public CLabelFeatureDataSubset(CLabelFeatureData labelFeatureData, Subset subset)
: base(labelFeatureData)
{
this.cSubSet = 0;
this.mapTbl = new int[labelFeatureData.NumDataPoint];
for (int i = 0; i < labelFeatureData.NumDataPoint; i++)
{
if (subset.Keep(i))
{
this.mapTbl[this.cSubSet++] = i;
}
}
}
static void Main(string[] args)
{
DataProcArgs cmd = new DataProcArgs(args);
IGroupBoundary boundary = null;
if (cmd.queryBoundary)
{
//we need to keep tract of the queries for ranking
boundary = new QueryBoundary();
}
else
{
//data boundary: no boundary
boundary = new OnelineGroup();
}
string[] labelName = { cmd.labelName };
IParser<float> RateParser = new MsnLabelParser(labelName, cmd.labelNameValueFile);
Console.WriteLine("Loading data from tsv file " + cmd.tsvFile);
MsnFeatureParser featureParser = null;
//read and process only a subset of activated features as specified in the activeFeatureFile
if (cmd.activeFeatureFile != null)
{
string[] FeatureNames = TsvFileLoader.ReadFeatureNames(cmd.activeFeatureFile);
featureParser = new MsnFeatureParser(FeatureNames);
}
TsvFileLoader tsvFileLoader = new TsvFileLoader(cmd.tsvFile, null, RateParser, featureParser, boundary);
Console.WriteLine("Finishing loading the tsv file");
Console.WriteLine("Create LabelFeatureData uncoded ...");
CLabelFeatureData labelFeatureData = new CLabelFeatureData(tsvFileLoader.FeatureName, tsvFileLoader.Labels, tsvFileLoader.GroupId, tsvFileLoader.Feature);
Console.WriteLine("Save LabelFeatureData uncoded ...");
if (cmd.binFile != null)
{
labelFeatureData.Save(cmd.binFile);
}
Console.WriteLine("Create LabelFeatureData coded ...");
CLabelFeatureDataCoded labelFeatureDataCoded = new CLabelFeatureDataCoded(labelFeatureData, cmd.cThreads, cmd.storeCodedFeature, cmd.fCodedFeatureSparse);
Console.WriteLine("Save LabelFeatureData coded ...");
if (cmd.binFileCoded != null)
{
labelFeatureDataCoded.Save(cmd.binFileCoded);
}
}
public CLabelFeatureDataCoded(CLabelFeatureData labelFeatureData, int cThreads)
: this(labelFeatureData, cThreads, true)
{
}
public CLabelFeatureDataCoded(CLabelFeatureData labelFeatureData, int cThreads, bool fStoreCodedFeature)
: this(labelFeatureData, cThreads, fStoreCodedFeature, true)
{
}
public CLabelFeatureDataCoded(CLabelFeatureData labelFeatureData, int cThreads, bool fStoreCodedFeature, bool fSparse)
: base(labelFeatureData)
{
//input data
int numRows = labelFeatureData.NumDataPoint;
int numCols = labelFeatureData.NumFeatures;
//compute the code book for each feature
this.codeBook = ComputeCodeBook(this.feature, cThreads);
//encode the original feature values and store them for speed if required
if (fStoreCodedFeature)
{
this.featureCoded = EncodeFeatureValues(this.feature, this.codeBook, cThreads, fSparse);
}
}
public CLabelFeatureData(CLabelFeatureData labelFeatureData)
: base(labelFeatureData)
{
this.feature = labelFeatureData.feature;
}
/// <summary>
/// Load LabelFeatureData file and distribute - static function for both CLabelFeatureData and CLabelFeatureDataCoded
/// load in the file and output a LabelFeatureData object that has the specified Type (type = {LabelFeatureData, LabelFeatureDataCode}
/// </summary>
/// <param name="inFileName">the name of the file: xxx.tsv == tsv file formation; xxx.bin == binary uncoded data format; xxx.dp == binary coded data format</param>
/// <param name="featureParser">parser that understand the feature values</param>
/// <param name="labelParser">parser that understand the label values</param>
/// <param name="dataGroupBoundary">data group boundaries</param>
/// <param name="outDatatype">the output data type LabelFeatureData or LabelFeatureDataCoded</param>
/// <param name="activeFeatureNames">only these feature values are loaded</param>
/// <param name="cThreads">number of threads used to code the original data</param>
/// <returns>the desired LabelFeatureData if no errors in loading; otherwise, null</returns>
static public LabelFeatureData DistributeLoad(string inFileName, IParser<float> featureParser, IParser<float> labelParser, IGroupBoundary dataGroupBoundary,
Type outDataType, string[] activeFeatureNames, int cThreads, bool fCacheCodedFeature, bool fSparseCoded, Random r, double rangeLower, double rangeUpper)
{
if (inFileName == null)
{
return null;
}
string[] fields = inFileName.Split('.');
if (fields.Length <= 0)
{
return null;
}
CLabelFeatureData labelFeatureData = null;
string sufix = fields[fields.Length - 1];
if (string.Compare(sufix, "tsv", true) == 0 || string.Compare(sufix, "gz", true) == 0)
{
TsvFileLoader tsvFileLoader = new TsvFileLoader(inFileName, null, labelParser, featureParser, dataGroupBoundary, r, rangeLower, rangeUpper);
labelFeatureData = new CLabelFeatureData(tsvFileLoader.FeatureName, tsvFileLoader.Labels, tsvFileLoader.GroupId, tsvFileLoader.Feature);
}
else if (string.Compare(sufix, "bin", true) == 0 || string.Compare(sufix, "dp", true) == 0)
{
//initially, only accept tsv file.
return null;
//BinaryReaderEx binReaderEx = new BinaryReaderEx(inFileName);
//Type t = binReaderEx.Read<Type>();
//labelFeatureData = (CLabelFeatureData)binReaderEx.Read(t);
//binReaderEx.Close();
//labelFeatureData.SetActiveFeatures(activeFeatureNames);
}
if (outDataType.Equals(typeof(CLabelFeatureDataCoded)))
{
if (labelFeatureData.GetType().Equals(typeof(CLabelFeatureDataCoded)))
{
if (fCacheCodedFeature)
{
((CLabelFeatureDataCoded)labelFeatureData).EncodeFeatureValues(cThreads, fSparseCoded);
}
}
else
{
//need to upgrade to coded
labelFeatureData = new CLabelFeatureDataCoded(labelFeatureData, cThreads, fCacheCodedFeature, fSparseCoded);
}
}
return labelFeatureData;
}
override public bool Evaluate(float[] features, float[] results)
{
float[][] testData = new float[1][];
testData[0] = features;
LabelFeatureData labelFeatureData = new CLabelFeatureData(testData);
float[][] prob = boostTree.Predict(labelFeatureData);
for (int i = 0; i < results.Length; i++)
{
results[i] = prob[i][0];
}
return true;
}