void RuleBlock::activate() {
FL_DBG("===================");
FL_DBG("ACTIVATING RULEBLOCK " << getName());
if (not getActivation()) {
setActivation(new General);
}
FL_DBG("Activation: " << getActivation()->className() << " " << getActivation()->parameters());
getActivation()->activate(this);
}
// Train network on dataset `data` with learning rate `learnRate` for `epochs` epochs
void Network::train(Dataset &data, double learnRate, int epochs) {
// Loop over epochs
for (int iter=0; iter < epochs; iter++) {
// Loop over samples
for (int i=0; i < data.size(); i++) {
std::vector<double> sampleFeatures;
std::vector<bool> sampleLabels;
std::vector<double> outputDelta(numOutput,0);
std::vector<double> hiddenDelta(numHidden,0);
// Read sample and calculate activations
data.sample(i, sampleFeatures, sampleLabels);
getActivation(sampleFeatures);
// Output layer delta
for (int j=0; j < numOutput; j++) {
outputDelta[j] = SIGDERIV(input[2][j]) * (sampleLabels[j] - activation[2][j]);
}
// Hidden layer delta
for (int j=0; j < numHidden; j++) {
for (int k=0; k < numOutput; k++) {
hiddenDelta[j] += outputWeights[k][j+1] * outputDelta[k];
}
hiddenDelta[j] *= SIGDERIV(input[1][j]);
}
// Update weights from hidden layer to output layer
for (int j=0; j < numOutput; j++) {
for (int k=0; k < numHidden; k++) {
outputWeights[j][k+1] = outputWeights[j][k+1] + learnRate * activation[1][k] * outputDelta[j];
}
outputWeights[j][0] = outputWeights[j][0] + learnRate * -1 * outputDelta[j]; // Bias
}
// Update weights from input layer to hidden layer
for (int j=0; j < numHidden; j++) {
for (int k=0; k < numInputs; k++) {
inputWeights[j][k+1] = inputWeights[j][k+1] + learnRate * activation[0][k] * hiddenDelta[j];
}
inputWeights[j][0] = inputWeights[j][0] + learnRate * -1 * hiddenDelta[j]; // Bias
}
// Update weight vector
weights[0] = inputWeights;
weights[1] = outputWeights;
}
}
}
// Predict on `data` with current network weights
void Network::test(Dataset &data) {
// Loop over samples
for (int i=0; i < data.size(); i++) {
std::vector<double> sampleFeatures;
std::vector<bool> sampleLabels;
// Load current sample and calculate activation
data.sample(i, sampleFeatures, sampleLabels);
getActivation(sampleFeatures);
// Loop over each class
for (int j=0; j < numOutput; j++) {
// Threshold activations to predict
data.classify(i, j, (activation[2][j] >= .5 ? true : false));
}
}
}
