private void ReadXmlSupportVectorFeatures(XmlReader rdr, out FeatureList supportFeatures)
{
var featurePoints = new List <double>();
rdr.MoveToContent();
//start walking down the tree
if (rdr.NodeType == XmlNodeType.Element && rdr.Name == XmlXsupTag)
{
//at elements
rdr.ReadStartElement(); // Read the xsup tag, to get to the contents
//first bias
while (rdr.NodeType != XmlNodeType.EndElement && !rdr.EOF)
{
switch (rdr.Name)
{
case XmlFeatureTag:
featurePoints.Add(rdr.ReadElementContentAsDouble());
break;
default:
rdr.Skip();
break;
}
}
if (!rdr.EOF)
{
rdr.ReadEndElement();
}
}
rdr.Close();
supportFeatures = new FeatureList(featurePoints);
}
private void ReadValues(int scanNum, double parentMz, double xScore2, double xScore3, double xScoreRatio,
double bScore2, double bScore3, double pk1Prob, double pk2Prob, double pk3Prob,
double pk4Prob, double fscore2, double fscore3, double xScore2_CO, double xScore3_CO, double xScore2_H2O,
double xScore3_H20, double xScore2_NH3, double xScore3_NH3)
{
var features = new List <double>
{
scanNum,
parentMz,
xScore2,
xScore3,
xScoreRatio,
bScore2,
bScore3,
pk1Prob,
pk2Prob,
pk3Prob,
pk4Prob,
fscore2,
fscore3,
xScore2_CO,
xScore3_CO,
xScore2_H2O,
xScore3_H20,
xScore2_NH3,
xScore3_NH3
};
var scanFeatures = new FeatureList(features);
_testVector.Add(scanFeatures);
for (var i = 0; i < NumFeatures; i++)
{
var val = features[i];
if (val <= _minValues[i])
{
_minValues[i] = val;
}
if (val >= _maxValues[i])
{
_maxValues[i] = val;
}
}
}
public void SVMClassification(int startIndex, int stopIndex, int k_bias)
{
var chunk_xtest = new List <FeatureList>();
var chunk_xsup = new List <FeatureList>();
var numSupport = stopIndex - startIndex;
var numTest = _testVector.Count;
const int chunksize = 100;
var ind1 = new List <int>();
var ind2 = new List <int>();
var chunks1 = (numSupport / chunksize) + 1;
var chunks2 = (numTest / chunksize) + 1;
// Performing y2(ind2)=y2(ind2)+ kchunk*w(ind1);
for (var ch1 = 1; ch1 <= chunks1; ch1++)
{
//Get ind1
var low_ind1_index = (ch1 - 1) * chunksize + startIndex;
var high_ind1_index = (ch1 * chunksize) - 1 + startIndex;
if (high_ind1_index > stopIndex)
{
high_ind1_index = stopIndex;
}
ind1.Clear();
for (var index = 0; index <= (high_ind1_index - low_ind1_index); index++)
{
ind1.Add(index + low_ind1_index);
}
//Get support vectors
chunk_xsup.Clear();
for (var j = 0; j < ind1.Count; j++)
{
var xsupIndex = ind1[j];
var thisSupportVector = new FeatureList(_supportVectors[xsupIndex]);
chunk_xsup.Add(thisSupportVector);
}
for (var ch2 = 1; ch2 <= chunks2; ch2++)
{
//Get ind2
var low_ind2_index = (ch2 - 1) * chunksize;
var high_ind2_index = (ch2 * chunksize) - 1;
if (high_ind2_index > numTest)
{
high_ind2_index = numTest - 1;
}
ind2.Clear();
for (var index2 = 0; index2 <= (high_ind2_index - low_ind2_index); index2++)
{
ind2.Add(index2 + low_ind2_index);
}
//Get X vector
chunk_xtest.Clear();
for (var j = 0; j < ind2.Count; j++)
{
var xIndex = ind2[j];
var thisTestVector = _testVector[xIndex];
chunk_xtest.Add(new FeatureList(thisTestVector));
}
//Get the kernel
var svmKernel = GetKernel(chunk_xtest, chunk_xsup);
//Read in the weights w(ind1)
var w = Matrix <double> .Build.Dense(ind1.Count, 1, 0);
for (var i = 0; i < ind1.Count; i++)
{
var index = ind1[i];
w[i, 0] = _weights[index];
}
// m1 = kchunk*w(ind1)
var m1 = svmKernel.Multiply(w);
//y2(ind2) += m1;
for (var i = 0; i < ind2.Count; i++)
{
var index = ind2[i];
_predictedY[index] += m1[i, 0];
}
}
}
//Add w0
for (var i = 0; i < _predictedY.Count; i++)
{
_predictedY[i] += _biases[k_bias];
}
}
public FeatureList(FeatureList a)
{
Array.Copy(a._features, _features, NumFeatures);
}
