public void Run()
{
// get iris file from resource stream
Assembly assembly = Assembly.GetExecutingAssembly();
var f = assembly.GetManifestResourceStream("HillClimbing.Resources.iris.csv");
StreamReader sr = new StreamReader(f);
// load all records into new dataset
DataSet ds = DataSet.Load(sr);
sr.Close();
Dictionary<string, int> species = ds.EncodeOneOfN(4); // encode specie names using one of n encoding
// get a list of records with data and encoded class name
List<Record> trainingData = ds.ExtractSupervised(0, 4, 4, 3);
Network net = new Network(4, 4, 2); // new RBF network
net.Randomize(); // initialize the networks state
Scorer score = new Scorer(trainingData); // new scorer
HillClimb hc = new HillClimb(net, score, 1.2, 1); // new hill climbing training algorithm
Iterate(hc, 100, 0.01); // iterate through hill climbing algorithm
QueryOneOfN(net, trainingData, species); // display ideal and actual specie names
}
/// <summary>
/// Determines a score for a RBF network.
/// </summary>
/// <param name="net">RBF network to determine a score for.</param>
/// <returns>Double that represents the score.</returns>
public double Score(Network net)
{
_errCalc.Clear(); // clear the error calculator
foreach(var data in _trainingData) {
double[] output = net.ComputeRegression(data.Input); // send input through RBFs in network
_errCalc.UpdateError(output, data.Ideals); // updates error based on new outputs
}
return _errCalc.Calculate(); // calculates MSE error
}
/// <summary>
/// Goes through each record in the training set and displays the ideal output and the actual output.
/// </summary>
/// <param name="net">RBF network.</param>
/// <param name="trainingData">Training data.</param>
/// <param name="specieNames">Specie names.</param>
private void QueryOneOfN(Network net, List<Record> trainingData, Dictionary<string, int> species)
{
// invert the specie list
Dictionary<int, string> invSpecies = new Dictionary<int, string>();
foreach(string key in species.Keys) {
int value = species[key];
invSpecies[value] = key;
}
// for each training record
foreach(Record rec in trainingData) {
double[] output = net.ComputeRegression(rec.Input); // run the input through the RBF network
int idealIdx = GetMax(rec.Ideals); // find the index of the ideal name
int actualIdx = GetMax(output); // find the index of the actual name
Console.WriteLine("Guess: " + invSpecies[actualIdx] + " Actual: " + invSpecies[idealIdx]);
}
}