public static BasicNetwork loadNetwork(byte[] networkBytes)
{
FilePersistence location = new FilePersistence();
location.AddStreamData(networkBytes);
EncogPersistedCollection collection = new EncogPersistedCollection(location, FileMode.Open);
BasicNetwork returnNetwork = (BasicNetwork)collection.Find("neuralNet");
//Make sure we have the flattended network cached.
returnNetwork.Structure.FinalizeStructure();
return(returnNetwork);
}
public static void saveNetwork(BasicNetwork network, string filename, string filePath)
{
if (filename.Substring(filename.Length - 4, 4) != ".eNN")
{
throw new Exception("Incorrect extension in filename");
}
EncogPersistedCollection collection = new EncogPersistedCollection(Path.Combine(filePath, filename), FileMode.Create);
if (filename.Contains("\\"))
{
int indexOfSlash = filename.IndexOf('\\');
collection.Add("neuralNet", network);
}
else
{
collection.Add("neuralNet", network);
}
}
public static void Go()
{
var encog = new EncogPersistedCollection("market-network.dat", FileMode.Open);
Console.WriteLine(@"Loading network");
var network = (BasicNetwork)encog.Find("market-network");
Console.WriteLine(@"Reading current data from db");
var market = new StockMarket();
market.Init(false);
var data = market.GetCurrentData();
Console.WriteLine(@"Running network on data");
var madness = new ModelMadness();
foreach (StockMarket.WorkableStockInfo info in data)
{
var input = InputOutputMadness.CreateInputs(info);
var neuralInput = new BasicMLData(input);
var output = network.Compute(neuralInput);
Console.WriteLine(@"Stock " + info.ViewToday.stock + @" will change " + output[0] + @"% in the next 20 trading days");
var future = new prediction
{
day = DateTime.Now.Date,
C20_Days_Close = 100 * (decimal)output[0],
stock = info.ViewToday.stock
};
madness.AddTopredictions(future);
}
madness.SaveChanges();
Console.WriteLine(@"Done - begin making $millions");
}
public static BasicNetwork loadNetwork(string filename, string filePath)
{
if (filename.Substring(filename.Length - 4, 4) != ".eNN")
{
throw new Exception("Unexpected network extension in filename");
}
string fullPath = Path.Combine(filePath, filename);
if (!File.Exists(fullPath))
{
throw new ArgumentException("Requested file does not exist - " + fullPath);
}
EncogPersistedCollection collection = new EncogPersistedCollection(fullPath, FileMode.Open);
BasicNetwork returnNetwork = (BasicNetwork)collection.Find("neuralNet");
//Make sure we have the flattended network cached.
returnNetwork.Structure.FinalizeStructure();
return(returnNetwork);
}
public static void Create()
{
var market = new StockMarket();
market.Init(true);
Console.WriteLine(@"Fetching training data from db");
var trainingData = market.GetTrainingData();
Console.WriteLine(@"Creating training set");
var inputs = new ArrayList();
var outputs = new ArrayList();
var cur = 1;
foreach (var info in trainingData)
{
Console.WriteLine(@"Adding record " + (cur++) + @" of " + trainingData.Count);
var input = InputOutputMadness.CreateInputs(info);
var output = new[] { InputOutputMadness.CreateOutput(info) };
inputs.Add(input);
outputs.Add(output);
}
Console.WriteLine(@"Created training set - saving");
var trainingSet = new BasicMLDataSet((double[][])inputs.ToArray(typeof(double[])), (double[][])outputs.ToArray(typeof(double[])));
var encog = new EncogPersistedCollection("market-training.dat", FileMode.Create);
encog.Add("market-training", trainingSet);
Console.WriteLine(@"saved");
}
public static void Run(bool useBPN, double?learnRate, double?momentum)
{
Log("Using " +
(useBPN ? "Back-Propagation training with learnRate = " + learnRate + ", momentum = " + momentum
: "Resilient training"));
Log("Loading training data");
var encog = new EncogPersistedCollection("market-training.dat", FileMode.Open);
var trainingSet = (BasicMLDataSet)encog.Find("market-training");
BasicNetwork network;
if (File.Exists("market-network.dat"))
{
Log("Loading network");
encog = new EncogPersistedCollection("market-network.dat", FileMode.Open);
network = (BasicNetwork)encog.Find("market-network");
}
else
{
Log("Creating new network");
network = new BasicNetwork();
// input & hiddens
foreach (var num in InputOutputMadness.InputAndHiddenLayerInfo)
{
network.AddLayer(new BasicLayer(num));
}
// output
network.AddLayer(new BasicLayer(1));
network.Structure.FinalizeStructure();
network.Reset();
}
BasicTraining train;
if (useBPN)
{
train = new Backpropagation(network, trainingSet, (double)learnRate, (double)momentum);
}
else
{
train = new ResilientPropagation(network, trainingSet);
}
//train = new NeuralGeneticAlgorithm(network, new FanInRandomizer(), new TrainingSetScore(trainingSet), 500, 0.1, 0.25);
Log("Beginning Training");
int epoch = 1;
do
{
train.Iteration();
Log("Epoch #" + epoch + " Error:" + train.Error);
epoch++;
if (epoch % 100 == 0)
{
Log("Saving current data");
encog = new EncogPersistedCollection("market-network.dat", FileMode.Create);
encog.Create();
encog.Add("market-network", network);
}
}while (true);
}
