public static void Main(string[] args)
{
var options = new Options();
var parser = Parser.Default;
string invokedCommand = "";
object invokedSubOptions = null;
if (!parser.ParseArgumentsStrict(args, options, (command, subOptions) =>
{
invokedCommand = command;
invokedSubOptions = subOptions;
}))
{
Environment.Exit(Parser.DefaultExitCodeFail);
}
var verbose = ((CommonSubOptions)invokedSubOptions).Verbose;
if (verbose)
{
Console.WriteLine(Options.GetHeading().ToString());
Console.WriteLine();
}
switch (invokedCommand)
{
case "train":
var trainSubOptions = (TrainSubOptions)invokedSubOptions;
if (!trainSubOptions.Validate()) Environment.Exit(Parser.DefaultExitCodeFail);
var learningRate = trainSubOptions.LearningRate;
var stepFunction = trainSubOptions.StepFunction;
var weightsLimit = trainSubOptions.InitialWeights;
var useAdaline = trainSubOptions.UseAdaline;
var perceptron = PerceptronTrainer.CreatePerceptron(
learningRate,
weightsLimit,
2,
stepFunction,
useAdaline
);
if (useAdaline)
{
var adalineThreshold = trainSubOptions.AdalineThreshold;
PerceptronTrainer.TrainPerceptron_And(perceptron, adalineThreshold, verbose);
}
else
{
PerceptronTrainer.TrainPerceptron_And(perceptron);
}
if (!PerceptronTrainer.Test_And(perceptron, verbose))
{
Console.Error.WriteLine("Test data failed!");
}
if (trainSubOptions.Octave)
{
ConsoleHelper.WriteYellowLine("Octave plot code:");
Console.WriteLine(perceptron.GenerateOctaveCode());
Console.WriteLine();
}
Console.Write(perceptron.ToJson());
break;
case "experiment":
var experimentSubOptions = (ExperimentSubOptions)invokedSubOptions;
if (!experimentSubOptions.Validate()) Environment.Exit(Parser.DefaultExitCodeFail);
switch (experimentSubOptions.Type)
{
case ExperimentRunner.ExperimentType.LearningRate:
ExperimentRunner.LearningRate(
experimentSubOptions.From,
experimentSubOptions.To,
experimentSubOptions.Step,
experimentSubOptions.Repetitions,
experimentSubOptions.InitialWeights,
experimentSubOptions.StepFunction,
2,
experimentSubOptions.UseAdaline,
experimentSubOptions.AdalineThreshold,
experimentSubOptions.Verbose
);
break;
case ExperimentRunner.ExperimentType.InitialWeights:
ExperimentRunner.WeightsRange(
experimentSubOptions.From,
experimentSubOptions.To,
experimentSubOptions.Step,
experimentSubOptions.Repetitions,
experimentSubOptions.LearningRate,
experimentSubOptions.StepFunction,
2,
experimentSubOptions.UseAdaline,
experimentSubOptions.AdalineThreshold,
experimentSubOptions.Verbose
);
break;
case ExperimentRunner.ExperimentType.AdalineThreshold:
ExperimentRunner.AdalineTreshold(
experimentSubOptions.From,
experimentSubOptions.To,
experimentSubOptions.Step,
experimentSubOptions.Repetitions,
experimentSubOptions.LearningRate,
experimentSubOptions.InitialWeights,
2,
experimentSubOptions.Verbose
);
break;
case ExperimentRunner.ExperimentType.AdalineError:
ExperimentRunner.AdalineError(
experimentSubOptions.LearningRate,
experimentSubOptions.InitialWeights,
experimentSubOptions.AdalineThreshold,
2
);
break;
case ExperimentRunner.ExperimentType.Error:
ExperimentRunner.Error(
experimentSubOptions.LearningRate,
experimentSubOptions.InitialWeights,
experimentSubOptions.StepFunction,
2
);
break;
default:
throw new ArgumentOutOfRangeException();
}
break;
case "test":
var testSubOptions = (TestSubOptions)invokedSubOptions;
if (!testSubOptions.Validate()) Environment.Exit(Parser.DefaultExitCodeFail);
var json = File.ReadAllText(testSubOptions.PerceptronJsonPath);
var p = PerceptronTrainer.CreatePerceptron(json);
if (p == null)
{
ConsoleHelper.WriteErrorLine("Perceptron json is invalid");
Environment.Exit(Parser.DefaultExitCodeFail);
}
if (testSubOptions.TestAnd) PerceptronTrainer.Test_And(p, true);
if (!double.IsNaN(testSubOptions.X1) && !double.IsNaN(testSubOptions.X2))
{
var input = new[] { testSubOptions.X1, testSubOptions.X2 };
Console.WriteLine(p.Feedforward(input));
}
if (testSubOptions.Octave) Console.Write(p.GenerateOctaveCode());
break;
default:
Environment.Exit(Parser.DefaultExitCodeFail);
break;
}
Environment.Exit(0);
}
public static void AdalineTreshold(
double start,
double end,
double step,
int repetitions,
double learningRate,
double weightsLimit,
int inputsCount,
bool verbose = false)
{
if (start > end || step > Math.Abs(end - start) || step == 0)
{
throw new ArgumentException();
}
CsvPrinter.DumpParams(
new KeyValuePair <string, object>("start", start),
new KeyValuePair <string, object>("end", end),
new KeyValuePair <string, object>("step", step),
new KeyValuePair <string, object>("repetitions", repetitions),
new KeyValuePair <string, object>("weightsLimit", weightsLimit),
new KeyValuePair <string, object>("learningRate", learningRate),
new KeyValuePair <string, object>("inputsCount", inputsCount),
new KeyValuePair <string, object>("verbose", verbose)
);
CsvPrinter.DumpHeaderLine("n", "adaline threshold", "min epochs", "max epochs", "avg epochs", "final error");
int experimentIndex = 0;
for (double adalineThreshold = start; adalineThreshold <= end; adalineThreshold += step)
{
experimentIndex++;
var minEpochs = int.MaxValue;
var epochsSum = 0;
var maxEpochs = 0;
var run = 0;
Perceptron p = null;
if (verbose)
{
ConsoleHelper.WriteYellowLine($"Experiment - {experimentIndex}");
}
try
{
for (int j = 0; j < repetitions; j++)
{
run++;
p = PerceptronTrainer.CreatePerceptron(learningRate, weightsLimit, inputsCount, StepFunction.Bipolar, true);
var epochs = PerceptronTrainer.TrainPerceptron_And(p, adalineThreshold, verbose);
if (epochs < minEpochs)
{
minEpochs = epochs;
}
if (epochs > maxEpochs)
{
maxEpochs = epochs;
}
epochsSum += epochs;
}
}
catch (PerceptronLearnException)
{
if (verbose)
{
ConsoleHelper.WriteErrorLine("Perceptron can't learn with params:");
}
p?.Dump();
}
var avarageEpochs = epochsSum / run;
var currentError = PerceptronTrainer.GetAdalineError(p);
CsvPrinter.DumpLine(experimentIndex, adalineThreshold, minEpochs, maxEpochs, avarageEpochs, currentError);
}
}
public static int TrainPerceptron_And(
Perceptron perceptron,
double adalineThreshold = 1,
bool verbose = false,
bool dumpData = false
)
{
bool isTrained = false;
int epoch = 0;
if (perceptron.IsAdaline)
{
if (verbose)
{
ConsoleHelper.WriteYellowLine($"Using adaline, error treshold - {adalineThreshold}");
}
double errorSum;
do
{
epoch++;
errorSum = 0;
if (verbose)
{
ConsoleHelper.WriteYellow($"Learning epoch - {epoch}");
}
foreach (var trainObject in andTraingData.Shuffle())
{
errorSum += Math.Pow(perceptron.Train(trainObject), 2);
}
errorSum /= andTraingData.Count;
if (verbose)
{
ConsoleHelper.WriteLine($" current error - {errorSum}");
}
if (dumpData)
{
CsvPrinter.DumpLine(epoch, errorSum);
}
if (epoch <= MaximumEpochs)
{
continue;
}
if (verbose)
{
ConsoleHelper.WriteErrorLine("Stopping!");
ConsoleHelper.WriteErrorLine("Did not learn nothing in 10000 epochs!");
ConsoleHelper.WriteErrorLine($"Using adaline, current values: error-{errorSum} > threshold-{adalineThreshold}");
}
return(0);
}while (errorSum > adalineThreshold);
}
else
{
while (!isTrained)
{
epoch++;
if (verbose)
{
ConsoleHelper.WriteLine($"Learning epoch - {epoch}");
}
double errorSum = 0;
errorSum = 0;
foreach (var trainObject in andTraingData)
{
var error = Math.Abs(perceptron.Train(trainObject));
errorSum += error;
}
if (Math.Abs(errorSum) < 0.000001)
{
isTrained = true;
}
if (dumpData)
{
CsvPrinter.DumpLine(epoch, errorSum);
}
if (epoch <= MaximumEpochs)
{
continue;
}
if (verbose)
{
ConsoleHelper.WriteErrorLine("Stopping!");
ConsoleHelper.WriteErrorLine("Did not learn nothing in 10000 epochs!");
}
return(0);
}
}
return(epoch);
}
