adamax_update.hpp

Go to the documentation of this file.

#ifndef MLPACK_CORE_OPTIMIZERS_ADAM_ADAMAX_UPDATE_HPP
#define MLPACK_CORE_OPTIMIZERS_ADAM_ADAMAX_UPDATE_HPP

#include <mlpack/prereqs.hpp>

namespace mlpack {
namespace optimization {

class AdaMaxUpdate
{
 public:
  AdaMaxUpdate(const double epsilon = 1e-8,
               const double beta1 = 0.9,
               const double beta2 = 0.999) :
    epsilon(epsilon),
    beta1(beta1),
    beta2(beta2),
    iteration(0)
  {
    // Nothing to do.
  }

  void Initialize(const size_t rows, const size_t cols)
  {
    m = arma::zeros<arma::mat>(rows, cols);
    u = arma::zeros<arma::mat>(rows, cols);
  }

  void Update(arma::mat& iterate,
              const double stepSize,
              const arma::mat& gradient)
  {
    // Increment the iteration counter variable.
    ++iteration;

    // And update the iterate.
    m *= beta1;
    m += (1 - beta1) * gradient;

    // Update the exponentially weighted infinity norm.
    u *= beta2;
    u = arma::max(u, arma::abs(gradient));

    const double biasCorrection1 = 1.0 - std::pow(beta1, iteration);

    if (biasCorrection1 != 0)
      iterate -= (stepSize / biasCorrection1 * m / (u + epsilon));
  }

  double Epsilon() const { return epsilon; }
  double& Epsilon() { return epsilon; }

  double Beta1() const { return beta1; }
  double& Beta1() { return beta1; }

  double Beta2() const { return beta2; }
  double& Beta2() { return beta2; }

 private:
  // The epsilon value used to initialise the squared gradient parameter.
  double epsilon;

  // The smoothing parameter.
  double beta1;

  // The second moment coefficient.
  double beta2;

  // The exponential moving average of gradient values.
  arma::mat m;

  // The exponentially weighted infinity norm.
  arma::mat u;

  // The number of iterations.
  double iteration;
};

} // namespace optimization
} // namespace mlpack

#endif