void SparseCoding<DictionaryInitializer>::Encode(const size_t maxIterations,
const double objTolerance,
const double newtonTolerance)
{
Timer::Start("sparse_coding");
double lastObjVal = DBL_MAX;
// Take the initial coding step, which has to happen before entering the main
// optimization loop.
Log::Info << "Initial Coding Step." << std::endl;
OptimizeCode();
arma::uvec adjacencies = find(codes);
Log::Info << " Sparsity level: " << 100.0 * ((double) (adjacencies.n_elem))
/ ((double) (atoms * data.n_cols)) << "%." << std::endl;
Log::Info << " Objective value: " << Objective() << "." << std::endl;
for (size_t t = 1; t != maxIterations; ++t)
{
// Print current iteration, and maximum number of iterations (if it isn't
// 0).
Log::Info << "Iteration " << t;
if (maxIterations != 0)
Log::Info << " of " << maxIterations;
Log::Info << "." << std::endl;
// First step: optimize the dictionary.
Log::Info << "Performing dictionary step... " << std::endl;
OptimizeDictionary(adjacencies, newtonTolerance);
Log::Info << " Objective value: " << Objective() << "." << std::endl;
// Second step: perform the coding.
Log::Info << "Performing coding step..." << std::endl;
OptimizeCode();
// Get the indices of all the nonzero elements in the codes.
adjacencies = find(codes);
Log::Info << " Sparsity level: " << 100.0 * ((double) (adjacencies.n_elem))
/ ((double) (atoms * data.n_cols)) << "%." << std::endl;
// Find the new objective value and improvement so we can check for
// convergence.
double curObjVal = Objective();
double improvement = lastObjVal - curObjVal;
Log::Info << " Objective value: " << curObjVal << " (improvement "
<< std::scientific << improvement << ")." << std::endl;
// Have we converged?
if (improvement < objTolerance)
{
Log::Info << "Converged within tolerance " << objTolerance << ".\n";
break;
}
lastObjVal = curObjVal;
}
Timer::Stop("sparse_coding");
}
void LocalCoordinateCoding::Train(
const arma::mat& data,
const DictionaryInitializer& initializer)
{
Timer::Start("local_coordinate_coding");
// Initialize the dictionary.
initializer.Initialize(data, atoms, dictionary);
double lastObjVal = DBL_MAX;
// Take the initial coding step, which has to happen before entering the main
// loop.
Log::Info << "Initial Coding Step." << std::endl;
arma::mat codes;
Encode(data, codes);
arma::uvec adjacencies = find(codes);
Log::Info << " Sparsity level: " << 100.0 * ((double)(adjacencies.n_elem)) /
((double)(atoms * data.n_cols)) << "%.\n";
Log::Info << " Objective value: " << Objective(data, codes, adjacencies)
<< "." << std::endl;
for (size_t t = 1; t != maxIterations; t++)
{
Log::Info << "Iteration " << t << " of " << maxIterations << "."
<< std::endl;
// First step: optimize the dictionary.
Log::Info << "Performing dictionary step..." << std::endl;
OptimizeDictionary(data, codes, adjacencies);
double dsObjVal = Objective(data, codes, adjacencies);
Log::Info << " Objective value: " << dsObjVal << "." << std::endl;
// Second step: perform the coding.
Log::Info << "Performing coding step..." << std::endl;
Encode(data, codes);
adjacencies = find(codes);
Log::Info << " Sparsity level: " << 100.0 * ((double) (adjacencies.n_elem))
/ ((double)(atoms * data.n_cols)) << "%.\n";
// Terminate if the objective increased in the coding step.
double curObjVal = Objective(data, codes, adjacencies);
if (curObjVal > dsObjVal)
{
Log::Warn << "Objective increased in coding step! Terminating."
<< std::endl;
break;
}
// Find the new objective value and improvement so we can check for
// convergence.
double improvement = lastObjVal - curObjVal;
Log::Info << "Objective value: " << curObjVal << " (improvement "
<< std::scientific << improvement << ")." << std::endl;
if (improvement < tolerance)
{
Log::Info << "Converged within tolerance " << tolerance << ".\n";
break;
}
lastObjVal = curObjVal;
}
Timer::Stop("local_coordinate_coding");
}
