Class TrainingState

• Defined in File training_state.h

Class Documentation

class TrainingState

Public Functions

void updateEta(float dynamicBaseLR)

TrainingState(float learnRate)

void registerObserver(Ptr<TrainingObserver> observer)

size_t getProgressIn(SchedulingUnit u) const

void rememberPreviousProgress()

size_t getPreviousProgressIn(SchedulingUnit u) const

bool enteredNewPeriodOf(std::string schedulingParam) const

void newEpoch()

void newUpdate(size_t batchesInUpdate)

void newStalled(size_t num)

void newLoad()

void load(const std::string &name)

void save(const std::string &name) const

std::string fillTemplate(const std::string &templ) const

Public Members

size_t epochs = {1}

size_t batches = {0}

size_t batchesEpoch = {0}

size_t samplesEpoch = {0}

size_t labelsTotal = {0}

size_t prevLabelsTotal = {0}

size_t prevBatches = {0}

size_t prevEpochs = {0}

size_t stalled = {0}

size_t maxStalled = {0}

std::string validator

YAML::Node validators

bool reset = {false}

float eta

float factor = {1.f}

SchedulingParameter warmupStart

float costSum = {0}

float costCount = {0}

size_t wordsDisp = {0}

size_t samplesDisp = {0}

size_t updatesDisp = {0}

float gradientNormAvg = {0}

float gradientNormVar = {0}

float logGradientNormAvg = {0}

float logGradientNormVar = {0}

std::string seedBatch

std::string seedCorpus

bool loaded = {false}

bool validated = {false}