void plotLearningCurves(NNModel* model, DataParser* trainingSet, DataParser* cvSet, double lambda){
/* getting training set and cross validation set */
mat& xTrain = trainingSet->getExampleSet();
mat& yTrain = trainingSet->getLabelSet();
mat& xCross = cvSet->getExampleSet();
mat& yCross = cvSet->getLabelSet();
/* Initializing some usefull variables */
int m = xTrain.n_rows;
int numEvaluations = ((m-stepSize) / stepSize)+1;
/* Initializing data containers */
double* errorCV = new double[numEvaluations];
double* errorT = new double[numEvaluations];
double* foldSizes = new double[numEvaluations];
int currentFoldSize = stepSize;
mat predictions;
for (int i = 0; i < numEvaluations; i++)
{
/* Initializing example fold(size is dependent on iteration) */
mat xFold = xTrain.head_rows(currentFoldSize);
mat yFold = yTrain.head_rows(currentFoldSize);
/* Training model over current example fold */
NNBackPropagation * backPropagation
= new NNBackPropagation(model, xFold, yFold, lambda);
backPropagation->optimize(numTrainingIterations);
model = backPropagation->getUpdatedModel();
/* Getting error over current fold and cross validation set */
predictions = model->predict(xFold);
errorT[i] = model->getCostOver(predictions, yFold);
predictions = model->predict(xCross);
errorCV[i] = model->getCostOver(predictions, yCross);
foldSizes[i] = currentFoldSize;
/* Showing the results of iteration */
cout << " Cost for crossVal. set: " << errorCV[i] << endl;
cout << " Cost for training fold: " << errorT[i] << endl;
cout << " Training finished for fold of: " << currentFoldSize << endl << endl;
/* Changing variables */
currentFoldSize += stepSize;
model->randomlyInitialize();
delete backPropagation;
}
cout << " == Computation finished." << endl;
cout << " == Visualizing data." << endl;
/* ================== Plotting the results ================== */
ostringstream plotNameStream;
plotNameStream << "\"" << "Learning curves: lambda = " << lambda << "\"";
string plotName = plotNameStream.str();
Visualizer *visualizer = new Visualizer("");
visualizer->addSeries("\"Crossvalidation set\"", numEvaluations, foldSizes, errorCV);
visualizer->addSeries("\"Training set\"", numEvaluations, foldSizes, errorT);
visualizer->visualize("\"Learning curves\"", (char*)plotName.c_str(), "\"Fold size\"", "error");
}
