ClassifierEngine::ResultVector ClassifierEngine::runOnBatch(const Bundle& input)
{
auto network = getAggregateNetwork();
network->setIsTraining(false);
auto bundle = network->runInputs(input);
auto& result = bundle["outputActivations"].get<matrix::MatrixVector>().front();
auto labels = convertActivationsToLabels(std::move(result), *getModel());
ClassifierEngine::ResultVector results;
if(_shouldUseLabeledData)
{
bundle = network->getCost(input);
auto cost = bundle["cost"].get<double>();
results = compareWithReference(cost * labels.size(), getIteration(), labels,
getReferenceLabels(bundle, *getModel()));
}
else
{
results = recordLabels(labels);
}
restoreAggregateNetwork();
return results;
}
NeuralNetwork::Matrix NeuralNetwork::runInputs(const Matrix& m) const
{
//util::log("NeuralNetwork") << "Running forward propagation on matrix (" << m.rows()
// << " rows, " << m.columns() << " columns).\n";
auto temp = convertToBlockSparseForLayerInput(front(), m);
return runInputs(temp).toMatrix();
}
float NeuralNetwork::computeAccuracy(const BlockSparseMatrix& input,
const BlockSparseMatrix& reference) const
{
assert(input.rows() == reference.rows());
assert(reference.columns() == getOutputCount());
auto result = runInputs(input);
float threshold = 0.5f;
auto resultActivations = result.greaterThanOrEqual(threshold);
auto referenceActivations = reference.greaterThanOrEqual(threshold);
util::log("NeuralNetwork") << "Result activations " << resultActivations.toString();
util::log("NeuralNetwork") << "Reference activations " << referenceActivations.toString();
auto matchingActivations = resultActivations.equals(referenceActivations);
float matches = matchingActivations.reduceSum();
return matches / result.size();
}
