public static void InitializeGenerationalDataSet(int timestepRange)
{
totalRun = 0;
MeanFitnessOfGeneration = 0.0;
BestFitnessOfGenerationGainPercent = 0.0;
BestFitnessOfGeneration = 0.0;
BestFitnessOfGenerationNumberTrades = 0;
WorstFitnessOfGeneration = 1.0;
//if(CURRENT_DATA_SET == null)
//{
CURRENT_DATA_SET = TradingData.fetchRandomSeries(timestepRange);
//}
}
public static TradingData fetchRandomSeries(int timestepRange)
{
TradingData series;
do
{
Random rnd = new Random();
int startIndex = rnd.Next(0, allData.Length - timestepRange + 1);
TimeStepDataPiece[] randomSet = new TimeStepDataPiece[timestepRange];
for (int i = 0; i < timestepRange; ++i)
{
randomSet[i] = allData[startIndex + i];
}
series = new TradingData(randomSet);
} while (series.MaximumGainFactor < 1.3 || series.MaximumLossFactor < 1.6);
return(series);
}
private TrainingSession()
{
// Initialise log4net (log to console).
XmlConfigurator.Configure(new FileInfo("log4net.properties"));
// Experiment classes encapsulate much of the nuts and bolts of setting up a NEAT search.
this.experiment = new TradingExperiment();
// Initialize the data to be used.
TradingData.initializeData();
// Load config XML.
XmlDocument xmlConfig = new XmlDocument();
xmlConfig.Load(StorageLayer.FILE_IO_PATH + "trading.config.xml");
this.experiment.Initialize("Trading", xmlConfig.DocumentElement);
}
/// <summary>
/// Evaluate the provided IBlackBox.
/// </summary>
public FitnessInfo Evaluate(IBlackBox box)
{
//Console.WriteLine("Evaluate being called");
double startingOutBalance = 1000.00;
double outBalance = startingOutBalance;
double inBalance = 0.0;
TradingData market = CURRENT_DATA_SET.clone();
if (!market.hasNextPrice())
{
throw new Exception("No market data in the CURRENT_DATA_SET");
}
TimeStepDataPiece currentData;
int index = 0;
int numberOfTrades = 0;
do
{
currentData = market.getNextData();
// Provide state info to the black box inputs.
// Unless I'm mistaken, this library will already provide a 1.0 as the first input to the black box at all times.
for (int i = 0; i < currentData.ccis.Count; ++i)
{
box.InputSignalArray[i] = currentData.ccis[i];
}
// Activate the network.
box.Activate();
// Read the network's trading recommendation signal output.
bool allIn = (box.OutputSignalArray[0] - 0.5) > 0.0;
//Console.WriteLine("output = " + box.OutputSignalArray[0]);
// Take the action being recommended by the network.
if (allIn && inBalance == 0.0)
{
inBalance = outBalance / currentData.price;
outBalance = 0.0;
numberOfTrades++;
}
else if (!allIn && outBalance == 0.0)
{
outBalance = inBalance * currentData.price;
inBalance = 0.0;
numberOfTrades++;
}
++index;
} while (market.hasNextPrice());
market.resetPricePointer();
_evalCount++;
double effectiveOutValue = outBalance > 0.0 ? outBalance : inBalance * currentData.price;
double percentChange = (effectiveOutValue - startingOutBalance) / startingOutBalance;
// Fitness function 1. Award profit.
//double fitness = (percentChange + 1.0) / 2.0;
// Fitness function 2. Award approaching of maximum gainFactor.
double gainFactor = effectiveOutValue / startingOutBalance;
double fitness = gainFactor / market.MaximumGainFactor;
// Fitness function 3. Award approaching of maximum gainFactor, Punish approaching of maximum lossFactor.
//double gainFactor = percentChange > 0 ? effectiveOutValue / startingOutBalance : 0.0;
//double lossFactor = percentChange < 0 ? startingOutBalance / effectiveOutValue : 0.0;
//double fitness = (gainFactor - 1) / market.MaximumGainFactor - (lossFactor - 1) / market.MaximumLossFactor;
//fitness = (fitness + 1) / 2.0;
lock (staticFieldsLock)
{
if (fitness > BestFitnessOfGeneration)
{
BestFitnessOfGeneration = fitness;
BestFitnessOfGenerationGainPercent = percentChange;
BestFitnessOfGenerationNumberTrades = numberOfTrades;
}
if (fitness < WorstFitnessOfGeneration)
{
WorstFitnessOfGeneration = fitness;
}
MeanFitnessOfGeneration *= totalRun;
MeanFitnessOfGeneration += fitness;
MeanFitnessOfGeneration /= (double)++totalRun;
}
return(new FitnessInfo(fitness, fitness));
}
