void linearRegressionExample()
{
// Load data into input/reference matrices
// divide into training set/validation set
auto inputData = loadInputData();
auto referenceData = loadReferenceData();
size_t inputCount = inputData.size()[0];
size_t inputSamples = inputData.size()[1];
size_t trainingSamples = 400;
//size_t validationSamples = inputSamples - trainingSamples;
auto trainingData = slice(inputData, {0, 0}, {inputCount, trainingSamples});
auto trainingReference = slice(referenceData, {0, 0}, {1, trainingSamples});
auto validationData = slice(inputData, {0, trainingSamples}, {inputCount, inputSamples});
auto validationReference = slice(referenceData, {0, trainingSamples}, {1, inputSamples});
setupParameters();
// Create the network
network::NeuralNetwork simpleNetwork;
simpleNetwork.addLayer(std::make_unique<network::FeedForwardLayer>(inputCount, 1, matrix::DoublePrecision()));
simpleNetwork.back()->setActivationFunction(network::ActivationFunctionFactory::create("NullActivationFunction"));
simpleNetwork.setCostFunction(network::CostFunctionFactory::create("SumOfSquaresCostFunction"));
simpleNetwork.initialize();
// Train it
simpleNetwork.train(trainingData, trainingReference);
// Evaluate it on test set
//auto trainingPredictions = simpleNetwork.runInputs(trainingData);
//compare(trainingPredictions, trainingReference);
// Evaluate it on validation set
auto predictions = simpleNetwork.runInputs(validationData);
compare(predictions, validationReference);
//for(size_t sample = 0; sample < validationSamples; ++ sample)
//{
// std::cout << sample << ", " << predictions(0, sample) << ", " << validationReference(0, sample) << "\n";
//}
}
void StrictMathTestCaseCos::setUp() {
loadReferenceData(TESTSRC "/StrictMathTestCase/cos.ref");
}
void StrictMathTestCaseATan::setUp() {
loadReferenceData(TESTSRC "/StrictMathTestCase/atan.ref");
}
