public void InitNewModel()
{
if (ExpData == null)
{
throw new Exception("Projects is not initialized in the model.");
}
string par = "";
//first time setting parameter
if (ExpData.GetOutputColumnType() == ColumnType.Binary)
{
par = initBinaryClassParameters();
}
else if (ExpData.GetOutputColumnType() == ColumnType.Category)
{
par = initMultiClassParameters();
}
else
{
par = initRegressionParameters();
}
//
var p = Parameters.FromString(par);
//
var fs = new Function[4] {
new Function()
{
Selected = true, Id = 2000, Weight = 1
},
new Function()
{
Selected = true, Id = 2001, Weight = 1
},
new Function()
{
Selected = true, Id = 2002, Weight = 1
},
new Function()
{
Selected = true, Id = 2003, Weight = 1
},
};
var t = terminalSet(p).ToArray();
//
Factory = new Factory();
Factory.FunctionSet = fs;
Factory.TerminalSet = t;
Factory.Parameters = p;
}
public double[] calculateOutput(bool isTraining, bool probValue = false)
{
if (Factory.ProgresReport.BestSolution == null)
{
return(null);
}
//
var ch = Factory.ProgresReport.BestSolution;//.expressionTree;
var par = Factory.Parameters;
if (par.RootFunctionNode == null)
{
setLearningType(par);
}
double[] yc = null;
if (ch != null)
{
yc = Inputs.CalculateOutput(ch, par, isTraining);//gp
//no testing data set
if (yc == null)
{
return(null);
}
//de normalize output
double[] output = yc;
for (int i = 0; i < yc.Length; i++)
{
//calculate de normalization
if (ExpData.GetOutputColumnType() == ColumnType.Numeric)
{
double[] normRow = new double[1] {
yc[i]
};
output[i] = ExpData.GetDecodedOutputRow(normRow)[0];
}
else
{
if (par.RootFunctionNode.Id == 2048) //sigmoid
{
if (probValue) //when model is evaluate we need probability f event in order to optimize threshold value
{
output[i] = yc[i];
}
else
{
output[i] = yc[i] > par.RootFunctionNode.Parameter ? 1 : 0;
}
}
else if (par.RootFunctionNode.Id == 2050)//step
{
output[i] = Math.Truncate(yc[i]);
}
else
{
output[i] = Math.Truncate(yc[i]);
}
}
}
return(output);
}
return(null);//no solution yet
}
public bool setLearningType(Parameters param)
{
//determine the type of ML
param.OutputType = ExpData.GetOutputColumnType();
if (param.OutputType == ColumnType.Binary)
{
if (param.RootName.StartsWith("Sigm"))
{
param.RootFunctionNode = new Function()
{
Id = 2048, Name = "Sigmoid", Arity = 1, HasParameter = true, Parameter = param.Threshold, Parameter2 = 2
}
}
;
else if (param.RootName.StartsWith("Step"))
{
param.RootFunctionNode = new Function()
{
Id = 2049, Name = "Step", Arity = 1, HasParameter = true, Parameter = param.Threshold, Parameter2 = 2
}
}
;
else if (param.RootName.StartsWith("Scal"))
{
param.RootFunctionNode = new Function()
{
Id = 2050, Name = "SSigmoid", Arity = 1, HasParameter = true, Parameter = param.Threshold, Parameter2 = 2
}
}
;
else if (param.RootName.StartsWith("Soft"))
{
param.RootFunctionNode = new Function()
{
Id = 2051, Name = "Softmax", Arity = 2, Parameter2 = 2
}
}
;
}
else if (param.OutputType == ColumnType.Category)
{
var clss = ExpData.GetColumnsFromOutput().FirstOrDefault().Statistics.Categories.Count;
if (param.RootName.StartsWith("Scal"))
{
param.RootFunctionNode = new Function()
{
Id = 2050, Name = "SSigmoid", Arity = 1, HasParameter = true, Parameter = clss, Parameter2 = clss
}
}
;
else if (param.RootName.StartsWith("Soft"))
{
param.RootFunctionNode = new Function()
{
Id = 2051, Name = "Softmax", Arity = clss, Parameter2 = clss
}
}
;
else
{
throw new Exception("Predefined Root Node is not compatible with multi class classification modeling!");
}
}
else if (param.OutputType == ColumnType.Numeric)
{
if (param.RootName.StartsWith("Pol3"))
{
param.RootFunctionNode = Globals.GetFunction(2039);// new Function() { Id = 2050, Name = "P3", Arity = 1, HasParameter = true, Parameter = clss, Parameter2 = clss };
}
}
return(true);
}