protected override OptimizationAlgorithm ConstructOptimizationAlgorithm(IChannel ch)
{
Contracts.CheckValue(ch, nameof(ch));
OptimizationAlgorithm optimizationAlgorithm;
IGradientAdjuster gradientWrapper = MakeGradientWrapper(ch);
switch (Args.OptimizationAlgorithm)
{
case BoostedTreeArgs.OptimizationAlgorithmType.GradientDescent:
optimizationAlgorithm = new GradientDescent(Ensemble, TrainSet, InitTrainScores, gradientWrapper);
break;
case BoostedTreeArgs.OptimizationAlgorithmType.AcceleratedGradientDescent:
optimizationAlgorithm = new AcceleratedGradientDescent(Ensemble, TrainSet, InitTrainScores, gradientWrapper);
break;
case BoostedTreeArgs.OptimizationAlgorithmType.ConjugateGradientDescent:
optimizationAlgorithm = new ConjugateGradientDescent(Ensemble, TrainSet, InitTrainScores, gradientWrapper);
break;
default:
throw ch.Except("Unknown optimization algorithm '{0}'", Args.OptimizationAlgorithm);
}
optimizationAlgorithm.TreeLearner = ConstructTreeLearner(ch);
optimizationAlgorithm.ObjectiveFunction = ConstructObjFunc(ch);
optimizationAlgorithm.Smoothing = Args.Smoothing;
optimizationAlgorithm.DropoutRate = Args.DropoutRate;
optimizationAlgorithm.DropoutRng = new Random(Args.RngSeed);
optimizationAlgorithm.PreScoreUpdateEvent += PrintTestGraph;
return(optimizationAlgorithm);
}
protected override OptimizationAlgorithm ConstructOptimizationAlgorithm(IChannel ch)
{
Host.CheckValue(ch, nameof(ch));
IGradientAdjuster gradientWrapper = MakeGradientWrapper(ch);
var optimizationAlgorithm = new RandomForestOptimizer(Ensemble, TrainSet, InitTrainScores, gradientWrapper);
optimizationAlgorithm.TreeLearner = ConstructTreeLearner(ch);
optimizationAlgorithm.ObjectiveFunction = ConstructObjFunc(ch);
optimizationAlgorithm.Smoothing = Args.Smoothing;
// No notion of dropout for non-boosting applications.
optimizationAlgorithm.DropoutRate = 0;
optimizationAlgorithm.DropoutRng = null;
optimizationAlgorithm.PreScoreUpdateEvent += PrintTestGraph;
return(optimizationAlgorithm);
}
public AcceleratedGradientDescent(TreeEnsemble ensemble, Dataset trainData, double[] initTrainScores, IGradientAdjuster gradientWrapper)
: base(ensemble, trainData, initTrainScores, gradientWrapper)
{
UseFastTrainingScoresUpdate = false;
}
// REVIEW: When the FastTree appliation is decoupled with tree learner and boosting logic, this class should be removed.
public RandomForestOptimizer(TreeEnsemble ensemble, Dataset trainData, double[] initTrainScores, IGradientAdjuster gradientWrapper)
: base(ensemble, trainData, initTrainScores, gradientWrapper)
{
_gradientWrapper = gradientWrapper;
}
internal GradientDescent(InternalTreeEnsemble ensemble, Dataset trainData, double[] initTrainScores, IGradientAdjuster gradientWrapper)
: base(ensemble, trainData, initTrainScores)
{
_gradientWrapper = gradientWrapper;
_treeScores = new List <double[]>();
}
public ConjugateGradientDescent(InternalTreeEnsemble ensemble, Dataset trainData, double[] initTrainScores, IGradientAdjuster gradientWrapper)
: base(ensemble, trainData, initTrainScores, gradientWrapper)
{
_currentDk = new double[trainData.NumDocs];
}
