ssRbm_test.cpp

#include <mlpack/core.hpp>

#include <mlpack/core/optimizers/rmsprop/rmsprop.hpp>
#include <mlpack/methods/ann/init_rules/gaussian_init.hpp>
#include <mlpack/methods/rbm/binary_layer.hpp>
#include <mlpack/methods/rbm/spike_slab_layer.hpp>
#include <mlpack/methods/rbm/rbm.hpp>
#include <mlpack/methods/rbm/binary_rbm.hpp>
#include <mlpack/methods/rbm/ssRBM.hpp>
#include <mlpack/methods/softmax_regression/softmax_regression.hpp>
#include <mlpack/core/optimizers/minibatch_sgd/minibatch_sgd.hpp>
#include <mlpack/core/optimizers/sgd/sgd.hpp>
#include <mlpack/core/optimizers/lbfgs/lbfgs.hpp>

using namespace mlpack;
using namespace mlpack::rbm;
using namespace mlpack::optimization;
using namespace mlpack::regression;

int main()
{
  // Hidden layer size 192
  int hiddenLayerSize = 100;
  arma::mat trainData;
  trainData.load("batch1patches.txt");
  trainData.resize(192, 800);

  std::cout << "batch rows = " << trainData.n_rows << std::endl;
  std::cout << "batch columns = " << trainData.n_cols << std::endl;
  // Intialise weight 10^-4
  GaussianInitialization gaussian(0, 1e-4);

  double radius = 0;
  double tempRadius = 0;
  for (size_t i = 0; i < trainData.n_cols; i++)
  {
    tempRadius = arma::norm(trainData.col(i));
    if (radius < tempRadius)
      radius = tempRadius;
  }
  // Set pool size
  size_t poolSize = 1;
  std::cout << "n rows = " << trainData.n_rows << std::endl;
  std::cout << "poolSize = " << poolSize << std::endl;
  BinaryLayer<> visible(64, 100, 1);
  BinaryLayer<> hidden(100, 64, 0);
  BinaryRBM binary_rbm(visible, hidden);
  SpikeSlabLayer<> spikeVisible(64, 100, 3, radius,
      1);
  SpikeSlabLayer<> spikeHidden(100, 64, 3, radius, 0);
  ssRBM ss_rbm(spikeVisible, spikeHidden);
  trainData = trainData.rows(0, 99);
  std::cout << arma::size(trainData) << std::endl;
  // ssRBM Policy
  RBM<GaussianInitialization, ssRBM> modelssRBM(trainData, gaussian, ss_rbm,
      2, true, true);
  MiniBatchSGD msgd(10, 0.06, trainData.n_cols * 20, 0, true);
  std::cout << "Here" << std::endl;
  modelssRBM.Reset();
  /*
  // Lambda bias = 0
  modelssRBM.Policy().VisibleLayer().LambdaBias().fill(0);
  // Spike Bias = -1
  modelssRBM.Policy().VisibleLayer().SpikeBias().fill(-1);
  // slab bias fixed at 1.5
  modelssRBM.Policy().VisibleLayer().SlabBias().fill(1.5);
  // Train the models
  std::cout << "Train" << std::endl;
  modelssRBM.Train(trainData, msgd);
  */
  return 0;
}