer1Strategy() && ((result.getPlayer1States() && !targetStates).isZero() || !result.getPlayer1Strategy().isZero()), "Unable to proceed without strategy.");
}
STORM_LOG_ASSERT(result.hasPlayer2Strategy() && (result.getPlayer2States().isZero() || !result.getPlayer2Strategy().isZero()), "Unable to proceed without strategy.");
return true;
}
template<storm::dd::DdType Type>
bool checkQualitativeStrategies(QualitativeGameResultMinMax<Type> const& qualitativeResult, storm::dd::Bdd<Type> const& targetStates) {
bool result = true;
result &= checkQualitativeStrategies(true, qualitativeResult.prob0Min, targetStates);
result &= checkQualitativeStrategies(false, qualitativeResult.prob1Min, targetStates);
result &= checkQualitativeStrategies(true, qualitativeResult.prob0Max, targetStates);
result &= checkQualitativeStrategies(false, qualitativeResult.prob1Max, targetStates);
return result;
}
template<storm::dd::DdType Type, typename ModelType>
QualitativeGameResultMinMax<Type> GameBasedMdpModelChecker<Type, ModelType>::computeProb01States(boost::optional<QualitativeGameResultMinMax<Type>> const& previousQualitativeResult, storm::abstraction::MenuGame<Type, ValueType> const& game, storm::OptimizationDirection player1Direction, storm::dd::Bdd<Type> const& transitionMatrixBdd, storm::dd::Bdd<Type> const& constraintStates, storm::dd::Bdd<Type> const& targetStates) {
QualitativeGameResultMinMax<Type> result;
if (reuseQualitativeResults) {
// Depending on the player 1 direction, we choose a different order of operations.
if (player1Direction == storm::OptimizationDirection::Minimize) {
// (1) min/min: compute prob0 using the game functions
result.prob0Min = storm::utility::graph::performProb0(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Minimize, true, true);
// (2) min/min: compute prob1 using the MDP functions
storm::dd::Bdd<Type> candidates = game.getReachableStates() && !result.prob0Min.player1States;
storm::dd::Bdd<Type> prob1MinMinMdp = storm::utility::graph::performProb1A(game, transitionMatrixBdd, previousQualitativeResult ? previousQualitativeResult.get().prob1Min.player1States : targetStates, candidates);
// (3) min/min: compute prob1 using the game functions
result.prob1Min = storm::utility::graph::performProb1(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Minimize, true, true, boost::make_optional(prob1MinMinMdp));
// (4) min/max: compute prob 0 using the game functions
result.prob0Max = storm::utility::graph::performProb0(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Maximize, true, true);
// (5) min/max: compute prob 1 using the game functions
// We know that only previous prob1 states can now be prob 1 states again, because the upper bound
// values can only decrease over iterations.
boost::optional<storm::dd::Bdd<Type>> prob1Candidates;
if (previousQualitativeResult) {
prob1Candidates = previousQualitativeResult.get().prob1Max.player1States;
}
result.prob1Max = storm::utility::graph::performProb1(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Maximize, true, true, prob1Candidates);
} else {
// (1) max/max: compute prob0 using the game functions
result.prob0Max = storm::utility::graph::performProb0(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Maximize, true, true);
// (2) max/max: compute prob1 using the MDP functions, reuse prob1 states of last iteration to constrain the candidate states.
storm::dd::Bdd<Type> candidates = game.getReachableStates() && !result.prob0Max.player1States;
if (previousQualitativeResult) {
candidates &= previousQualitativeResult.get().prob1Max.player1States;
}
storm::dd::Bdd<Type> prob1MaxMaxMdp = storm::utility::graph::performProb1E(game, transitionMatrixBdd, constraintStates, targetStates, candidates);
// (3) max/max: compute prob1 using the game functions, reuse prob1 states from the MDP precomputation
result.prob1Max = storm::utility::graph::performProb1(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Maximize, true, true, boost::make_optional(prob1MaxMaxMdp));
// (4) max/min: compute prob0 using the game functions
result.prob0Min = storm::utility::graph::performProb0(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Minimize, true, true);
// (5) max/min: compute prob1 using the game functions, use prob1 from max/max as the candidate set
result.prob1Min = storm::utility::graph::performProb1(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Minimize, true, true, boost::make_optional(prob1MaxMaxMdp));
}
} else {
result.prob0Min = storm::utility::graph::performProb0(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Minimize, true, true);
result.prob1Min = storm::utility::graph::performProb1(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Minimize, true, true);
result.prob0Max = storm::utility::graph::performProb0(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Maximize, true, true);
result.prob1Max = storm::utility::graph::performProb1(game, transitionMatrixBdd, constraintStates, targetStates, player1Direction, storm::OptimizationDirection::Maximize, true, true);
}
STORM_LOG_TRACE("Qualitative precomputation completed.");
STORM_LOG_TRACE("[" << player1Direction << ", " << storm::OptimizationDirection::Minimize << "]: " << result.prob0Min.player1States.getNonZeroCount() << " 'no', " << result.prob1Min.player1States.getNonZeroCount() << " 'yes'.");
STORM_LOG_TRACE("[" << player1Direction << ", " << storm::OptimizationDirection::Maximize << "]: " << result.prob0Max.player1States.getNonZeroCount() << " 'no', " << result.prob1Max.player1States.getNonZeroCount() << " 'yes'.");
STORM_LOG_ASSERT(checkQualitativeStrategies(result, targetStates), "Qualitative strategies appear to be broken.");
return result;
}
template<storm::dd::DdType Type, typename ModelType>
void GameBasedMdpModelChecker<Type, ModelType>::printStatistics(storm::abstraction::MenuGameAbstractor<Type, ValueType> const& abstractor, storm::abstraction::MenuGame<Type, ValueType> const& game) const {
if (storm::settings::getModule<storm::settings::modules::CoreSettings>().isShowStatisticsSet()) {
storm::abstraction::AbstractionInformation<Type> const& abstractionInformation = abstractor.getAbstractionInformation();
std::cout << std::endl;
std::cout << "Statistics:" << std::endl;
std::cout << " * player 1 states (final game): " << game.getReachableStates().getNonZeroCount() << std::endl;
std::cout << " * transitions (final game): " << game.getTransitionMatrix().getNonZeroCount() << std::endl;
std::cout << " * predicates used in abstraction: " << abstractionInformation.getNumberOfPredicates() << std::endl;
}
}
template<storm::dd::DdType Type, typename ModelType>
storm::expressions::Expression GameBasedMdpModelChecker<Type, ModelType>::getExpression(storm::logic::Formula const& formula) {
STORM_LOG_THROW(formula.isBooleanLiteralFormula() || formula.isAtomicExpressionFormula() || formula.isAtomicLabelFormula(), storm::exceptions::InvalidPropertyException, "The target states have to be given as label or an expression.");
storm::expressions::Expression result;
if (formula.isAtomicLabelFormula()) {
result = preprocessedModel.asPrismProgram().getLabelExpression(formula.asAtomicLabelFormula().getLabel());
} else if (formula.isAtomicExpressionFormula()) {
result = formula.asAtomicExpressionFormula().getExpression();
} else {
result = formula.asBooleanLiteralFormula().isTrueFormula() ? preprocessedModel.getManager().boolean(true) : preprocessedModel.getManager().boolean(false);
}
return result;
}
template class GameBasedMdpModelChecker<storm::dd::DdType::CUDD, storm::models::symbolic::Dtmc<storm::dd::DdType::CUDD, double>>;
template class GameBasedMdpModelChecker<storm::dd::DdType::CUDD, storm::models::symbolic::Mdp<storm::dd::DdType::CUDD, double>>;
template class GameBasedMdpModelChecker<storm::dd::DdType::Sylvan, storm::models::symbolic::Dtmc<storm::dd::DdType::Sylvan, double>>;
template class GameBasedMdpModelChecker<storm::dd::DdType::Sylvan, storm::models::symbolic::Mdp<storm::dd::DdType::Sylvan, double>>;
}
}