void ComputeNearestNeighbors()
{
lock (_lock)
{
if (_trees != null)
{
return;
}
using (var ch = _host.Start("Build k-d tree"))
{
ch.Info(MessageSensitivity.None, "ComputeNearestNeighbors: build a k-d tree.");
int featureIndex, labelIndex, idIndex, weightIndex;
featureIndex = GetColumnIndex(ch, _args.column);
if (featureIndex == -1)
{
throw ch.Except($"Unable to find column '{_args.column}' in {SchemaHelper.ToString(Schema)}.");
}
labelIndex = GetColumnIndex(ch, _args.labelColumn);
weightIndex = GetColumnIndex(ch, _args.weightColumn);
idIndex = GetColumnIndex(ch, _args.colId);
Dictionary <long, Tuple <long, float> > merged;
_trees = NearestNeighborsBuilder.NearestNeighborsBuild <long>(ch, _input, featureIndex, labelIndex,
idIndex, weightIndex, out merged, _args);
ch.Info(MessageSensitivity.UserData, "Done. Tree size: {0} points.", _trees.Count());
}
}
}
private INearestNeighborsPredictor TrainPredictorLabel <TLabel>(IChannel ch, RoleMappedData data)
where TLabel : IComparable <TLabel>
{
int featureIndex = data.Schema.Feature.Index;
int labelIndex = data.Schema.Label.Index;
int idIndex = -1;
int weightIndex = data.Schema.Weight == null ? -1 : data.Schema.Weight.Index;
var indexes = new HashSet <int>()
{
featureIndex, labelIndex, weightIndex
};
if (!string.IsNullOrEmpty(_args.colId) && data.Schema.Schema.TryGetColumnIndex(_args.colId, out idIndex))
{
indexes.Add(idIndex);
}
if (idIndex != -1)
{
var colType = data.Schema.Schema.GetColumnType(idIndex);
if (colType.IsVector() || colType.RawKind() != DataKind.I8)
{
throw ch.Except("Column '{0}' must be of type '{1}' not '{2}'", _args.colId, DataKind.I8, colType);
}
}
Dictionary <long, Tuple <TLabel, float> > merged;
var kdtrees = NearestNeighborsBuilder.NearestNeighborsBuild <TLabel>(ch, data.Data, featureIndex, labelIndex,
idIndex, weightIndex, out merged, _args);
// End.
return(CreateTrainedPredictor(kdtrees.Trees, merged));
}
void ComputeNearestNeighbors()
{
lock (_lock)
{
if (_trees != null)
{
return;
}
using (var ch = _host.Start("Build k-d tree"))
{
ch.Info("ComputeNearestNeighbors: build a k-d tree.");
int featureIndex, labelIndex, idIndex, weightIndex;
if (!_input.Schema.TryGetColumnIndex(_args.column, out featureIndex))
{
throw ch.Except("Unable to find column '{0}'.", _args.column);
}
labelIndex = GetColumnIndex(ch, _args.labelColumn);
weightIndex = GetColumnIndex(ch, _args.weightColumn);
idIndex = GetColumnIndex(ch, _args.colId);
Dictionary <long, Tuple <long, float> > merged;
_trees = NearestNeighborsBuilder.NearestNeighborsBuild <long>(ch, _input, featureIndex, labelIndex,
idIndex, weightIndex, out merged, _args);
ch.Info("Done. Tree size: {0} points.", _trees.Count());
}
}
}