public void LearnOneExample(Example example)
{
if (savedExamples.Count >= n)
{
//delete the oldest example
savedExamples.RemoveAt(1);
}
savedExamples.Add(example);
}
public void PropogateInput(Example example)
{
List<BestExample> best = new List<BestExample>();
double maxBest = 0.0;
foreach (Example e in savedExamples)
{
double d = distance_sq(e, example);
if (best.Count < k)
{
best.Add(new BestExample(e, d));
if (d > maxBest) { maxBest = d; }
}
else if (d < maxBest)
{
// at this point, the Best list contains k elements
best.Sort();
maxBest = best[k - 1].distance;
best[k - 1] = new BestExample(e, d);
}
}
if (best.Count < 1)
throw new LearnerNotReady();
// perform a weighted average
List<Double> result = new List<Double>();
for (int i = 0; i < best[0].example.Labels.Count; i++)
{
result.Add(0.0);
}
double sumWeight = 0.0;
foreach (BestExample be in best)
{
double thisWeight = 1 / (1 + be.distance);
for (int i = 0; i < result.Count; i++)
{
result[i] += be.example.Labels[i] * thisWeight;
}
sumWeight += thisWeight;
}
for (int i = 0; i < result.Count; i++)
{
result[i] /= sumWeight;
}
example.Predictions = result;
}
public override void LearnOneExample(Example example)
{
Network.LearnOneExample(example);
}
public abstract void LearnOneExample(Example example);
public override void LearnOneExample(Example example)
{
learner.LearnOneExample(example);
}
private void PropogateErrors(Example example)
{
AssertValidLabels(example);
Neuron.UnfeedAll(Neurons);
int label = 0;
foreach (Neuron neuron in Outputs)
{
neuron.UpdateErrorTerm(example.Labels[label++]);
neuron.Fed = true;
}
foreach (Neuron neuron in Neurons)
recur_BackPropogation(neuron);
}
private void AssertValidLabels(Example example)
{
if (example.Labels.Count != Outputs.Count)
throw new Exception("The number of labels must match the number of Output Neurons.\r\n" + "Labels: " + example.Features.Count + " Outputs: " + Inputs.Count);
}
private void AssertValidForTraining(Example example)
{
AssertValidFeatures(example);
AssertValidLabels(example);
}
private void AssertValidFeatures(Example example)
{
if (example.Features.Count != Inputs.Count)
throw new Exception("The number of features must match the number of Input Neurons.\r\n" + "Features: " + example.Features.Count + " Inputs: " + Inputs.Count);
}
/// <summary>
/// Propogates input through network (FeedForward). Updates example.predictions.
/// </summary>
public void PropogateInput(Example example)
{
AssertValidFeatures(example);
Neuron.UnfeedAll(Neurons);
int feature = 0;
foreach (Neuron neuron in Inputs)
{
neuron.Value = example.Features[feature++];
neuron.Fed = true;
}
foreach (Neuron neuron in Constants)
{
neuron.Value = 1;
neuron.Fed = true;
}
foreach (Neuron neuron in Neurons)
recur_FeedForward(neuron);
example.Predictions = Outputs.Select(o => o.Value).ToList();
}
public void LearnOneExample(Example example)
{
PropogateInput(example);
PropogateErrors(example);
UpdateWeights();
}
private double distance_sq(Example e1, Example e2)
{
double sum = 0.0;
Debug.Assert(e1.Features.Count == e2.Features.Count);
for (int i = 0; i < e1.Features.Count; i++)
{
double x1 = e1.Features[i];
double x2 = e2.Features[i];
sum += Math.Pow(x1 - x2, 2.0);
}
return sum;
}
public BestExample(Example e, double d)
{
example = e;
distance = d;
}