#include "src/models/sparse/MarkovAutomaton.h"
#include "src/models/sparse/StandardRewardModel.h"
#include "src/exceptions/InvalidArgumentException.h"
#include "src/utility/constants.h"
#include "src/adapters/CarlAdapter.h"
#include "src/storage/FlexibleSparseMatrix.h"
#include "src/models/sparse/Dtmc.h"
#include "src/solver/stateelimination/MAEliminator.h"
#include "src/utility/vector.h"
namespace storm {
namespace models {
namespace sparse {
template <typename ValueType, typename RewardModelType>
MarkovAutomaton<ValueType, RewardModelType>::MarkovAutomaton(storm::storage::SparseMatrix<ValueType> const& transitionMatrix,
storm::models::sparse::StateLabeling const& stateLabeling,
storm::storage::BitVector const& markovianStates,
std::vector<ValueType> const& exitRates,
std::unordered_map<std::string, RewardModelType> const& rewardModels,
boost::optional<std::vector<LabelSet>> const& optionalChoiceLabeling)
: NondeterministicModel<ValueType, RewardModelType>(storm::models::ModelType::MarkovAutomaton, transitionMatrix, stateLabeling, rewardModels, optionalChoiceLabeling), markovianStates(markovianStates), exitRates(exitRates), closed(this->checkIsClosed()) {
this->turnRatesToProbabilities();
}
template <typename ValueType, typename RewardModelType>
MarkovAutomaton<ValueType, RewardModelType>::MarkovAutomaton(storm::storage::SparseMatrix<ValueType>&& transitionMatrix,
storm::models::sparse::StateLabeling&& stateLabeling,
storm::storage::BitVector const& markovianStates,
std::vector<ValueType> const& exitRates,
std::unordered_map<std::string, RewardModelType>&& rewardModels,
boost::optional<std::vector<LabelSet>>&& optionalChoiceLabeling)
: NondeterministicModel<ValueType, RewardModelType>(storm::models::ModelType::MarkovAutomaton, std::move(transitionMatrix), std::move(stateLabeling), std::move(rewardModels), std::move(optionalChoiceLabeling)), markovianStates(markovianStates), exitRates(std::move(exitRates)), closed(this->checkIsClosed()) {
this->turnRatesToProbabilities();
}
template <typename ValueType, typename RewardModelType>
MarkovAutomaton<ValueType, RewardModelType>::MarkovAutomaton(storm::storage::SparseMatrix<ValueType>&& transitionMatrix,
storm::models::sparse::StateLabeling&& stateLabeling,
storm::storage::BitVector const& markovianStates,
std::vector<ValueType> const& exitRates,
bool probabilities,
std::unordered_map<std::string, RewardModelType>&& rewardModels,
boost::optional<std::vector<LabelSet>>&& optionalChoiceLabeling)
: NondeterministicModel<ValueType, RewardModelType>(storm::models::ModelType::MarkovAutomaton, std::move(transitionMatrix), std::move(stateLabeling), std::move(rewardModels), std::move(optionalChoiceLabeling)), markovianStates(markovianStates), exitRates(std::move(exitRates)), closed(this->checkIsClosed()) {
assert(probabilities);
assert(this->getTransitionMatrix().isProbabilistic());
}
template <typename ValueType, typename RewardModelType>
bool MarkovAutomaton<ValueType, RewardModelType>::isClosed() const {
return closed;
}
template <typename ValueType, typename RewardModelType>
bool MarkovAutomaton<ValueType, RewardModelType>::isHybridState(storm::storage::sparse::state_type state) const {
return isMarkovianState(state) && (this->getTransitionMatrix().getRowGroupSize(state) > 1);
}
template <typename ValueType, typename RewardModelType>
bool MarkovAutomaton<ValueType, RewardModelType>::isMarkovianState(storm::storage::sparse::state_type state) const {
return this->markovianStates.get(state);
}
template <typename ValueType, typename RewardModelType>
bool MarkovAutomaton<ValueType, RewardModelType>::isProbabilisticState(storm::storage::sparse::state_type state) const {
return !this->markovianStates.get(state);
}
template <typename ValueType, typename RewardModelType>
std::vector<ValueType> const& MarkovAutomaton<ValueType, RewardModelType>::getExitRates() const {
return this->exitRates;
}
template <typename ValueType, typename RewardModelType>
std::vector<ValueType>& MarkovAutomaton<ValueType, RewardModelType>::getExitRates() {
return this->exitRates;
}
template <typename ValueType, typename RewardModelType>
ValueType const& MarkovAutomaton<ValueType, RewardModelType>::getExitRate(storm::storage::sparse::state_type state) const {
return this->exitRates[state];
}
template <typename ValueType, typename RewardModelType>
ValueType MarkovAutomaton<ValueType, RewardModelType>::getMaximalExitRate() const {
ValueType result = storm::utility::zero<ValueType>();
for (auto markovianState : this->markovianStates) {
result = std::max(result, this->exitRates[markovianState]);
}
return result;
}
template <typename ValueType, typename RewardModelType>
storm::storage::BitVector const& MarkovAutomaton<ValueType, RewardModelType>::getMarkovianStates() const {
return this->markovianStates;
}
template <typename ValueType, typename RewardModelType>
void MarkovAutomaton<ValueType, RewardModelType>::close() {
if (!closed) {
// First, count the number of hybrid states to know how many Markovian choices
// will be removed.
uint_fast64_t numberOfHybridStates = 0;
for (uint_fast64_t state = 0; state < this->getNumberOfStates(); ++state) {
if (this->isHybridState(state)) {
++numberOfHybridStates;
}
}
// Create the matrix for the new transition relation and the corresponding nondeterministic choice vector.
storm::storage::SparseMatrixBuilder<ValueType> newTransitionMatrixBuilder(0, 0, 0, false, true, this->getNumberOfStates());
// Now copy over all choices that need to be kept.
uint_fast64_t currentChoice = 0;
for (uint_fast64_t state = 0; state < this->getNumberOfStates(); ++state) {
// Record the new beginning of choices of this state.
newTransitionMatrixBuilder.newRowGroup(currentChoice);
// If the state is a hybrid state, closing it will make it a probabilistic state, so we remove the Markovian marking.
// Additionally, we need to remember whether we need to skip the first choice of the state when
// we assemble the new transition matrix.
uint_fast64_t offset = 0;
if (this->isHybridState(state)) {
this->markovianStates.set(state, false);
offset = 1;
}
for (uint_fast64_t row = this->getTransitionMatrix().getRowGroupIndices()[state] + offset; row < this->getTransitionMatrix().getRowGroupIndices()[state + 1]; ++row) {
for (auto const& entry : this->getTransitionMatrix().getRow(row)) {
newTransitionMatrixBuilder.addNextValue(currentChoice, entry.getColumn(), entry.getValue());
}
++currentChoice;
}
}
// Finalize the matrix and put the new transition data in place.
this->setTransitionMatrix(newTransitionMatrixBuilder.build());
// Mark the automaton as closed.
closed = true;
}
}
template <typename ValueType, typename RewardModelType>
void MarkovAutomaton<ValueType, RewardModelType>::writeDotToStream(std::ostream& outStream, bool includeLabeling, storm::storage::BitVector const* subsystem, std::vector<ValueType> const* firstValue, std::vector<ValueType> const* secondValue, std::vector<uint_fast64_t> const* stateColoring, std::vector<std::string> const* colors, std::vector<uint_fast64_t>* scheduler, bool finalizeOutput) const {
NondeterministicModel<ValueType, RewardModelType>::writeDotToStream(outStream, includeLabeling, subsystem, firstValue, secondValue, stateColoring, colors, scheduler, false);
// Write the probability distributions for all the states.
for (uint_fast64_t state = 0; state < this->getNumberOfStates(); ++state) {
uint_fast64_t rowCount = this->getTransitionMatrix().getRowGroupIndices()[state + 1] - this->getTransitionMatrix().getRowGroupIndices()[state];
bool highlightChoice = true;
// For this, we need to iterate over all available nondeterministic choices in the current state.
for (uint_fast64_t choice = 0; choice < rowCount; ++choice) {
typename storm::storage::SparseMatrix<ValueType>::const_rows row = this->getTransitionMatrix().getRow(this->getTransitionMatrix().getRowGroupIndices()[state] + choice);
if (scheduler != nullptr) {
// If the scheduler picked the current choice, we will not make it dotted, but highlight it.
if ((*scheduler)[state] == choice) {
highlightChoice = true;
} else {
highlightChoice = false;
}
}
// If it's not a Markovian state or the current row is the first choice for this state, then we
// are dealing with a probabilitic choice.
if (!markovianStates.get(state) || choice != 0) {
// For each nondeterministic choice, we draw an arrow to an intermediate node to better display
// the grouping of transitions.
outStream << "\t\"" << state << "c" << choice << "\" [shape = \"point\"";
// If we were given a scheduler to highlight, we do so now.
if (scheduler != nullptr) {
if (highlightChoice) {
outStream << ", fillcolor=\"red\"";
}
}
outStream << "];" << std::endl;
outStream << "\t" << state << " -> \"" << state << "c" << choice << "\"";
// If we were given a scheduler to highlight, we do so now.
if (scheduler != nullptr) {
if (highlightChoice) {
outStream << " [color=\"red\", penwidth = 2]";
} else {
outStream << " [style = \"dotted\"]";
}
}
outStream << ";" << std::endl;
// Now draw all probabilitic arcs that belong to this nondeterminstic choice.
for (auto const& transition : row) {
if (subsystem == nullptr || subsystem->get(transition.getColumn())) {
outStream << "\t\"" << state << "c" << choice << "\" -> " << transition.getColumn() << " [ label= \"" << transition.getValue() << "\" ]";
// If we were given a scheduler to highlight, we do so now.
if (scheduler != nullptr) {
if (highlightChoice) {
outStream << " [color=\"red\", penwidth = 2]";
} else {
outStream << " [style = \"dotted\"]";
}
}
outStream << ";" << std::endl;
}
}
} else {
// In this case we are emitting a Markovian choice, so draw the arrows directly to the target states.
for (auto const& transition : row) {
if (subsystem == nullptr || subsystem->get(transition.getColumn())) {
outStream << "\t\"" << state << "\" -> " << transition.getColumn() << " [ label= \"" << transition.getValue() << " (" << this->exitRates[state] << ")\" ]";
}
}
}
}
}
if (finalizeOutput) {
outStream << "}" << std::endl;
}
}
template <typename ValueType, typename RewardModelType>
std::size_t MarkovAutomaton<ValueType, RewardModelType>::getSizeInBytes() const {
return NondeterministicModel<ValueType, RewardModelType>::getSizeInBytes() + markovianStates.getSizeInBytes() + exitRates.size() * sizeof(ValueType);
}
template <typename ValueType, typename RewardModelType>
void MarkovAutomaton<ValueType, RewardModelType>::turnRatesToProbabilities() {
for (auto state : this->markovianStates) {
for (auto& transition : this->getTransitionMatrix().getRow(this->getTransitionMatrix().getRowGroupIndices()[state])) {
transition.setValue(transition.getValue() / this->exitRates[state]);
}
}
}
template <typename ValueType, typename RewardModelType>
bool MarkovAutomaton<ValueType, RewardModelType>::hasOnlyTrivialNondeterminism() const {
// Check every state
for (uint_fast64_t state = 0; state < this->getNumberOfStates(); ++state) {
// Get number of choices in current state
uint_fast64_t numberChoices = this->getTransitionMatrix().getRowGroupIndices()[state + 1] - this->getTransitionMatrix().getRowGroupIndices()[state];
if (isMarkovianState(state)) {
assert(numberChoices == 1);
}
if (numberChoices > 1) {
assert(isProbabilisticState(state));
return false;
}
}
return true;
}
template <typename ValueType, typename RewardModelType>
bool MarkovAutomaton<ValueType, RewardModelType>::checkIsClosed() const {
for (auto state : markovianStates) {
if (this->getTransitionMatrix().getRowGroupSize(state) > 1) {
return false;
}
}
return true;
}
template <typename ValueType, typename RewardModelType>
std::shared_ptr<storm::models::sparse::Ctmc<ValueType, RewardModelType>> MarkovAutomaton<ValueType, RewardModelType>::convertToCTMC() {
STORM_LOG_TRACE("MA matrix:" << std::endl << this->getTransitionMatrix());
STORM_LOG_TRACE("Markovian states: " << getMarkovianStates());
// Eliminate all probabilistic states by state elimination
// Initialize
storm::storage::FlexibleSparseMatrix<ValueType> flexibleMatrix(this->getTransitionMatrix());
storm::storage::FlexibleSparseMatrix<ValueType> flexibleBackwardTransitions(this->getTransitionMatrix().transpose());
storm::solver::stateelimination::MAEliminator<storm::models::sparse::Dtmc<ValueType>> stateEliminator(flexibleMatrix, flexibleBackwardTransitions);
for (uint_fast64_t state = 0; state < this->getNumberOfStates(); ++state) {
assert(!this->isHybridState(state));
if (this->isProbabilisticState(state)) {
// Eliminate this probabilistic state
stateEliminator.eliminateState(state, true);
STORM_LOG_TRACE("Flexible matrix after eliminating state " << state << ":" << std::endl << flexibleMatrix);
}
}
// Create the rate matrix for the CTMC
storm::storage::SparseMatrixBuilder<ValueType> transitionMatrixBuilder(0, 0, 0, false, false);
// Remember state to keep
storm::storage::BitVector keepStates(this->getNumberOfStates(), true);
for (uint_fast64_t state = 0; state < this->getNumberOfStates(); ++state) {
if (storm::utility::isZero(flexibleMatrix.getRowSum(state))) {
// State is eliminated and can be discarded
keepStates.set(state, false);
} else {
assert(this->isMarkovianState(state));
// Copy transitions
for (uint_fast64_t row = flexibleMatrix.getRowGroupIndices()[state]; row < flexibleMatrix.getRowGroupIndices()[state + 1]; ++row) {
for (auto const& entry : flexibleMatrix.getRow(row)) {
// Convert probabilities into rates
transitionMatrixBuilder.addNextValue(state, entry.getColumn(), entry.getValue() * exitRates[state]);
}
}
}
}
storm::storage::SparseMatrix<ValueType> rateMatrix = transitionMatrixBuilder.build();
rateMatrix = rateMatrix.getSubmatrix(false, keepStates, keepStates, false);
STORM_LOG_TRACE("New CTMC matrix:" << std::endl << rateMatrix);
// Construct CTMC
storm::models::sparse::StateLabeling stateLabeling = this->getStateLabeling().getSubLabeling(keepStates);
boost::optional<std::vector<LabelSet>> optionalChoiceLabeling = this->getOptionalChoiceLabeling();
if (optionalChoiceLabeling) {
optionalChoiceLabeling = storm::utility::vector::filterVector(optionalChoiceLabeling.get(), keepStates);
}
//TODO update reward models according to kept states
std::unordered_map<std::string, RewardModelType> rewardModels = this->getRewardModels();
return std::make_shared<storm::models::sparse::Ctmc<ValueType, RewardModelType>>(std::move(rateMatrix), std::move(stateLabeling), std::move(rewardModels), std::move(optionalChoiceLabeling));
}
template class MarkovAutomaton<double>;
// template class MarkovAutomaton<float>;
#ifdef STORM_HAVE_CARL
template class MarkovAutomaton<double, storm::models::sparse::StandardRewardModel<storm::Interval>>;
template class MarkovAutomaton<storm::RationalFunction>;
#endif
} // namespace sparse
} // namespace models
} // namespace storm