public static void LoadData(string modelFileName, string waveDataFileName)
{
LocalLogManager.AddLine("Machine Learning manager initialization");
_modelFileName = modelFileName;
_waveDataFileName = waveDataFileName;
InputCount = (MaxPathingPoints * CountPerPathing) +
(MaxWaterPoints * CountPerWater) +
(MaxTowers * CountPerTower) +
(MaxEnemies * CountPerEnemy);
_machineLearningModel = new DeepBeliefNetworkModel();
lock (FileLocker)
{
var canLoadExistingModel = _machineLearningModel.Load(_modelFileName);
if (!canLoadExistingModel)
{
_machineLearningModel.Initialize(InputCount, (int)(InputCount * 0.8));
}
}
var waveDataExists = FileManager.FileExists(_waveDataFileName);
if (waveDataExists)
{
_waveData = FileManager.BinaryDeserialize(typeof(WaveData), _waveDataFileName) as WaveData;
}
_waveData = _waveData ?? new WaveData
{
WaveInputs = new List <double[]>(),
WaveScores = new List <double>(),
};
}
private static void LoadTestData()
{
CSV = new StringBuilder();
CSV.Append("SampleSize, AVMMSE, MSE");
CSV.Append(Environment.NewLine);
var slr = 0.8;
var sm = 0.8;
var uslr = 0.9;
var usm = 0.2;
var usd = 0.21;
InputCount = (MaxPathingPoints * CountPerPathing) +
(MaxWaterPoints * CountPerWater) +
(MaxTowers * CountPerTower) +
(MaxEnemies * CountPerEnemy);
_machineLearningModel = new DeepBeliefNetworkModel();
_machineLearningModel.Initialize(InputCount, (int)(InputCount * 0.8), uslr, usm, usd, slr, sm);
var outputList = new List <double>();
var inputList = new List <double[]>();
_waveData = new WaveData
{
WaveInputs = inputList,
WaveScores = outputList,
};
var priorMSE = 0.0;
using (var reader = new StreamReader(@"wavedata.csv"))
{
var sampleSize = 0;
while (!reader.EndOfStream)
{
var line = reader.ReadLine();
var values = line.Split('\t');
var output = double.Parse(values[0]);
outputList.Add(output);
var inputCount = values.Length - 1;
var inputArray = new double[inputCount];
for (var i = 1; i < inputCount; i++)
{
inputArray[i - 1] = double.Parse(values[i]);
}
inputList.Add(inputArray);
_machineLearningModel.LearnAll(_waveData);
CSV.Append($"{++sampleSize}, {_machineLearningModel.AVMMSE}, {_machineLearningModel.LastMSE}");
CSV.Append(Environment.NewLine);
if ((sampleSize / 50 >= 1 && sampleSize % 50 == 0) || (sampleSize > 20 && Math.Abs(priorMSE - _machineLearningModel.LastMSE) > 0.05))
{
var wait = _machineLearningModel.LastMSE;
}
priorMSE = _machineLearningModel.LastMSE;
}
}
var fileName = "ModelData.csv";
File.WriteAllText(fileName, CSV.ToString());
}
private static void IncrementalTest()
{
var slr = 0.8;
var sm = 0.8;
var uslr = 0.9;
var usm = 0.2;
var usd = 0.21;
InputCount = (MaxPathingPoints * CountPerPathing) +
(MaxWaterPoints * CountPerWater) +
(MaxTowers * CountPerTower) +
(MaxEnemies * CountPerEnemy);
var outputList = new List <double>();
var inputList = new List <double[]>();
using (var reader = new StreamReader(@"wavedata.csv"))
{
while (!reader.EndOfStream)
{
var line = reader.ReadLine();
var values = line.Split('\t');
var output = double.Parse(values[0]);
outputList.Add(output);
var inputCount = values.Length - 1;
var inputArray = new double[inputCount];
for (var i = 1; i < inputCount; i++)
{
inputArray[i - 1] = double.Parse(values[i]);
}
inputList.Add(inputArray);
}
}
for (var modelTest = 5; modelTest <= 9; modelTest++)
{
_machineLearningModel = new DeepBeliefNetworkModel();
_machineLearningModel.Initialize(InputCount, (int)(InputCount * 0.8), uslr, usm, usd, slr, sm);
CSV = new StringBuilder();
CSV.Append("SampleSize, AVMMSE, MSE, LearnTime");
CSV.Append(Environment.NewLine);
var partialOutput = new List <double>();
var partialInput = new List <double[]>();
_waveData = new WaveData
{
WaveInputs = partialInput,
WaveScores = partialOutput,
};
var sampleSize = FlatRedBallServices.Random.Next(150, 630);
for (var i = 0; i <= sampleSize; i++)
{
var randomIndex = FlatRedBallServices.Random.Next(0, outputList.Count - 1);
partialInput.Add(inputList[randomIndex]);
partialOutput.Add(outputList[randomIndex]);
_machineLearningModel.LearnAll(_waveData);
CSV.Append($"{i + 1}, {_machineLearningModel.AVMMSE}, {_machineLearningModel.LastMSE}, {_machineLearningModel.LastLearnTime}");
CSV.Append(Environment.NewLine);
}
var fileName = $"Player{modelTest}Data.csv";
File.WriteAllText(fileName, CSV.ToString());
}
}
private static void SearchTest()
{
var fileName = "CollectedData.csv";
InputCount = (MaxPathingPoints * CountPerPathing) +
(MaxWaterPoints * CountPerWater) +
(MaxTowers * CountPerTower) +
(MaxEnemies * CountPerEnemy);
var slrList = new List <double>();
var smList = new List <double>();
var uslrList = new List <double>();
var usmList = new List <double>();
var usdList = new List <double>();
CSV = new StringBuilder();
File.WriteAllText(fileName, CSV.ToString());
for (var slr = 0.0; slr <= 1; slr += 0.1)
{
slrList.Add(slr);
}
for (var sm = 0.0; sm <= 1; sm += 0.1)
{
smList.Add(sm);
}
for (var uslr = 0.0; uslr <= 1; uslr += 0.1)
{
uslrList.Add(uslr);
}
for (var usm = 0.0; usm <= 1; usm += 0.1)
{
usmList.Add(usm);
}
for (var usd = 0.01; usd <= 0.3; usd += 0.025)
{
usdList.Add(usd);
}
var triedCombos = new List <string>();
var outputList = new List <double>();
var inputList = new List <double[]>();
_waveData = new WaveData
{
WaveInputs = inputList,
WaveScores = outputList,
};
using (var reader = new StreamReader(@"wavedata.csv"))
{
while (!reader.EndOfStream)
{
var line = reader.ReadLine();
var values = line.Split('\t');
var output = double.Parse(values[0]);
outputList.Add(output);
var inputCount = values.Length - 1;
var inputArray = new double[inputCount];
for (var i = 1; i < inputCount; i++)
{
inputArray[i - 1] = double.Parse(values[i]);
}
inputList.Add(inputArray);
}
}
DeepBeliefNetworkModel baseModel = new DeepBeliefNetworkModel();
baseModel.Initialize(InputCount, (int)(InputCount * 0.8), 0, 0, 0, 0, 0);
baseModel.Save("BaseModel.model");
DeepBeliefNetworkModel newModel = baseModel;
while (true)
{
var slr = FlatRedBallServices.Random.In(slrList);
var sm = FlatRedBallServices.Random.In(smList);
var uslr = FlatRedBallServices.Random.In(uslrList);
var usm = FlatRedBallServices.Random.In(usmList);
var usd = FlatRedBallServices.Random.In(usdList);
var combo = $"{slr}{sm}{uslr}{usm}{usd}";
if (triedCombos.Contains(combo))
{
continue;
}
triedCombos.Add(combo);
CSV.Clear();
newModel.Initialize(InputCount, (int)(InputCount * 0.8), uslr, usm, usd, slr, sm);
newModel.Load("BaseModel.model");
CSV.Append($"{slr},{sm},{uslr},{usm},{usd},");
newModel.LearnAll(_waveData);
CSV.Append(newModel.LastMSE);
CSV.Append(Environment.NewLine);
File.AppendAllText(fileName, CSV.ToString());
CSV.Clear();
if (newModel.LastMSE < 0.03)
{
newModel.Save("BESTMODEL" + newModel.LastMSE + "SAVE.model");
}
}
}
