void MapTokens(const std::vector<std::string>& tokens,
size_t& row,
arma::Mat<eT>& matrix,
MapType& maps,
std::vector<Datatype>& types)
{
// MissingPolicy allows double type matrix only, because it uses NaN.
static_assert(std::is_same<eT, double>::value, "You must use double type "
" matrix in order to apply MissingPolicy");
std::stringstream token;
for (size_t i = 0; i != tokens.size(); ++i)
{
token.str(tokens[i]);
token>>matrix.at(row, i);
// if the token is not number, map it.
// or if token is a number, but is included in the missingSet, map it.
if (token.fail() || missingSet.find(tokens[i]) != std::end(missingSet))
{
const eT val = static_cast<eT>(this->MapString(tokens[i], row, maps,
types));
matrix.at(row, i) = val;
}
token.clear();
}
}
static typename std::enable_if<
std::is_same<Border, ValidConvolution>::value, void>::type
Convolution(const arma::Mat<eT>& input,
const arma::Mat<eT>& filter,
arma::Mat<eT>& output,
const size_t dW = 1,
const size_t dH = 1)
{
output = arma::zeros<arma::Mat<eT> >((input.n_rows - filter.n_rows + 1) /
dW, (input.n_cols - filter.n_cols + 1) / dH);
// It seems to be about 3.5 times faster to use pointers instead of
// filter(ki, kj) * input(leftInput + ki, topInput + kj) and output(i, j).
eT* outputPtr = output.memptr();
for (size_t j = 0; j < output.n_cols; ++j)
{
for (size_t i = 0; i < output.n_rows; ++i, outputPtr++)
{
const eT* kernelPtr = filter.memptr();
for (size_t kj = 0; kj < filter.n_cols; ++kj)
{
const eT* inputPtr = input.colptr(kj + j * dW) + i * dH;
for (size_t ki = 0; ki < filter.n_rows; ++ki, ++kernelPtr, ++inputPtr)
*outputPtr += *kernelPtr * (*inputPtr);
}
}
}
}
void
convert_matrix(const arma::Mat<Tfrom>& densemat,
csc_matrix<Tto, Idxto>& cscmat) {
Idxto m = densemat.num_rows();
Idxto n = densemat.num_cols();
arma::Col<Idxto> new_col_offsets(n + 1, 0);
arma::Col<Idxto> new_row_indices;
arma::Col<Tto> new_values;
// Compute column sizes, and fill vectors of row indices and values.
for (Idxto j(0); j < n; ++j) {
for (Idxto i(0); i < m; ++i) {
if (densemat(i,j) != 0) {
++new_col_offsets[j];
new_row_indices.push_back(i);
new_values.push_back(densemat(i,j));
}
}
}
// Compute offsets.
for (Idxto i(0); i < n; ++i)
new_col_offsets[i+1] += new_col_offsets[i];
for (Idxto i(n); i > 0; --i)
new_col_offsets[i] = new_col_offsets[i-1];
new_col_offsets[0] = 0;
// Copy data over
cscmat.reset_nocopy(m, n, new_col_offsets, new_row_indices, new_values);
}
void MainWindow::loadSeries() {
QString seriesFile = __fileManager.openFile(0, this, "Open Series File...", "*.mat");
if (seriesFile != "") {
cout << "Loading start ..." << flush;
_data.load(seriesFile.toStdString().c_str());
cout << " done. " << endl;
_dataT = arma::trans(_data);
cout << _data.n_cols << "," << _data.n_rows << endl;
_imageBuffer.set_size(_mapid.n_rows, _mapid.n_cols);
for (int i = 0; i < _mapid.n_rows; i++) {
for (int j = 0; j < _mapid.n_cols; j++) {
if (_mapid(i,j) != _mapid(i,j)) {
_imageBuffer(i,j) = NAN;
} else {
_imageBuffer(i,j) = 0;
}
}
}
ui.slider->setMaximum(_data.n_cols - 1);
ui.minValue->setValue(_data.min());
ui.maxValue->setValue(_data.max());
_mapItem->setImage(_imageBuffer.memptr(), _imageBuffer.n_rows, _imageBuffer.n_cols);
_mapItem->setRange(ui.minValue->value(), ui.maxValue->value());
_mapItem->setInteraction(this);
selectValue(0);
}
}
inline ParallelKinematicMachine3PUPS::ParallelKinematicMachine3PUPS() noexcept {
setMinimalActiveJointActuations({0.39, 0.39, 0.39});
setMaximalActiveJointActuations({0.95, 0.95, 0.95});
setEndEffectorJointPositions({
-0.025561381023353, 0.086293776138137, 0.12,
0.087513292835791, -0.021010082747031, 0.12,
-0.061951911812438, -0.065283693391106, 0.12});
setRedundantJointStartPositions({
-0.463708870031622, 0.417029254828353, -0.346410161513775,
0.593012363818459, 0.193069033993384, -0.346410161513775,
-0.129303493786837, -0.610098288821738, -0.346410161513775});
setRedundantJointEndPositions({
-0.247202519085512, 0.292029254828353, 0.086602540378444,
0.376506012872349, 0.068069033993384, 0.086602540378444,
-0.129303493786837, -0.360098288821738, 0.086602540378444});
redundantJointStartToEndPositions_ = redundantJointEndPositions_ - redundantJointStartPositions_;
redundantJointIndicies_ = arma::find(arma::any(redundantJointStartToEndPositions_));
redundantJointAngles_.set_size(3, redundantJointIndicies_.n_elem);
for (std::size_t n = 0; n < redundantJointIndicies_.n_elem; ++n) {
const double& redundantJointXAngle = std::atan2(redundantJointStartToEndPositions_(1, n), redundantJointStartToEndPositions_(0, n));
const double& redundantJointYAngle = std::atan2(redundantJointStartToEndPositions_(2, n), redundantJointStartToEndPositions_(1, n));
redundantJointAngles_.col(n) = arma::Col<double>::fixed<3>({std::cos(redundantJointXAngle) * std::cos(redundantJointYAngle), std::sin(redundantJointXAngle) * std::cos(redundantJointYAngle), std::sin(redundantJointYAngle)});
}
}
inline void gpu_train_batch(FeedForward_Network<activation, error>& network,
arma::Mat<float> inputs, arma::Mat<float> targets, int batch_size, float learning_rate = 0.8f, float momentum = 0.8f) {
network.resize_activation(batch_size);
Raw_FeedForward_Network<activation, error> raw_net = convert_to_raw(network);
Raw_FeedForward_Network<activation, error> * d_network = network_to_gpu(raw_net);
int batches_in_train = targets.n_rows/batch_size - 1;
for (int i = 0; i < batches_in_train; ++i) {
arma::Mat<float> input_slice = inputs.rows(i*batch_size, (i+1) * batch_size - 1);
Raw_Matrix raw_input = to_raw(input_slice);
Raw_Matrix * d_input = matrix_to_gpu(raw_input);
int num_trials = input_slice.n_rows;
calculate_activation(num_trials, network.layer_sizes, d_network, d_input);
//TODO make this memory shared as to not realloc
free_gpu_matrix(d_input);
arma::Mat<float> targets_slice = targets.rows(i*batch_size, (i+1) * batch_size - 1);
Raw_Matrix raw_targets = to_raw(targets_slice);
Raw_Matrix * d_targets = matrix_to_gpu(raw_targets);
backprop(num_trials, network.layer_sizes, d_network, d_targets, learning_rate, momentum);
free_gpu_matrix(d_targets);
}
network_to_cpu_free(d_network, raw_net);
update_from_raw(network, raw_net);
}
MainWindow::MainWindow(QWidget* parent) {
ui.setupUi(this);
ui.graphicsView->setScene(&_scene);
_mapid.load("./MapID.mat");
cout << _mapid.n_cols << "," << _mapid.n_rows << endl;
connect(ui.slider, SIGNAL(valueChanged(int)), SLOT(selectValue(int)));
connect(ui.clearGraphButton, SIGNAL(pressed()), SLOT(clearGraphs()));
connect(ui.loadSeries, SIGNAL(pressed()), SLOT(loadSeries()));
_mapItem = new QFloatImageItem();
_mapItem->setImage(_imageBuffer.memptr(), _imageBuffer.n_rows, _imageBuffer.n_cols);
_mapItem->setRange(ui.minValue->value(), ui.maxValue->value());
_mapItem->setInteraction(this);
_scene.addItem(_mapItem);
_traceItem = _scene.addPath(_tracePath);
_traceItem->setPen(QPen(Qt::black));
QTransform matrix;
matrix.scale(5, 5);
ui.graphicsView->setTransform(matrix);
ui.customPlot->setInteraction(QCustomPlot::iSelectLegend, true);
ui.customPlot->setInteraction(QCustomPlot::iSelectPlottables, true);
connect(ui.customPlot, SIGNAL(selectionChangedByUser()), SLOT(graphSelected()));
_plotMarker = new QGraphicsEllipseItem(_mapItem);
_plotMarker->setVisible(false);
}
/**
* Impute function searches through the input looking for mappedValue and
* remove the whole row or column. The result is overwritten to the input.
*
* @param input Matrix that contains mappedValue.
* @param mappedValue Value that the user wants to get rid of.
* @param dimension Index of the dimension of the mappedValue.
* @param columnMajor State of whether the input matrix is columnMajor or not.
*/
void Impute(arma::Mat<T>& input,
const T& mappedValue,
const size_t dimension,
const bool columnMajor = true)
{
std::vector<arma::uword> colsToKeep;
if (columnMajor)
{
for (size_t i = 0; i < input.n_cols; ++i)
{
if (!(input(dimension, i) == mappedValue ||
std::isnan(input(dimension, i))))
{
colsToKeep.push_back(i);
}
}
input = input.cols(arma::uvec(colsToKeep));
}
else
{
for (size_t i = 0; i < input.n_rows; ++i)
{
if (!(input(i, dimension) == mappedValue ||
std::isnan(input(i, dimension))))
{
colsToKeep.push_back(i);
}
}
input = input.rows(arma::uvec(colsToKeep));
}
}
void
save(OutputArchive& ar, const arma::Mat<T>& mat, const unsigned int version) {
size_t size = mat.size();
ar & mat.n_cols;
ar & mat.n_rows;
ar & make_array(mat.memptr(), size);
}
arma::Col<double> compute_column_rms(const arma::Mat<double>& data) {
const long n_cols = data.n_cols;
arma::Col<double> rms(n_cols);
for (long i=0; i<n_cols; ++i) {
const double dot = arma::dot(data.col(i), data.col(i));
rms(i) = std::sqrt(dot / (data.col(i).n_rows-1));
}
return std::move(rms);
}
void ConcreteGridView<DuneGridView>::getRawElementDataImpl(
arma::Mat<CoordinateType>& vertices,
arma::Mat<int>& elementCorners,
arma::Mat<char>& auxData) const
{
typedef typename DuneGridView::Grid DuneGrid;
typedef typename DuneGridView::IndexSet DuneIndexSet;
const int dimGrid = DuneGrid::dimension;
const int dimWorld = DuneGrid::dimensionworld;
const int codimVertex = dimGrid;
const int codimElement = 0;
typedef Dune::LeafMultipleCodimMultipleGeomTypeMapper<DuneGrid,
Dune::MCMGElementLayout> DuneElementMapper;
typedef typename DuneGridView::template Codim<codimVertex>::Iterator
DuneVertexIterator;
typedef typename DuneGridView::template Codim<codimElement>::Iterator
DuneElementIterator;
typedef typename DuneGridView::template Codim<codimVertex>::Geometry
DuneVertexGeometry;
typedef typename DuneGridView::template Codim<codimElement>::Geometry
DuneElementGeometry;
typedef typename DuneGrid::ctype ctype;
const DuneIndexSet& indexSet = m_dune_gv.indexSet();
vertices.set_size(dimWorld, indexSet.size(codimVertex));
for (DuneVertexIterator it = m_dune_gv.template begin<codimVertex>();
it != m_dune_gv.template end<codimVertex>(); ++it)
{
size_t index = indexSet.index(*it);
const DuneVertexGeometry& geom = it->geometry();
Dune::FieldVector<ctype, dimWorld> vertex = geom.corner(0);
for (int i = 0; i < dimWorld; ++i)
vertices(i, index) = vertex[i];
}
const int MAX_CORNER_COUNT = dimWorld == 2 ? 2 : 4;
DuneElementMapper elementMapper(m_dune_gv.grid());
elementCorners.set_size(MAX_CORNER_COUNT, elementMapper.size());
for (DuneElementIterator it = m_dune_gv.template begin<codimElement>();
it != m_dune_gv.template end<codimElement>(); ++it)
{
size_t index = elementMapper.map(*it);
const Dune::GenericReferenceElement<ctype, dimGrid>& refElement =
Dune::GenericReferenceElements<ctype, dimGrid>::general(it->type());
const int cornerCount = refElement.size(codimVertex);
assert(cornerCount <= MAX_CORNER_COUNT);
for (int i = 0; i < cornerCount; ++i)
elementCorners(i, index) = indexSet.subIndex(*it, i, codimVertex);
for (int i = cornerCount; i < MAX_CORNER_COUNT; ++i)
elementCorners(i, index) = -1;
}
auxData.set_size(0, elementCorners.n_cols);
}
// Predict the rating for a group of user/item combinations.
void CF::Predict(const arma::Mat<size_t>& combinations,
arma::vec& predictions) const
{
// First, for nearest neighbor search, stretch the H matrix.
arma::mat l = arma::chol(w.t() * w);
arma::mat stretchedH = l * h; // Due to the Armadillo API, l is L^T.
// Now, we must determine those query indices we need to find the nearest
// neighbors for. This is easiest if we just sort the combinations matrix.
arma::Mat<size_t> sortedCombinations(combinations.n_rows,
combinations.n_cols);
arma::uvec ordering = arma::sort_index(combinations.row(0).t());
for (size_t i = 0; i < ordering.n_elem; ++i)
sortedCombinations.col(i) = combinations.col(ordering[i]);
// Now, we have to get the list of unique users we will be searching for.
arma::Col<size_t> users = arma::unique(combinations.row(0).t());
// Assemble our query matrix from the stretchedH matrix.
arma::mat queries(stretchedH.n_rows, users.n_elem);
for (size_t i = 0; i < queries.n_cols; ++i)
queries.col(i) = stretchedH.col(users[i]);
// Now calculate the neighborhood of these users.
neighbor::KNN a(stretchedH);
arma::mat distances;
arma::Mat<size_t> neighborhood;
a.Search(queries, numUsersForSimilarity, neighborhood, distances);
// Now that we have the neighborhoods we need, calculate the predictions.
predictions.set_size(combinations.n_cols);
size_t user = 0; // Cumulative user count, because we are doing it in order.
for (size_t i = 0; i < sortedCombinations.n_cols; ++i)
{
// Could this be made faster by calculating dot products for multiple items
// at once?
double rating = 0.0;
// Map the combination's user to the user ID used for kNN.
while (users[user] < sortedCombinations(0, i))
++user;
for (size_t j = 0; j < neighborhood.n_rows; ++j)
rating += arma::as_scalar(w.row(sortedCombinations(1, i)) *
h.col(neighborhood(j, user)));
rating /= neighborhood.n_rows;
predictions(ordering[i]) = rating;
}
}
void enforce_positive_sign_by_column(arma::Mat<double>& data) {
for (long i=0; i<long(data.n_cols); ++i) {
const double max = arma::max(data.col(i));
const double min = arma::min(data.col(i));
bool change_sign = false;
if (std::abs(max)>=std::abs(min)) {
if (max<0) change_sign = true;
} else {
if (min<0) change_sign = true;
}
if (change_sign) data.col(i) *= -1;
}
}
void calculateJacobian(const arma::Mat<std::complex<double> >& myOffsets,
arma::Mat<double>& myJacobian,
arma::Col<std::complex<double> >& myTargetsCalculated,
arma::Col<std::complex<double> >& myCurrentGuess,
void myCalculateDependentVariables(const arma::Mat<std::complex<double> >&, const arma::Col<std::complex<double> >&, arma::Col<std::complex<double> >&))
{
//Calculate a temporary, unperturbed target evaluation, such as is needed for solving for the updated guess
//formula
arma::Col<std::complex<double> > unperturbedTargetsCalculated(NUMDIMENSIONS);
unperturbedTargetsCalculated.fill(0.0);
myCalculateDependentVariables(myOffsets, myCurrentGuess, unperturbedTargetsCalculated);
std::complex<double> oldGuessValue(0.0, 0.0);
//Each iteration fills a column in the Jacobian
//The Jacobian takes this form:
//
// dF0/dx0 dF0/dx1
// dF1/dx0 dF1/dx1
//
for(int j = 0; j< NUMDIMENSIONS; j++)
{
//Store old element value, perturb the current value
oldGuessValue = myCurrentGuess[j];
myCurrentGuess[j] += std::complex<double>(0.0, PROBEDISTANCE);
//Evaluate functions for perturbed guess
myCalculateDependentVariables(myOffsets, myCurrentGuess, myTargetsCalculated);
//The column of the Jacobian that goes with the independent variable we perturbed
//can be determined using the finite-difference formula
//The arma::Col allows this to be expressed as a single vector operation
//note slice works as: std::slice(start_index, number_of_elements_to_access, index_interval_between_selections)
//std::cout << "Jacobian column " << j << " with:" << std::endl;
//std::cout << "myTargetsCalculated" << std::endl;
//std::cout << myTargetsCalculated << std::endl;
//std::cout << "unperturbedTargetsCalculated" << std::endl;
//std::cout << unperturbedTargetsCalculated << std::endl;
myJacobian.col(j) = arma::imag(myTargetsCalculated);
myJacobian.col(j) *= pow(PROBEDISTANCE, -1.0);
//std::cout << "The jacobian: " << std::endl;
//std::cout << myJacobian << std::endl;
myCurrentGuess[j] = oldGuessValue;
}
//Reset to unperturbed, so we dont waste a function evaluation
myTargetsCalculated = unperturbedTargetsCalculated;
}
arma::Col<double> compute_column_means(const arma::Mat<double>& data) {
const long n_cols = data.n_cols;
arma::Col<double> means(n_cols);
for (long i=0; i<n_cols; ++i)
means(i) = arma::mean(data.col(i));
return std::move(means);
}
void Arma_mat_to_cv_mat(const arma::Mat<T>& arma_mat_in, cv::Mat_<T>& cv_mat_out)
{
cv::transpose(cv::Mat_<T>(static_cast<int>(arma_mat_in.n_cols),
static_cast<int>(arma_mat_in.n_rows),
const_cast<T*>(arma_mat_in.memptr())),
cv_mat_out);
};
void FastMKS<KernelType, TreeType>::Search(TreeType* queryTree,
const size_t k,
arma::Mat<size_t>& indices,
arma::mat& kernels)
{
// If either naive mode or single mode is specified, this must fail.
if (naive || singleMode)
{
throw std::invalid_argument("can't call Search() with a query tree when "
"single mode or naive search is enabled");
}
// No remapping will be necessary because we are using the cover tree.
indices.set_size(k, queryTree->Dataset().n_cols);
kernels.set_size(k, queryTree->Dataset().n_cols);
kernels.fill(-DBL_MAX);
Timer::Start("computing_products");
typedef FastMKSRules<KernelType, TreeType> RuleType;
RuleType rules(referenceSet, queryTree->Dataset(), indices, kernels,
metric.Kernel());
typename TreeType::template DualTreeTraverser<RuleType> traverser(rules);
traverser.Traverse(*queryTree, *referenceTree);
Log::Info << rules.BaseCases() << " base cases." << std::endl;
Log::Info << rules.Scores() << " scores." << std::endl;
Timer::Stop("computing_products");
}
void Backward(const arma::Cube<eT>& input,
const arma::Mat<eT>& gy,
arma::Cube<eT>& g)
{
// Generate a cube using the backpropagated error matrix.
arma::Cube<eT> mappedError = arma::zeros<arma::cube>(input.n_rows,
input.n_cols, input.n_slices);
for (size_t s = 0, j = 0; s < mappedError.n_slices; s+= gy.n_cols, j++)
{
for (size_t i = 0; i < gy.n_cols; i++)
{
arma::Col<eT> temp = gy.col(i).subvec(
j * input.n_rows * input.n_cols,
(j + 1) * input.n_rows * input.n_cols - 1);
mappedError.slice(s + i) = arma::Mat<eT>(temp.memptr(),
input.n_rows, input.n_cols);
}
}
arma::Cube<eT> derivative;
Deriv(input, derivative);
g = mappedError % derivative;
}
inline force_inline
double LSHSearch<SortPolicy>::BaseCase(arma::mat& distances,
arma::Mat<size_t>& neighbors,
const size_t queryIndex,
const size_t referenceIndex)
{
// If the datasets are the same, then this search is only using one dataset
// and we should not return identical points.
if ((&querySet == &referenceSet) && (queryIndex == referenceIndex))
return 0.0;
const double distance = metric::EuclideanDistance::Evaluate(
querySet.unsafe_col(queryIndex), referenceSet.unsafe_col(referenceIndex));
// If this distance is better than any of the current candidates, the
// SortDistance() function will give us the position to insert it into.
arma::vec queryDist = distances.unsafe_col(queryIndex);
arma::Col<size_t> queryIndices = neighbors.unsafe_col(queryIndex);
size_t insertPosition = SortPolicy::SortDistance(queryDist, queryIndices,
distance);
// SortDistance() returns (size_t() - 1) if we shouldn't add it.
if (insertPosition != (size_t() - 1))
InsertNeighbor(distances, neighbors, queryIndex, insertPosition,
referenceIndex, distance);
return distance;
}
void FeedBackward(const arma::Cube<eT>& inputActivation,
const arma::Mat<eT>& error,
arma::Cube<eT>& delta)
{
delta = delta % mask * scale;
// Generate a cube from the error matrix.
arma::Cube<eT> mappedError = arma::zeros<arma::cube>(inputActivation.n_rows,
inputActivation.n_cols, inputActivation.n_slices);
for (size_t s = 0, j = 0; s < mappedError.n_slices; s+= error.n_cols, j++)
{
for (size_t i = 0; i < error.n_cols; i++)
{
arma::Col<eT> temp = error.col(i).subvec(
j * inputActivation.n_rows * inputActivation.n_cols,
(j + 1) * inputActivation.n_rows * inputActivation.n_cols - 1);
mappedError.slice(s + i) = arma::Mat<eT>(temp.memptr(),
inputActivation.n_rows, inputActivation.n_cols);
}
}
delta = mappedError;
}
inline Raw_Matrix to_raw(arma::Mat<float> & mat) {
Raw_Matrix matrix;
matrix.n_rows = mat.n_rows;
matrix.n_cols= mat.n_cols;
matrix.data = mat.memptr();
return matrix;
}
static typename std::enable_if<
std::is_same<Border, FullConvolution>::value, void>::type
Convolution(const arma::Mat<eT>& input,
const arma::Mat<eT>& filter,
arma::Mat<eT>& output)
{
// In case of the full convolution outputRows and outputCols doesn't
// represent the true output size when the padLastDim parameter is set,
// instead it's the working size.
const size_t outputRows = input.n_rows + 2 * (filter.n_rows - 1);
size_t outputCols = input.n_cols + 2 * (filter.n_cols - 1);
if (padLastDim)
outputCols++;
// Pad filter and input to the working output shape.
arma::Mat<eT> inputPadded = arma::zeros<arma::Mat<eT> >(outputRows,
outputCols);
inputPadded.submat(filter.n_rows - 1, filter.n_cols - 1,
filter.n_rows - 1 + input.n_rows - 1,
filter.n_cols - 1 + input.n_cols - 1) = input;
arma::Mat<eT> filterPadded = filter;
filterPadded.resize(outputRows, outputCols);
// Perform FFT and IFFT
output = arma::real(ifft2(arma::fft2(inputPadded) % arma::fft2(
filterPadded)));
// Extract the region of interest. We don't need to handle the padLastDim
// parameter in a special way we just cut it out from the output matrix.
output = output.submat(filter.n_rows - 1, filter.n_cols - 1,
2 * (filter.n_rows - 1) + input.n_rows - 1,
2 * (filter.n_cols - 1) + input.n_cols - 1);
}
void Forward(const arma::Mat<eT>& input, arma::Mat<eT>& output)
{
arma::mat maxInput = arma::repmat(arma::max(input), input.n_rows, 1);
output = (maxInput - input);
// Approximation of the hyperbolic tangent. The acuracy however is
// about 0.00001 lower as using tanh. Credits go to Leon Bottou.
output.transform( [](double x)
{
//! Fast approximation of exp(-x) for x positive.
static constexpr double A0 = 1.0;
static constexpr double A1 = 0.125;
static constexpr double A2 = 0.0078125;
static constexpr double A3 = 0.00032552083;
static constexpr double A4 = 1.0172526e-5;
if (x < 13.0)
{
double y = A0 + x * (A1 + x * (A2 + x * (A3 + x * A4)));
y *= y;
y *= y;
y *= y;
y = 1 / y;
return y;
}
return 0.0;
} );
output = input - (maxInput + std::log(arma::accu(output)));
}
std::vector<double> extract_column_vector(const arma::Mat<double>& data, long index) {
if (index<0 || index >= long(data.n_cols))
throw std::range_error(join("Index out of range: ", index));
const long n_rows = data.n_rows;
const double* memptr = data.colptr(index);
std::vector<double> result(memptr, memptr + n_rows);
return std::move(result);
}
arma::Row<T> column_apply(const arma::Mat<T>& matrix, Functor functor) {
arma::Row<T> result(matrix.n_cols);
std::size_t index = 0;
std::generate(result.begin(), result.end(), [&matrix, &index, &functor] {
return functor(matrix.col(index++));
});
return result;
}
/**
* Given a Reber string, return a Reber string with all reachable next symbols.
*
* @param transitions The Reber transistion matrix.
* @param reber The Reber string used to generate all reachable next symbols.
* @param nextReber All reachable next symbols.
*/
void GenerateNextReber(const arma::Mat<char>& transitions,
const std::string& reber, std::string& nextReber)
{
size_t idx = 0;
for (size_t grammer = 1; grammer < reber.length(); grammer++)
{
const int grammerIdx = arma::as_scalar(arma::find(
transitions.row(idx) == reber[grammer], 1, "first"));
idx = arma::as_scalar(transitions.submat(idx, grammerIdx + 2, idx,
grammerIdx + 2)) - '0';
}
nextReber = arma::as_scalar(transitions.submat(idx, 0, idx, 0));
nextReber += arma::as_scalar(transitions.submat(idx, 1, idx, 1));
}
/**
* Generate a random Reber grammar.
*
* For more information, see the following thesis.
*
* @code
* @misc{Gers2001,
* author = {Felix Gers},
* title = {Long Short-Term Memory in Recurrent Neural Networks},
* year = {2001}
* }
* @endcode
*
* @param transitions Reber grammar transition matrix.
* @param reber The generated Reber grammar string.
*/
void GenerateReber(const arma::Mat<char>& transitions, std::string& reber)
{
size_t idx = 0;
reber = "B";
do
{
const int grammerIdx = rand() % 2;
reber += arma::as_scalar(transitions.submat(idx, grammerIdx, idx,
grammerIdx));
idx = arma::as_scalar(transitions.submat(idx, grammerIdx + 2, idx,
grammerIdx + 2)) - '0';
} while (idx != 0);
reber = "BPTVVE";
}
//This function is specific to a single problem
void calculateDependentVariables(const arma::Mat<std::complex<double> >& myOffsets,
const arma::Col<std::complex<double> >& myCurrentGuess,
arma::Col<std::complex<double> >& targetsCalculated)
{
//Evaluate a dependent variable for each iteration
//The arma::Col allows this to be expressed as a vector operation
for(int i = 0; i < NUMDIMENSIONS; i++)
{
targetsCalculated[i] = arma::sum(pow(myCurrentGuess.subvec(0,1) - myOffsets.row(i).subvec(0,1).t(),2.0));
targetsCalculated[i] = targetsCalculated[i] + myCurrentGuess[2]*pow(-1.0, i) - myOffsets.row(i)[2];
//std::cout << targetsCalculated[i] << std::endl;
}
//std::cout << "model evaluated *************************" << std::endl;
//std::cout << targetsCalculated << std::endl;
//std::cout << myOffsets << std::endl;
}
std::vector<double> extract_row_vector(const arma::Mat<double>& data, long index) {
if (index<0 || index >= long(data.n_rows))
throw std::range_error(join("Index out of range: ", index));
const arma::Row<double> row(data.row(index));
const double* memptr = row.memptr();
std::vector<double> result(memptr, memptr + row.n_elem);
return std::move(result);
}
void normalize_by_column(arma::Mat<double>& data, const arma::Col<double>& rms) {
if (data.n_cols != rms.n_elem)
throw std::range_error("Number of elements of rms is not equal to the number of columns of data");
for (long i=0; i<long(data.n_cols); ++i) {
if (rms(i)==0)
throw std::runtime_error("At least one of the entries of rms equals to zero");
data.col(i) *= 1./rms(i);
}
}