protected virtual Matrix <double> ExtractQueryDataAsMatrix(
IDataFrame queryDataFrame,
IKnnPredictionModel <TPredictionResult> knnModel,
int dependentFeatureIdx)
{
var dependentFetureName = (dependentFeatureIdx < queryDataFrame.ColumnsCount && dependentFeatureIdx >= 0)
? queryDataFrame.ColumnNames[dependentFeatureIdx]
: string.Empty;
var queryMatrix = queryDataFrame.GetSubsetByColumns(
queryDataFrame.ColumnNames.Where(colName => colName != dependentFetureName).ToList()).GetAsMatrix();
return(queryMatrix);
}
protected virtual Tuple <Matrix <double>, Matrix <double> > NormalizeData(IDataFrame queryDataFrame, IKnnPredictionModel <TPredictionResult> knnModel, int dependentFeatureIdx)
{
var modelMatrix = knnModel.TrainingData;
var queryMatrix = ExtractQueryDataAsMatrix(queryDataFrame, knnModel, dependentFeatureIdx);
return(PerformNormalization(modelMatrix, queryMatrix));
}
protected override Tuple <Matrix <double>, Matrix <double> > NormalizeData(IDataFrame queryDataFrame, IKnnPredictionModel <TPredictionResult> knnModel, int dependentFeatureIdx)
{
var backwardsEliminationModel = knnModel as IBackwardsEliminationKnnModel <TPredictionResult>;
var featureIndicesToRemove = backwardsEliminationModel.RemovedFeaturesData.Select(f => queryDataFrame.ColumnNames.IndexOf(f.FeatureName)).OrderBy(i => i).ToList();
var relevantFeatures =
queryDataFrame.ColumnNames.Where((colName, colIdx) => !featureIndicesToRemove.Contains(colIdx) && colIdx != dependentFeatureIdx).ToList();
var modelMatrix = GetModelMatrixWithOnlyRelevantColumns(knnModel.TrainingData, featureIndicesToRemove);
var queryDataFrameWithoutRedundantColumns = queryDataFrame.GetSubsetByColumns(relevantFeatures).GetAsMatrix();
return(base.PerformNormalization(modelMatrix, queryDataFrameWithoutRedundantColumns));
}
