public GbmTree Train(double[] gradients, double[] hessians)
{
_frame.RandomizeTrainingIndices(_algorithmSettings.RowSamplingRate);
var sampledFeatures = _frame.GetRandomInputList(_algorithmSettings.ColumnSamplingRate);
var sampledTrainingIndices = _frame.GetRandomTrainingIndices();
var rootBinInfo = new BinInfo();
foreach (var idx in sampledTrainingIndices)
{
rootBinInfo.Add(1, gradients[idx], hessians[idx]);
}
var tree = GbmTree.CreateEmptyTree(RootNodeIdx);
var splitComputers = new Dictionary <string, SplitComputer>();
foreach (var feature in sampledFeatures)
{
var bin = _frame.GetBin(feature);
splitComputers.Add(feature, new SplitComputer(feature, bin.NumberOfBins));
}
while (true)
{
foreach (var feature in sampledFeatures)
{
}
}
throw new NotImplementedException();
}
private double GetLeafSplitGain(BinInfo info)
{
var sumGradients = info.SumGradients;
var sumHessians = info.SumHessians;
var absSumGradients = Math.Abs(sumGradients);
return((absSumGradients * absSumGradients) / (sumHessians));
}
public BinInfoViewModel(BinInfo bi, INavigation navigation) : base(navigation)
{
Caption = bi.Caption;
Description = bi.Description;
Detail = bi.Detail;
Value1 = bi.Value1;
Value2 = bi.Value2;
value3 = bi.Value3;
ImageURL = bi.ImageURL;
}
public SplitInfo()
{
Feature = String.Empty;
IsNumerical = false;
IsNull = true;
IntegerThreshold = -1;
LeftInfoTraining = new BinInfo();
RightInfoTraining = new BinInfo();
LeftCategoryIndices = null;
LeftPrediction = Double.NaN;
RightPrediction = Double.NaN;
Gain = Double.MinValue;
DefaultOnLeft = true;
DoesDefaultExist = false;
}
private void FindBestCategoricalSplit(BinInfo parent, double parentInformation, BinInfo[] bins, SplitInfo split,
bool doesDefaultExist, GbmAlgorithmSettings algorithmSettings)
{
throw new NotImplementedException();
}
private void ScanFromLeft(BinInfo parentT, double parentTGain, BinInfo[] binsT, SplitInfo split, bool doesDefaultExist, GbmAlgorithmSettings algorithmSettings)
{
throw new NotImplementedException();
}
