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

DirectEncodingParser: Added support for Action-based rewards.

tempestpy_adaptions
Tim Quatmann 6 years ago
parent
429c91ff13
commit
42b7865e7e
2 changed files with 66 additions and 10 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. 62
      src/storm-parsers/parser/DirectEncodingParser.cpp
  2. 14
      src/test/storm/parser/DirectEncodingParserTest.cpp

62
src/storm-parsers/parser/DirectEncodingParser.cpp
View File

@ -114,6 +114,7 @@ namespace storm {
modelComponents->observabilityClasses = std::vector<uint32_t>();
modelComponents->observabilityClasses->resize(stateSize);
std::vector<std::vector<ValueType>> stateRewards;
std::vector<std::vector<ValueType>> actionRewards;
if (continuousTime) {
modelComponents->exitRates = std::vector<ValueType>(stateSize);
if (type == storm::models::ModelType::MarkovAutomaton) {
@ -128,6 +129,7 @@ namespace storm {
// Iterate over all lines
std::string line;
size_t row = 0;
size_t firstRowOfState = 0;
size_t state = 0;
bool firstState = true;
bool firstActionForState = true;
@ -160,6 +162,7 @@ namespace storm {
STORM_LOG_TRACE("new Row Group starts at " << row << ".");
builder.newRowGroup(row);
}
firstRowOfState = row;
if (type == storm::models::ModelType::Ctmc || type == storm::models::ModelType::MarkovAutomaton) {
// Parse exit rate for CTMC or MA
@ -190,14 +193,19 @@ namespace storm {
std::vector<std::string> rewards;
boost::split(rewards, rewardsStr, boost::is_any_of(","));
if (stateRewards.size() < rewards.size()) {
stateRewards.resize(rewards.size(), std::vector<ValueType>(stateSize, storm::utility::zero<ValueType>()));
stateRewards.resize(rewards.size());
}
auto stateRewardsIt = stateRewards.begin();
for (auto const& rew : rewards) {
(*stateRewardsIt)[state] = valueParser.parseValue(rew);
auto rewardValue = valueParser.parseValue(rew);
if (!storm::utility::isZero(rewardValue)) {
if (stateRewardsIt->empty()) {
stateRewardsIt->resize(stateSize, storm::utility::zero<ValueType>());
}
(*stateRewardsIt)[row] = std::move(rewardValue);
}
++stateRewardsIt;
}
line = line.substr(posEndReward+1);
}
@ -258,15 +266,39 @@ namespace storm {
}
STORM_LOG_TRACE("New action: " << row);
line = line.substr(8); //Remove "\taction "
std::string curString = line;
size_t posEnd = line.find(" ");
if (posEnd != std::string::npos) {
curString = line.substr(0, posEnd);
line = line.substr(posEnd+1);
} else {
line = "";
}
size_t parsedId = parseNumber<size_t>(curString);
STORM_LOG_ASSERT(row == firstRowOfState + parsedId, "Action ids do not correspond.");
// Check for rewards
if (boost::starts_with(line, "[")) {
// Rewards found
size_t posEndReward = line.find(']');
STORM_LOG_THROW(posEndReward != std::string::npos, storm::exceptions::WrongFormatException, "] missing.");
std::string rewards = line.substr(1, posEndReward-1);
STORM_LOG_TRACE("Transition rewards: " << rewards);
STORM_LOG_WARN("Transition rewards [" << rewards << "] not parsed.");
// TODO save rewards
std::string rewardsStr = line.substr(1, posEndReward-1);
STORM_LOG_TRACE("Action rewards: " << rewardsStr);
std::vector<std::string> rewards;
boost::split(rewards, rewardsStr, boost::is_any_of(","));
if (actionRewards.size() < rewards.size()) {
actionRewards.resize(rewards.size());
}
auto actionRewardsIt = actionRewards.begin();
for (auto const& rew : rewards) {
auto rewardValue = valueParser.parseValue(rew);
if (!storm::utility::isZero(rewardValue)) {
if (actionRewardsIt->size() <= row) {
actionRewardsIt->resize(row + 1, storm::utility::zero<ValueType>());
}
(*actionRewardsIt)[row] = std::move(rewardValue);
}
++actionRewardsIt;
}
line = line.substr(posEndReward+1);
}
// TODO import choice labeling when the export works
@ -286,17 +318,27 @@ namespace storm {
STORM_LOG_TRACE("Finished parsing");
modelComponents->transitionMatrix = builder.build(row + 1, stateSize, nonDeterministic ? stateSize : 0);
STORM_LOG_TRACE("Built matrix");
for (uint64_t i = 0; i < stateRewards.size(); ++i) {
uint64_t numRewardModels = std::max(stateRewards.size(), actionRewards.size());
for (uint64_t i = 0; i < numRewardModels; ++i) {
std::string rewardModelName;
if (rewardModelNames.size() <= i) {
rewardModelName = "rew" + std::to_string(i);
} else {
rewardModelName = rewardModelNames[i];
}
modelComponents->rewardModels.emplace(rewardModelName, storm::models::sparse::StandardRewardModel<ValueType>(std::move(stateRewards[i])));
boost::optional<std::vector<ValueType>> stateRewardVector, actionRewardVector;
if (!stateRewards[i].empty()) {
stateRewardVector = std::move(stateRewards[i]);
}
if (!actionRewards[i].empty()) {
actionRewards[i].resize(row + 1, storm::utility::zero<ValueType>());
actionRewardVector = std::move(actionRewards[i]);
}
modelComponents->rewardModels.emplace(rewardModelName, storm::models::sparse::StandardRewardModel<ValueType>(std::move(stateRewardVector), std::move(actionRewardVector)));
}
STORM_LOG_TRACE("Built matrix");
STORM_LOG_TRACE("Built reward models");
return modelComponents;
}

14
src/test/storm/parser/DirectEncodingParserTest.cpp
View File

@ -35,6 +35,11 @@ TEST(DirectEncodingParserTest, MdpParsing) {
ASSERT_EQ(5ul, modelPtr->getStates("six").getNumberOfSetBits());
ASSERT_TRUE(modelPtr->hasLabel("eleven"));
ASSERT_EQ(2ul, modelPtr->getStates("eleven").getNumberOfSetBits());
ASSERT_EQ(1ul, modelPtr->getNumberOfRewardModels());
ASSERT_TRUE(modelPtr->hasRewardModel("coinflips"));
ASSERT_TRUE(!modelPtr->getRewardModel("coinflips").hasStateRewards());
ASSERT_TRUE(modelPtr->getRewardModel("coinflips").hasStateActionRewards());
ASSERT_TRUE(!modelPtr->getRewardModel("coinflips").isAllZero());
}
TEST(DirectEncodingParserTest, CtmcParsing) {
@ -50,6 +55,11 @@ TEST(DirectEncodingParserTest, CtmcParsing) {
ASSERT_EQ(64ul, modelPtr->getStates("premium").getNumberOfSetBits());
ASSERT_TRUE(modelPtr->hasLabel("minimum"));
ASSERT_EQ(132ul, modelPtr->getStates("minimum").getNumberOfSetBits());
ASSERT_EQ(1ul, modelPtr->getNumberOfRewardModels());
ASSERT_TRUE(modelPtr->hasRewardModel("num_repairs"));
ASSERT_TRUE(!modelPtr->getRewardModel("num_repairs").hasStateRewards());
ASSERT_TRUE(modelPtr->getRewardModel("num_repairs").hasStateActionRewards());
ASSERT_TRUE(!modelPtr->getRewardModel("num_repairs").isAllZero());
}
TEST(DirectEncodingParserTest, MarkovAutomatonParsing) {
@ -63,7 +73,11 @@ TEST(DirectEncodingParserTest, MarkovAutomatonParsing) {
ASSERT_EQ(19ul, ma->getNumberOfChoices());
ASSERT_EQ(10ul, ma->getMarkovianStates().getNumberOfSetBits());
ASSERT_EQ(5, ma->getMaximalExitRate());
ASSERT_EQ(1ul, ma->getNumberOfRewardModels());
ASSERT_TRUE(ma->hasRewardModel("avg_waiting_time"));
ASSERT_TRUE(ma->getRewardModel("avg_waiting_time").hasStateRewards());
ASSERT_TRUE(!ma->getRewardModel("avg_waiting_time").hasStateActionRewards());
ASSERT_TRUE(!ma->getRewardModel("avg_waiting_time").isAllZero());
ASSERT_TRUE(modelPtr->hasLabel("init"));
ASSERT_EQ(1ul, modelPtr->getInitialStates().getNumberOfSetBits());
ASSERT_TRUE(modelPtr->hasLabel("one_job_finished"));

Write Preview
Loading…
Cancel
Save