internal LearningResultUpdatedEventArgs(LearningEpoch epoch, double mse)
{
Contract.Requires(epoch != null);
Contract.Requires(mse >= 0.0);
Epoch = epoch;
MSE = mse;
}
internal LearningResult(LearningEpoch epoch, bool bestHolder = false)
: base(epoch.SyncRoot)
{
Contract.Requires(epoch != null);
Epoch = epoch;
this.bestHolder = bestHolder;
}
private static void Begin()
{
bool recurrent = false;
var trainingProv = CreateProvider(10000, recurrent);
var trainingStrat = new GaussianBatchingStrategy(.5);
//var trainingStrat = new MonteCarloBatchingStrategy();
var trainingBatcher = new ScriptCollectionBatcher(trainingStrat, trainingProv, 200, 500);
var validProv = CreateProvider(1000, recurrent);
var validStrat = new MonteCarloBatchingStrategy();
var validBatcher = new ScriptCollectionBatcher(validStrat, validProv, 25, 10000);
trainingBatcher.Initialize();
validBatcher.Initialize();
Console.WriteLine("Training samples: " + trainingProv.Count);
Console.WriteLine("Validation samples: " + validProv.Count);
// Rules:
Console.WriteLine("Creating learning rules ...");
var weightInitRule = new NoisedWeightInitializationRule { Noise = 0.5, IsEnabled = true };
//var learningRule = new QuickpropRule { StepSize = 0.001 };
var learningRule = new SCGRule();
//var learningRule = new LMRule();
//var learningRule = new MetaQSARule { Mode = LearningMode.Stochastic, Momentum = 0.1, StepSizeRange = new DoubleRange(0.0, 0.005), StepSize = 0.001, StochasticAdaptiveStateUpdate = true };
//var learningRule = new SuperSABRule { Mode = LearningMode.Batch, Momentum = 0.8, StepSizeRange = new DoubleRange(0.0, 0.01), StepSize = 0.005, StochasticAdaptiveStateUpdate = false };
//var learningRule = new SignChangesRule { Mode = LearningMode.Stochastic, Momentum = 0.2, StepSizeRange = new DoubleRange(0.0, 0.001), StepSize = 0.001, StochasticAdaptiveStateUpdate = false };
//var learningRule = new GradientDescentRule { Mode = LearningMode.Stochastic, Momentum = 0.2, StepSize = 0.001 };
//var learningRule = new QSARule();
//var learningRule = new MAQRule();
//var learningRule = new AdaptiveAnnealingRule { WeightGenMul = 0.1, AcceptProbMul = 0.05 };
//var learningRule = new RpropRule { Momentum = 0.01, StepSize = 0.01 };
//var learningRule = new CrossEntropyRule { PopulationSize = 50, NumberOfElites = 10, MutationChance = 0.01, MutationStrength = 0.01, DistributionType = DistributionType.Gaussian };
//var learningRule = new GARule { PopulationSize = 40, MutationStrength = 0.01, MutationChance = 0.01 };
var wdRule = (ILearningRule)learningRule as IWeightDecayedLearningRule;
if (wdRule != null)
{
wdRule.WeightDecay = new WeightDecay { Factor = -0.0001, IsEnabled = false };
}
IterationRepeatPars iterationRepeatPars = new IterationRepeatPars(5, 10);
// Net:
Console.WriteLine("Creating Neural Network ...");
var network = CreateNetwork(recurrent, weightInitRule, learningRule);
var exec = new LearningExecution(network, iterationRepeatPars);
// Epoch:
Console.WriteLine("Initializing epoch ...");
var epoch = new LearningEpoch(exec, trainingBatcher, validBatcher, 1);
epoch.Initialize();
epoch.CurrentResult.Updated += (sender, e) => WriteResult(epoch);
epoch.BestValidationResult.Updated += (sender, e) => vbestNet = network.Clone();
// Training loop:
Console.WriteLine("Starting ...");
bool done = false;
do
{
//CodeBench.By("Epoch").Do = () =>
//{
// epoch.Step();
//};
//CodeBench.By("Epoch").WriteToConsole();
epoch.Step();
//WriteResult(epoch);
if (Console.KeyAvailable)
{
var key = Console.ReadKey();
switch (key.Key)
{
case ConsoleKey.Escape:
done = true;
break;
case ConsoleKey.S:
Save(network.Clone());
break;
case ConsoleKey.V:
if (vbestNet != null) Save(vbestNet);
break;
}
}
}
while (!done);
}
private static void WriteResult(LearningEpoch epoch)
{
Console.WriteLine("{0}: Current: {1}/{2} Validation: {3}/{4}",
epoch.CurrentIteration.ToString("0000"),
epoch.BestResult.MSE.ToString("0.000000"),
epoch.CurrentResult.MSE.ToString("0.000000"),
epoch.BestValidationResult.MSE.ToString("0.000000"),
epoch.CurrentValidationResult.MSE.ToString("0.000000"));
}