internal override InternalRegressionTree TrainingIteration(IChannel ch, bool[] activeFeatures)
{
Contracts.CheckValue(ch, nameof(ch));
// Fit a regression tree to the gradient using least squares.
InternalRegressionTree tree = TreeLearner.FitTargets(ch, activeFeatures, AdjustTargetsAndSetWeights(ch));
if (tree == null)
{
return(null); // Could not learn a tree. Exit.
}
// Adjust output values of tree by performing a Newton step.
// REVIEW: This should be part of OptimizingAlgorithm.
using (Timer.Time(TimerEvent.TreeLearnerAdjustTreeOutputs))
{
double[] backupScores = null;
// when doing dropouts we need to replace the TrainingScores with the scores without the dropped trees
if (DropoutRate > 0)
{
backupScores = TrainingScores.Scores;
TrainingScores.Scores = _scores;
}
if (AdjustTreeOutputsOverride != null)
{
AdjustTreeOutputsOverride.AdjustTreeOutputs(ch, tree, TreeLearner.Partitioning, TrainingScores);
}
else if (ObjectiveFunction is IStepSearch)
{
(ObjectiveFunction as IStepSearch).AdjustTreeOutputs(ch, tree, TreeLearner.Partitioning, TrainingScores);
}
else
{
throw ch.Except("No AdjustTreeOutputs defined. Objective function should define IStepSearch or AdjustTreeOutputsOverride should be set");
}
if (DropoutRate > 0)
{
// Returning the original scores.
TrainingScores.Scores = backupScores;
}
}
if (Smoothing != 0.0)
{
SmoothTree(tree, Smoothing);
UseFastTrainingScoresUpdate = false;
}
if (DropoutRate > 0)
{
// Don't do shrinkage if you do dropouts.
double scaling = (1.0 / (1.0 + _numberOfDroppedTrees));
tree.ScaleOutputsBy(scaling);
_treeScores.Add(tree.GetOutputs(TrainingScores.Dataset));
}
UpdateAllScores(ch, tree);
Ensemble.AddTree(tree);
return(tree);
}
// Divides output values of leaves to bag count.
// This brings back the final scores generated by model on a same
// range as when we didn't use bagging
internal void ScaleEnsembleLeaves(int numTrees, int bagSize, InternalTreeEnsemble ensemble)
{
int bagCount = GetBagCount(numTrees, bagSize);
for (int t = 0; t < ensemble.NumTrees; t++)
{
InternalRegressionTree tree = ensemble.GetTreeAt(t);
tree.ScaleOutputsBy(1.0 / bagCount);
}
}
void IStepSearch.AdjustTreeOutputs(IChannel ch, InternalRegressionTree tree, DocumentPartitioning partitioning,
ScoreTracker previousScores)
{
_lo.Initialize(tree, partitioning, previousScores);
_hi.Initialize(tree, partitioning, previousScores);
_left.Initialize(tree, partitioning, previousScores);
_right.Initialize(tree, partitioning, previousScores);
_lo.Step = _historicStepSize / _phi;
_left.Step = _historicStepSize;
if (_lo.Loss.CompareTo(_left.Loss) == 1) // backtrack
{
do
{
Rotate(ref _hi, ref _left, ref _lo);
if (_hi.Step <= _minStepSize)
{
goto FINISHED;
}
_lo.Step = _left.Step / _phi;
} while (_lo.Loss.CompareTo(_left.Loss) == 1);
}
else // extend (or stay)
{
_hi.Step = _historicStepSize * _phi;
while (_hi.Loss.CompareTo(_left.Loss) == 1)
{
Rotate(ref _lo, ref _left, ref _hi);
_hi.Step = _left.Step * _phi;
}
}
if (_numPostbracketSteps > 0)
{
_right.Step = _lo.Step + (_hi.Step - _lo.Step) / _phi;
for (int step = 0; step < _numPostbracketSteps; ++step)
{
int cmp = _right.Loss.CompareTo(_left.Loss);
if (cmp == 0)
{
break;
}
if (cmp == 1) // move right
{
Rotate(ref _lo, ref _left, ref _right);
_right.Step = _lo.Step + (_hi.Step - _lo.Step) / _phi;
}
else // move left
{
Rotate(ref _hi, ref _right, ref _left);
if (_hi.Step <= _minStepSize)
{
goto FINISHED;
}
_left.Step = _hi.Step - (_hi.Step - _lo.Step) / _phi;
}
}
// prepare to return _left
if (_right.Loss.CompareTo(_left.Loss) == 1)
{
Swap(ref _left, ref _right);
}
}
FINISHED:
if (_hi.Step < _minStepSize)
{
_left.Step = _minStepSize;
}
else if (_hi.Step == _minStepSize)
{
Swap(ref _hi, ref _left);
}
double bestStep = _left.Step;
ch.Info("multiplier: {0}", bestStep);
_historicStepSize = bestStep;
tree.ScaleOutputsBy(bestStep);
}
