private static void TestNetwork(
PerceptronNetwork network,
List <DataSet> validationDataSet)
{
int nbGood = 0;
int nbBad = 0;
foreach (var dataSet in validationDataSet)
{
Console.Write("Input: " +
string.Join(" ", dataSet.Values
.Select(m => string.Format("{0:0.0}", m))));
var result = network.Compute(dataSet.Values.ToArray());
Console.Write(" Expected: " +
string.Join(" ", dataSet.Targets
.Select(m => string.Format("{0:0.0}", m))));
Console.Write(" Computed: " +
string.Join(" ", result
.Select(m => string.Format("{0:0.0}", m))));
int indexMaxExpected = !dataSet.Values.Any()
? -1
: dataSet.Targets
.Select((value, index) => new { Value = value, Index = index })
.Aggregate((a, b) => (a.Value > b.Value) ? a : b)
.Index;
int indexMaxComputed = !result.Any()
? -1
: result
.Select((value, index) => new { Value = value, Index = index })
.Aggregate((a, b) => (a.Value > b.Value) ? a : b)
.Index;
if (indexMaxComputed == indexMaxExpected)
{
Console.WriteLine(" GOOD RESULT");
nbGood++;
}
else
{
Console.WriteLine(" BAD RESULT");
nbBad++;
}
}
Console.WriteLine();
Console.WriteLine("# Good : " + nbGood);
Console.WriteLine("# Bad : " + nbBad);
Console.WriteLine("% Good : " +
string.Format("{0:0.0}", 100.0 * nbGood / (nbGood + nbBad)));
Console.WriteLine();
}
static void Main(string[] args)
{
Stopwatch stopwatch = new Stopwatch();
stopwatch.Start();
// Csv file.
var fileName = "iris.csv";
// sepal_length,sepal_width,petal_length,petal_width
int numberOfValues = 4; // Fields in csv file.
// is_setosa,is_versicolor,is_virginica (can be: 0 = false or 1 = true)
int numberOfTargets = 3; // Fields in csv file.
PerceptronNetwork network = new PerceptronNetwork(
inputSize: numberOfValues,
hiddenSize: 3,
outputSize: numberOfTargets,
learnRate: 0.4,
momentum: 0.95);
List <DataSet> fullDataset = LoadCsv(fileName, numberOfValues, numberOfTargets);
Shuffle(fullDataset);
int validationProportion = 15; // %
int crossValidationProportion = 15; // %
List <DataSet> validationDataSet = fullDataset
.Take(fullDataset.Count * validationProportion / 100)
.ToList();
List <DataSet> crossValidationDataSet = fullDataset
.Skip(fullDataset.Count * validationProportion / 100)
.Take(fullDataset.Count * crossValidationProportion / 100)
.ToList();
List <DataSet> trainDataset = fullDataset
.Skip(fullDataset.Count *
(100 - validationProportion - crossValidationProportion) / 100)
.ToList();
Console.WriteLine("Training started...\n");
int iterations = 10;
for (int i = 0; i < iterations; i++)
{
network.Train(
values: trainDataset.Select(m => m.Values).ToArray(),
targets: trainDataset.Select(m => m.Targets).ToArray(),
numEpochs: 100);
var errors = new List <double>();
foreach (var dataSet in trainDataset)
{
network.Compute(dataSet.Values);
errors.Add(network.CalculateError(dataSet.Targets));
}
var error = errors.Average();
Console.WriteLine("Iteration #" + (i + 1) + ", cost = " + error);
}
Console.WriteLine("\nTraining completed.\n");
Console.WriteLine("Testing network with train dataset:\n");
TestNetwork(network, trainDataset);
Console.WriteLine("Testing network with validation dataset:\n");
TestNetwork(network, validationDataSet);
Console.WriteLine("Testing network with cross validation dataset:\n");
TestNetwork(network, crossValidationDataSet);
stopwatch.Stop();
Console.WriteLine("Duration: " + stopwatch.Elapsed);
Console.WriteLine("Press any key to continue...");
Console.ReadLine();
}
