static void Main(string[] args)
{
ServiceEvaNN serviceEvaNN = new ServiceEvaNN();
NetworkStructure netStructure = new NetworkStructure
{
InputVectorLength = 10,
NeuronsByLayers = new[] { 230, 150, 120, 1 }
};
TrainingConfiguration trainConfig = new TrainingConfiguration
{
TrainingAlgorithmType = TrainingAlgorithmType.RProp,
StartIteration = 0,
EndIteration = 10,
InputDatasetFilename = "TrainingSets//inputSets.txt",
OutputDatasetFilename = "TrainingSets//outputSets.txt",
MemoryFolder = "Memory"
};
bool creatingSucceed = serviceEvaNN.CreateNetwork(trainConfig.MemoryFolder, netStructure);
if (creatingSucceed)
{
//serviceEvaNN.CalculateStatistic(trainConfig);
serviceEvaNN.Train(trainConfig,
true,
ProcessPriorityClass.Normal,
true);
serviceEvaNN.CalculateStatistic(trainConfig);
}
Console.WriteLine("Done!");
Console.ReadKey();
}
/// <summary>
/// Training FeedForward - NeuralNetwork
/// </summary>
/// <param name="trainingConfiguration"></param>
/// <param name="printLearnStatistic"></param>
/// <param name="processPriorityClass"></param>
/// <param name="unsafeTrainingMode"></param>
/// <param name="iterationsToPause"></param>
public void Train(TrainingConfiguration trainingConfiguration,
bool printLearnStatistic = false,
ProcessPriorityClass processPriorityClass = ProcessPriorityClass.Normal,
bool unsafeTrainingMode = false,
int iterationsToPause = -1)
{
// Check for unexistent network:
if (_networkTeacher == null)
{
Logger.LogError(ErrorType.OperationWithNonexistentNetwork, "Training failed!");
return;
}
// Check for set of iterations to pause:
if (iterationsToPause == -1)
{
iterationsToPause = trainingConfiguration.EndIteration - trainingConfiguration.StartIteration;
}
// Print warning about using unsafe mode:
if (unsafeTrainingMode)
{
Logger.LogWarning(WarningType.UsingUnsafeTrainingMode);
}
trainingConfiguration.MemoryFolder = trainingConfiguration.MemoryFolder == "" ? "Memory" : trainingConfiguration.MemoryFolder;
// Start process timer:
Stopwatch stopWatch = new Stopwatch();
stopWatch.Start();
// Set the process priority class:
Process thisProc = Process.GetCurrentProcess();
thisProc.PriorityClass = processPriorityClass;
if (_networkTeacher.CheckMemory(trainingConfiguration.MemoryFolder) && _networkTeacher.CheckDatasets(trainingConfiguration.InputDatasetFilename, trainingConfiguration.OutputDatasetFilename, _networkStructure))
{
_networkTeacher.TrainNet(trainingConfiguration, iterationsToPause, unsafeTrainingMode);
// Stopping timer and print spend time in [HH:MM:SS]:
stopWatch.Stop();
TimeSpan ts = stopWatch.Elapsed;
string elapsedTime = String.Format("{0:00}:{1:00}:{2:00}", ts.Hours, ts.Minutes, ts.Seconds);
Console.WriteLine("Time spend: " + elapsedTime);
if (printLearnStatistic)
{
_networkTeacher.PrintLearningStatistic(trainingConfiguration, true, elapsedTime);
}
}
else
{
stopWatch.Stop();
Console.WriteLine("Training failed!");
}
}
public void CalculateStatistic(TrainingConfiguration trainingConfig)
{
if (_networkTeacher == null)
{
Logger.LogError(ErrorType.OperationWithNonexistentNetwork, "Calculate statistic failed!");
return;
}
_networkTeacher.PrintLearningStatistic(trainingConfig, true);
}
private void RefreshAfterMerge()
{
DataAccess.Configure("WaitForNonStaleData", true);
RebuildList(true);
DataAccess.Configure("WaitForNonStaleData", false);
_availableCategories = null;
Notify(() => AvailableCategories);
TrainingConfiguration.RefreshCategories();
SearchPanel.RefreshCategories();
}
//[InlineData(10000, 0.0000351085)]
public void When_Then(int iterationsCount, double expectedError)
{
var net = NeuralNetTestSetup.GetConfiguredNet();
var trainingSet = new TrainingSet(new[] { new TrainingSample(new[] { 0.05, 0.10 }, new[] { 0.01, 0.99 }) });
_trainingCfg = new TrainingConfiguration {
LearningRate = 0.5, MaxEpoch = iterationsCount
};
Backpropagation.Train(net, _trainingCfg, trainingSet, _ => { });
var actualError = ErrorFunctions.TotalError(net, trainingSet);
actualError.Should().BeApproximately(expectedError, Precision);
}
public void When_Training_Iteration_Performed_Then_TotalError_Decreases()
{
var net = new NeuralNetwork(2, 2, 1, ActivationFunctions.Sigmoid);
var trainingSet = PopularProblems.Xor;
var trainingCfg = new TrainingConfiguration {
LearningRate = 0.1, MaxEpoch = 10000
};
var trainingErrors = new List <double>();
Backpropagation.Train(net, trainingCfg, trainingSet, trainingErrors.Add);
trainingErrors.Should().BeInDescendingOrder();
}
public static void LogTrainingResults(int testPassed, int testFailed, TrainingConfiguration trainConfig, string elapsedTime)
{
// Check for existing this logs-directory:
if (!Directory.Exists(TrainLogsDirectoryName))
{
Directory.CreateDirectory(TrainLogsDirectoryName);
}
// Save log:
using (StreamWriter fileWriter = new StreamWriter(TrainLogsDirectoryName + "/" + trainConfig.EndIteration + ".txt"))
{
fileWriter.WriteLine("Test passed: " + testPassed);
fileWriter.WriteLine("Test failed: " + testFailed);
fileWriter.WriteLine("Percent learned: {0:f2}", (double)testPassed * 100 / (testPassed + testFailed));
// Writing additional training data:
fileWriter.WriteLine("\n============================\n");
// Writing elapsed time:
if (elapsedTime != "")
{
fileWriter.WriteLine("Time spend: " + elapsedTime);
fileWriter.WriteLine("Start iteration: " + trainConfig.StartIteration);
fileWriter.WriteLine("End iteration: " + trainConfig.EndIteration);
fileWriter.WriteLine("============================\n");
}
// Writing system characteristics:
fileWriter.WriteLine("System characteristics:");
fileWriter.WriteLine("OS Version: {0} {1}", Environment.OSVersion, Environment.Is64BitOperatingSystem ? "64 bit" : "32 bit");
fileWriter.WriteLine("Processors count: " + Environment.ProcessorCount);
ManagementClass myManagementClass = new ManagementClass("Win32_Processor");
PropertyDataCollection myProperties = myManagementClass.Properties;
foreach (var myProperty in myProperties)
{
switch (myProperty.Name)
{
case "CurrentClockSpeed":
case "Name":
fileWriter.WriteLine($"{myProperty.Name}: { myProperty.Value ?? "-"}");
break;
}
}
}
Console.WriteLine("Learn statistic logs saved in {0}!", TrainLogsDirectoryName);
}
public void TestTrainingConfiguration()
{
// Arrange:
TrainingConfiguration trainingConfig = new TrainingConfiguration
{
StartIteration = 0,
EndIteration = 1,
InputDatasetFilename = "test",
OutputDatasetFilename = "test",
MemoryFolder = "test",
TrainingAlgorithmType = TrainingAlgorithmType.BProp
};
// Act & Assert:
Assert.IsNotNull(trainingConfig);
}
private List <TrainingConfiguration> InitializeTrainingSubConfigs(TrainingConfiguration trainingConfig, int iterationsToPause)
{
List <TrainingConfiguration> trainingConfigs = new List <TrainingConfiguration>();
int currentIterPosition = trainingConfig.StartIteration;
while (true)
{
if (trainingConfig.EndIteration - currentIterPosition - 1 >= iterationsToPause)
{
var trainingConfigItem = new TrainingConfiguration
{
TrainingAlgorithmType = trainingConfig.TrainingAlgorithmType,
StartIteration = currentIterPosition,
EndIteration = currentIterPosition + iterationsToPause,
MemoryFolder = trainingConfig.MemoryFolder,
InputDatasetFilename = trainingConfig.InputDatasetFilename,
OutputDatasetFilename = trainingConfig.OutputDatasetFilename
};
trainingConfigs.Add(trainingConfigItem);
currentIterPosition += iterationsToPause;
}
else
{
var trainingConfigItem = new TrainingConfiguration
{
TrainingAlgorithmType = trainingConfig.TrainingAlgorithmType,
StartIteration = currentIterPosition,
EndIteration = trainingConfig.EndIteration,
MemoryFolder = trainingConfig.MemoryFolder,
InputDatasetFilename = trainingConfig.InputDatasetFilename,
OutputDatasetFilename = trainingConfig.OutputDatasetFilename
};
trainingConfigs.Add(trainingConfigItem);
break;
}
}
Console.WriteLine("Train sub-configuration objects created!");
return(trainingConfigs);
}
public static void Main()
{
var net = new NeuralNetwork(2, 2, 1, ActivationFunctions.Sigmoid);
var trainingCfg = new TrainingConfiguration {
LearningRate = 0.1, MaxEpoch = 512000
};
var trainingSet = PopularProblems.Xor;
var chartWindow = SimpleChart.LaunchInNewThread();
chartWindow.Title = "Learning progress";
Backpropagation.Train(net, trainingCfg, trainingSet, x => Task.Run(() => chartWindow.AddPoint(x)));
Console.WriteLine("Press any key to continue...");
Console.ReadKey();
}
private void CreateDataSets(TrainingConfiguration trainingConfiguration)
{
// Creating input sets:
using (StreamWriter fileWriter = new StreamWriter(trainingConfiguration.InputDatasetFilename))
{
fileWriter.WriteLine("0 0");
fileWriter.WriteLine("0 1");
fileWriter.WriteLine("1 0");
fileWriter.Write("1 1");
}
// Creating input sets:
using (StreamWriter fileWriter = new StreamWriter(trainingConfiguration.OutputDatasetFilename))
{
fileWriter.WriteLine("0");
fileWriter.WriteLine("0");
fileWriter.WriteLine("0");
fileWriter.Write("1");
}
}
/// <summary>
/// Обучение сети
/// </summary>
/// <param name="trainingConfig"></param>
/// <param name="iterationsToPause"></param>
/// <param name="unsafeTrainingMode"></param>
public void TrainNet(TrainingConfiguration trainingConfig, int iterationsToPause, bool unsafeTrainingMode = false)
{
Iteration = trainingConfig.EndIteration;
#region Load data from file
List <double[]> inputDataSets;
List <double[]> outputDataSets;
try
{
inputDataSets = FileManager.LoadTrainingDataset(trainingConfig.InputDatasetFilename);
outputDataSets = FileManager.LoadTrainingDataset(trainingConfig.OutputDatasetFilename);
}
catch (Exception ex)
{
Logger.LogError(ErrorType.SetMissing, ex);
return;
}
#endregion
Console.WriteLine("Training start...");
try
{
List <TrainingConfiguration> trainingConfigs = InitializeTrainingSubConfigs(trainingConfig, iterationsToPause);
// Initialize teachers:
SingleNetworkTeacher netSubTeacher = new SingleNetworkTeacher
{
Network = _net,
NetworkStructure = _networkStructure,
TrainingConfiguration = trainingConfig,
InputDatasets = inputDataSets.ToArray(),
OutputDatasets = outputDataSets.ToArray(),
SafeTrainingMode = !unsafeTrainingMode
};
// Iteration multithreading train:
for (int j = 0; j < trainingConfigs.Count; j++)
{
netSubTeacher.TrainingConfiguration = trainingConfigs[j];
Thread thread = new Thread(netSubTeacher.Train);
thread.Start();
Wait(thread);
if (!netSubTeacher.LastTrainingSuccess)
{
return;
}
if (j != trainingConfigs.Count - 1)
{
Console.WriteLine("Iterations already finished: " + iterationsToPause * (j + 1));
}
else
{
Console.WriteLine("Iterations already finished: " + trainingConfig.EndIteration);
}
// Test after this iteration's part
CommonTest(true);
}
// Проведение завершающих операций после обучения модели:
switch (trainingConfig.TrainingAlgorithmType)
{
// В случае обучения по генетическому алгоритму - загрузка памяти из файла:
case TrainingAlgorithmType.GeneticAlg:
case TrainingAlgorithmType.RProp:
// Загрузка только что сохраненной памяти:
_net = new NeuralNetwork(_networkStructure.NeuronsByLayers, "memory.txt", _networkStructure.Alpha);
break;
// В общем случае - получение данных обученной сети от "подучителя":
case TrainingAlgorithmType.BProp:
default:
_net = netSubTeacher.Network;
break;
}
Console.WriteLine("Training success!");
}
catch (Exception ex)
{
Logger.LogError(ErrorType.TrainError, ex);
}
}
/// <summary>
/// Printing network's learning statistic
/// </summary>
/// <param name="trainingConfig"></param>
/// <param name="withLogging"></param>
public void PrintLearningStatistic(TrainingConfiguration trainingConfig, bool withLogging = false, string elapsedTime = "")
{
Console.WriteLine("Start calculating statistic...");
// Loading memory:
try
{
_net = new NeuralNetwork(_networkStructure.NeuronsByLayers, "memory.txt", _networkStructure.Alpha);
}
catch (Exception ex)
{
Logger.LogError(ErrorType.MemoryInitializeError, ex);
}
// Testing:
int testPassed = 0;
int testFailed = 0;
#region Load data from file
List <double[]> inputDataSets;
List <double[]> outputDataSets;
try
{
inputDataSets = FileManager.LoadTrainingDataset(trainingConfig.InputDatasetFilename);
outputDataSets = FileManager.LoadTrainingDataset(trainingConfig.OutputDatasetFilename);
}
catch (Exception ex)
{
Logger.LogError(ErrorType.SetMissing, ex);
return;
}
#endregion
for (int i = 0; i < inputDataSets.Count; i++)
{
// Получение ответа:
string handlingErrorText = "";
double[] netResult = _net.Handle(inputDataSets[i], ref handlingErrorText);
if (netResult != null)
{
if (IsVectorsRoughlyEquals(outputDataSets[i], netResult, 0.3))
{
testPassed++;
}
else
{
testFailed++;
}
}
else
{
Logger.LogError(ErrorType.NonEqualsInputLengths, handlingErrorText);
return;
}
}
// Logging (optional):
if (withLogging)
{
Logger.LogTrainingResults(testPassed, testFailed, trainingConfig, elapsedTime);
}
Console.WriteLine("Test passed: {0}\nTest failed: {1}\nPercent learned: {2:f2}", testPassed,
testFailed,
(double)testPassed * 100 / (testPassed + testFailed));
}
public void TestRProp()
{
// Arrange:
ServiceEvaNN serviceEvaNN = new ServiceEvaNN();
NetworkStructure netStructure = new NetworkStructure
{
InputVectorLength = 2,
NeuronsByLayers = new[] { 23, 15, 1 }
};
TrainingConfiguration trainConfig = new TrainingConfiguration
{
TrainingAlgorithmType = TrainingAlgorithmType.RProp,
StartIteration = 0,
EndIteration = 100,
InputDatasetFilename = "TrainingSetsTest//inputSets.txt",
OutputDatasetFilename = "TrainingSetsTest//outputSets.txt",
MemoryFolder = "MemoryRProp"
};
if (!Directory.Exists("TrainingSetsTest"))
{
Directory.CreateDirectory("TrainingSetsTest");
CreateDataSets(trainConfig);
}
if (File.Exists(trainConfig.MemoryFolder + "//memory.txt"))
{
File.Delete(trainConfig.MemoryFolder + "//memory.txt");
}
bool creatingSucceed = serviceEvaNN.CreateNetwork(trainConfig.MemoryFolder, netStructure);
if (creatingSucceed)
{
serviceEvaNN.Train(trainConfig,
true,
ProcessPriorityClass.Normal,
true);
}
// Asserts:
double tolerance = 0.15;
double[] result0 = serviceEvaNN.Handle(new double[] { 0, 0 });
Assert.IsNotNull(result0);
Assert.AreEqual(0.0, result0[0], tolerance);
double[] result1 = serviceEvaNN.Handle(new double[] { 0, 1 });
Assert.IsNotNull(result1);
Assert.AreEqual(0.0, result1[0], tolerance);
double[] result2 = serviceEvaNN.Handle(new double[] { 1, 0 });
Assert.IsNotNull(result2);
Assert.AreEqual(0.0, result2[0], tolerance);
double[] result3 = serviceEvaNN.Handle(new double[] { 1, 1 });
Assert.IsNotNull(result3);
Assert.AreEqual(1.0, result3[0], tolerance);
}
