Java example source code file (ParameterAveragingElementCombineFunction.java)

This example Java source code file (ParameterAveragingElementCombineFunction.java) is included in the alvinalexander.com "Java Source Code Warehouse" project. The intent of this project is to help you "Learn Java by Example" ^TM.

Learn more about this Java project at its project page.

Java - Java tags/keywords

exception, function2, indarray, override, parameteraveragingaggregationtuple, parameteraveragingelementcombinefunction, sparktrainingstats, updateraggregator

The ParameterAveragingElementCombineFunction.java Java example source code

package org.deeplearning4j.spark.impl.paramavg.aggregator;

import org.apache.spark.api.java.function.Function2;
import org.deeplearning4j.nn.updater.aggregate.UpdaterAggregator;
import org.deeplearning4j.nn.updater.graph.ComputationGraphUpdater;
import org.deeplearning4j.spark.api.stats.SparkTrainingStats;
import org.nd4j.linalg.api.ndarray.INDArray;

/**
 * Function used in ParameterAveraging TrainingMaster, for doing parameter averaging, and handling updaters
 *
 * @author Alex Black
 */
public class ParameterAveragingElementCombineFunction implements Function2<ParameterAveragingAggregationTuple, ParameterAveragingAggregationTuple, ParameterAveragingAggregationTuple> {
    @Override
    public ParameterAveragingAggregationTuple call(ParameterAveragingAggregationTuple v1, ParameterAveragingAggregationTuple v2) throws Exception {
        if(v1 == null) return v2;
        else if(v2 == null) return v1;

        //Handle edge case of less data than executors: in this case, one (or both) of v1 and v2 might not have any contents...
        if(v1.getParametersSum() == null) return v2;
        else if(v2.getParametersSum() == null) return v1;

        INDArray newParams = v1.getParametersSum().addi(v2.getParametersSum());

        UpdaterAggregator updaterAggregator = v1.getUpdaterAggregator();
        UpdaterAggregator updaterAggregator2 = v2.getUpdaterAggregator();
        UpdaterAggregator combinedAggregator;
        if(updaterAggregator == null) combinedAggregator = updaterAggregator2;
        else if(updaterAggregator2 == null) combinedAggregator = updaterAggregator;
        else{
            updaterAggregator.merge(updaterAggregator2);
            combinedAggregator = updaterAggregator;
        }

        ComputationGraphUpdater.Aggregator uAGraph1 = v1.getUpdaterAggregatorGraph();
        ComputationGraphUpdater.Aggregator uaGraph2 = v2.getUpdaterAggregatorGraph();
        ComputationGraphUpdater.Aggregator uaGraphCombined;
        if(uAGraph1 == null) uaGraphCombined = uaGraph2;
        else if(uaGraph2 == null) uaGraphCombined = uAGraph1;
        else {
            uAGraph1.merge(uaGraph2);
            uaGraphCombined = uAGraph1;
        }

        double scoreSum = v1.getScoreSum() + v2.getScoreSum();
        int aggregationCount = v1.getAggregationsCount() + v2.getAggregationsCount();

        SparkTrainingStats stats = v1.getSparkTrainingStats();
        if(v2.getSparkTrainingStats() != null){
            if(stats == null) stats = v2.getSparkTrainingStats();
            else stats.addOtherTrainingStats(v2.getSparkTrainingStats());
        }

        return new ParameterAveragingAggregationTuple(newParams, combinedAggregator, uaGraphCombined, scoreSum, aggregationCount, stats);
    }
}