public VirtualTable(PhysicalTable _source)
{
int i;
int rc = _source.RowCount;
source = _source;
row = new DynArrInt(rc);
for(i=0; i < rc; ++i)
row.Add(i);
}
public VirtualTable(DynArrInt subRows)
{
row = subRows;
}
public VirtualTable SubSetUpTo(int col, double threshold)
{
DynArrInt rowsSelected = new DynArrInt();
for(int y=0; y < RowCount; ++y)
{
if(GetNumber(col, y) <= threshold)
try
{
rowsSelected.Add(row[y]);
}
catch(Exception e)
{
MessageBox.Show(e.Message);
}
}
VirtualTable subVT = new VirtualTable(rowsSelected);
return subVT;
}
public VirtualTable SubSet(int col, string val)
{
DynArrInt rowsSelected = new DynArrInt();
for(int y=0; y < RowCount; ++y)
{
if(this[col, y].ToString().CompareTo(val) == 0)
rowsSelected.Add(row[y]);
}
VirtualTable subVT = new VirtualTable(rowsSelected);
return subVT;
}
public static double InfoContinuousAttribute(Node nd, Attribute att, ref double thresholdMinInfo, ref bool splitValid)
{
Case classCs;
double infoActual, infoActualUptoT, infoActualUptoTp2, infoActualHigherT, infoActualHigherTp2;
double infoAttributeMin;
int attributeMatchClass=0;
int y;
int DependentVariableAttributeCaseLCount = nd.DependentVariableAttribute.CaseL.Count;
bool infoAttributeIsSet = false;
infoAttributeMin=0;
DynArrInt threshold = new DynArrInt();
SetThresholdIndexes(threshold, nd, att.VariableIndex);
if(threshold.Count==0)
{
splitValid=false;
return -1;
}
for(int ti=0; ti < threshold.Count; ++ti)
{
infoActual=0;
infoActualUptoT = (double) (threshold[ti] + 1) / nd.VT.RowCount ;
infoActualUptoTp2 = 0;
attributeMatchClass = 0;
infoActualHigherT = (double) (nd.VT.RowCount - (threshold[ti] + 1)) / nd.VT.RowCount ;
infoActualHigherTp2 = 0;
attributeMatchClass = 0;
for (int classIndex = 0; classIndex < DependentVariableAttributeCaseLCount; ++classIndex)
{
//Calculation of the info up to threshold second part
attributeMatchClass = 0;
classCs = (Case) nd.DependentVariableAttribute.CaseL[classIndex];
for(y = 0; y <= threshold[ti]; ++y)
{
if(classCs.Val.ToString().CompareTo(nd.VT[Variable.DependentVariableIndex, y].ToString()) == 0)
++attributeMatchClass;
}
if(attributeMatchClass > 0)
infoActualUptoTp2 += -1 * Math.Log( (double) attributeMatchClass / (double) (threshold[ti] + 1), 2);
//Calculation of the info higher threshold second part
attributeMatchClass = 0;
for(y = threshold[ti] + 1; y < nd.VT.RowCount; ++y)
{
if(classCs.Val.ToString().CompareTo(nd.VT[Variable.DependentVariableIndex, y].ToString()) == 0)
++attributeMatchClass;
}
if(attributeMatchClass > 0)
infoActualHigherTp2 += -1 * Math.Log( (double) attributeMatchClass / (double) (nd.VT.RowCount - (threshold[ti] + 1)), 2);
}
infoActualUptoT = infoActualUptoT * infoActualUptoTp2;
infoActualHigherT = infoActualHigherT * infoActualHigherTp2;
infoActual = infoActualUptoT + infoActualHigherT;
if(infoAttributeIsSet == false)
{
infoAttributeMin = infoActual;
infoAttributeIsSet = true;
//This is the real threshold as a value
thresholdMinInfo = nd.VT.GetNumber(att.VariableIndex, threshold[ti]);
}
else
if(infoActual < infoAttributeMin)
{
infoAttributeMin = infoActual;
//This is the real threshold as a value
thresholdMinInfo = nd.VT.GetNumber(att.VariableIndex, threshold[ti]);
}
}
return infoAttributeMin;
}
private static void SetThresholdIndexes(DynArrInt threshold, Node nd, int attIndex)
{
int uniqueLastIdx=0;
double uniqueLast = nd.VT.GetNumber(attIndex, uniqueLastIdx);
int uniqueIdx, thIdx, i, rc = nd.VT.RowCount;
double unique, th, valMin;
//Ends when there are not enough rows at the end
for(uniqueIdx = Option.C45MinNumOfCasesPerContinuosSplit -1; uniqueIdx + Option.C45MinNumOfCasesPerContinuosSplit < rc; ++uniqueIdx)
{
//outdated can be optimised
if(nd.VT.GetNumber(attIndex, uniqueIdx) != nd.VT.GetNumber(attIndex, uniqueIdx+1))
threshold.Add(uniqueIdx);
}
}