public void Simulate()
{
var simulationNetwork = Workspace.Instance.GetNetworkFile(NetworkName);
var simulationPlugin = new NeuralSimulationPlugin(simulationNetwork, null, 0);
SimulationTemplate.CreateSimulationForm(simulationPlugin, "", SpaceTemplateFactory, TestingSimulationConfiguration, 50).Show();
}
private void Test(NeuralSimulationPlugin trainingPlugin, FileInfo testingNetworkFile, FileInfo testingBestNetworkFile, ref double testingBestReward, ref int testingBestReachedGoals, ChartForm chartForm)
{
trainingPlugin.Network.Save(testingNetworkFile);
var testingSeed = Guid.NewGuid().GetHashCode();
var testingPlugin = new NeuralSimulationPlugin(testingNetworkFile, null, testingSeed);
var testingSimulation = SimulationTemplate.CreateSimulation(testingPlugin, "Testing", SpaceTemplateFactory, TestingSimulationConfiguration);
testingSimulation.Initialise();
while (testingSimulation.Iterate())
{
;
}
foreach (var serie in testingSimulation.Results.Series)
{
chartForm.InvokeAction(() => chartForm.AddPoint(serie.Key, serie.Value.Last()));
}
var predictions = string.Join(", ", testingPlugin.Predictions.Select(p => $"{p,4}"));
if (lines % 20 == 0)
{
var headers = testingSimulation.Results.ToHeaderStrings();
WriteLine($" {headers[0]} Predictions (-,N,S,E,W)");
WriteLine($" {headers[1]} ============================");
}
if (testingPlugin.TotalReward - testingBestReward > -0.00001)
{
testingBestReward = testingPlugin.TotalReward;
testingBestReachedGoals = testingSimulation.TotalReachedGoals;
trainingPlugin.Network.Save(testingBestNetworkFile.Ensure());
WriteLine($"Saved {testingSimulation.Results.ToDataString()} {predictions}");
}
else
{
WriteLine($" {testingSimulation.Results.ToDataString()} {predictions}");
}
}
public void Train()
{
var trainingNetwork = Workspace.Instance.GetNetworkFile(NetworkName);
var trainingPlugin = new NeuralSimulationPlugin(trainingNetwork, TrainingPluginConfiguration, 0);
var trainingSimulation = SimulationTemplate.CreateSimulation(trainingPlugin, "Training", SpaceTemplateFactory, TrainingSimulationConfiguration);
if (TrainingProgressTrackerConfiguration.Enabled)
{
var testingNetworkFile = TrainingOuputDirectory.GetFile("Last.model");
var testingBestNetworkFile = TrainingOuputDirectory.GetFile("Best.model");
trainingPlugin.Network.Save(testingNetworkFile);
var testingBestReward = 0.0;
var testingBestReachedGoals = 0;
var chartForm = new ChartForm("Neural Agent Reinforcement Learning");
Task.Run(() =>
{
try
{
DateTime startTime = DateTime.Now;
WriteLine($"Training Start: {startTime}");
while (trainingSimulation.Iterate())
{
Console.Title = trainingSimulation.ToString();
if (trainingSimulation.Iteration % TrainingProgressTrackerConfiguration.Interval == 0)
{
Test(trainingPlugin, testingNetworkFile, testingBestNetworkFile, ref testingBestReward, ref testingBestReachedGoals, chartForm);
lines++;
}
}
WriteLine($"Training Finished: {DateTime.Now}");
WriteLine($"Training Time: {DateTime.Now - startTime}");
chartForm.InvokeAction(() =>
{
Bitmap chartScreenshot = new Bitmap(chartForm.Width, chartForm.Height);
chartForm.DrawToBitmap(chartScreenshot, new Rectangle(0, 0, chartForm.Width, chartForm.Height));
var chartScreenshotFile = TrainingOuputDirectory.GetFile("Progress.png");
chartScreenshot.Save(chartScreenshotFile.FullName);
});
if (testingBestReward > 0)
{
string message = $"Do you want to update network '{NetworkName}'?\r\n\r\nTotal Reached Goals: {testingBestReachedGoals}\r\nTotal Reward: {testingBestReward}";
if (MessageBox.Show(message, "Training Results", MessageBoxButtons.YesNo, MessageBoxIcon.Question) == DialogResult.Yes)
{
new AgentNetwork(testingBestNetworkFile).Save(trainingPlugin.NetworkFile);
}
}
}
catch (Exception exception)
{
WriteLine(exception.ToString());
}
});
chartForm.ShowDialog();
}
}