Class UCBPoolBasedActiveDyadRanker


  • public class UCBPoolBasedActiveDyadRanker
    extends ARandomlyInitializingDyadRanker
    A prototypical active dyad ranker based on the UCB decision rule. During the learning procedure, it keeps track over the standard deviation of the skill values predicted for a dyad. First a constant number of random queries is sampled at the beginning. Then the sampling strategy randomly selects problem instances and picks the two dyads with largest skill + standard deviation for pairwise comparison. On each query step, this is repeated a constant number of times to create a minibatch.
    • Constructor Detail

      • UCBPoolBasedActiveDyadRanker

        public UCBPoolBasedActiveDyadRanker​(PLNetDyadRanker ranker,
                                            IDyadRankingPoolProvider poolProvider,
                                            int seed,
                                            int numberRandomQueriesAtStart,
                                            int minibatchSize)
    • Method Detail

      • activelyTrainWithOneInstance

        public void activelyTrainWithOneInstance()
                                          throws org.api4.java.ai.ml.core.exception.TrainingException,
                                                 java.lang.InterruptedException
        Specified by:
        activelyTrainWithOneInstance in class ARandomlyInitializingDyadRanker
        Throws:
        org.api4.java.ai.ml.core.exception.TrainingException
        java.lang.InterruptedException