main.cpp

#include "TrainingData.h"
#include "Net.h"
#define ETA 0.5 // net learning rate, [0.0..1.0]
#define trainingFile "trainingData.txt"

void showVectorVals(std::string label, std::vector<double> &v)
{
    std::cout << label << " ";
    for (unsigned i = 0; i < v.size(); ++i) {
        std::cout << v[i] << " ";
    }

    std::cout << std::endl;
}

double activationFunc(double net) {
  return 1.0/(1.0+exp(-net));
}

double activationFuncDerivative(double net) {
  return activationFunc(net)*(1-activationFunc(net));
}


int main()
{
    TrainingData trainData(trainingFile);
    // e.g., { 2, 4, 1 }
    std::vector<unsigned> topology;

    // get the structure of the net
    trainData.getTopology(topology);

    Net myNet(topology, ETA, &activationFunc, &activationFuncDerivative);

    std::vector<double> inputVals, targetVals, resultVals;
    int trainingIteration = 0;

    while (!trainData.isEof()) {
        ++trainingIteration;
        std::cout << std::endl << "Iteration " << trainingIteration ;

        // Get new input data and feed it forward:
        if (trainData.getNextInputs(inputVals) != topology[0]) {
            break;
        }
        showVectorVals(": Inputs:", inputVals);

	// feed Forward
        myNet.feedForward(inputVals);

        // Collect the net's actual output results:
        myNet.getResults(resultVals);
        showVectorVals("Outputs:", resultVals);

        // Train the net what the outputs should have been:
        trainData.getTargetOutputs(targetVals);
        showVectorVals("Targets:", targetVals);
        assert(targetVals.size() == topology.back());

        myNet.backProp(targetVals);

        // Report how well the training is working, average over recent samples:
        std::cout << "loss: " << myNet.getError() << std::endl;
        //if (trainingIteration==10000)
        //  break;

    }

    std::cout << std::endl << "Training Complete" << std::endl;
}