libcaffe-cpu-dev 1.0.0~rc4-1 / usr / include / caffe / sgd_solvers.hpp

This file is indexed.

/usr/include/caffe/sgd_solvers.hpp is in libcaffe-cpu-dev 1.0.0~rc4-1.

This file is owned by root:root, with mode 0o644.

The actual contents of the file can be viewed below. 

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
#ifndef CAFFE_SGD_SOLVERS_HPP_
#define CAFFE_SGD_SOLVERS_HPP_

#include <string>
#include <vector>

#include "caffe/solver.hpp"

namespace caffe {

/**
 * @brief Optimizes the parameters of a Net using
 *        stochastic gradient descent (SGD) with momentum.
 */
template <typename Dtype>
class SGDSolver : public Solver<Dtype> {
 public:
  explicit SGDSolver(const SolverParameter& param)
      : Solver<Dtype>(param) { PreSolve(); }
  explicit SGDSolver(const string& param_file)
      : Solver<Dtype>(param_file) { PreSolve(); }
  virtual inline const char* type() const { return "SGD"; }

  const vector<shared_ptr<Blob<Dtype> > >& history() { return history_; }

 protected:
  void PreSolve();
  Dtype GetLearningRate();
  virtual void ApplyUpdate();
  virtual void Normalize(int param_id);
  virtual void Regularize(int param_id);
  virtual void ComputeUpdateValue(int param_id, Dtype rate);
  virtual void ClipGradients();
  virtual void SnapshotSolverState(const string& model_filename);
  virtual void SnapshotSolverStateToBinaryProto(const string& model_filename);
  virtual void SnapshotSolverStateToHDF5(const string& model_filename);
  virtual void RestoreSolverStateFromHDF5(const string& state_file);
  virtual void RestoreSolverStateFromBinaryProto(const string& state_file);
  // history maintains the historical momentum data.
  // update maintains update related data and is not needed in snapshots.
  // temp maintains other information that might be needed in computation
  //   of gradients/updates and is not needed in snapshots
  vector<shared_ptr<Blob<Dtype> > > history_, update_, temp_;

  DISABLE_COPY_AND_ASSIGN(SGDSolver);
};

template <typename Dtype>
class NesterovSolver : public SGDSolver<Dtype> {
 public:
  explicit NesterovSolver(const SolverParameter& param)
      : SGDSolver<Dtype>(param) {}
  explicit NesterovSolver(const string& param_file)
      : SGDSolver<Dtype>(param_file) {}
  virtual inline const char* type() const { return "Nesterov"; }

 protected:
  virtual void ComputeUpdateValue(int param_id, Dtype rate);

  DISABLE_COPY_AND_ASSIGN(NesterovSolver);
};

template <typename Dtype>
class AdaGradSolver : public SGDSolver<Dtype> {
 public:
  explicit AdaGradSolver(const SolverParameter& param)
      : SGDSolver<Dtype>(param) { constructor_sanity_check(); }
  explicit AdaGradSolver(const string& param_file)
      : SGDSolver<Dtype>(param_file) { constructor_sanity_check(); }
  virtual inline const char* type() const { return "AdaGrad"; }

 protected:
  virtual void ComputeUpdateValue(int param_id, Dtype rate);
  void constructor_sanity_check() {
    CHECK_EQ(0, this->param_.momentum())
        << "Momentum cannot be used with AdaGrad.";
  }

  DISABLE_COPY_AND_ASSIGN(AdaGradSolver);
};


template <typename Dtype>
class RMSPropSolver : public SGDSolver<Dtype> {
 public:
  explicit RMSPropSolver(const SolverParameter& param)
      : SGDSolver<Dtype>(param) { constructor_sanity_check(); }
  explicit RMSPropSolver(const string& param_file)
      : SGDSolver<Dtype>(param_file) { constructor_sanity_check(); }
  virtual inline const char* type() const { return "RMSProp"; }

 protected:
  virtual void ComputeUpdateValue(int param_id, Dtype rate);
  void constructor_sanity_check() {
    CHECK_EQ(0, this->param_.momentum())
        << "Momentum cannot be used with RMSProp.";
    CHECK_GE(this->param_.rms_decay(), 0)
        << "rms_decay should lie between 0 and 1.";
    CHECK_LT(this->param_.rms_decay(), 1)
        << "rms_decay should lie between 0 and 1.";
  }

  DISABLE_COPY_AND_ASSIGN(RMSPropSolver);
};

template <typename Dtype>
class AdaDeltaSolver : public SGDSolver<Dtype> {
 public:
  explicit AdaDeltaSolver(const SolverParameter& param)
      : SGDSolver<Dtype>(param) { AdaDeltaPreSolve(); }
  explicit AdaDeltaSolver(const string& param_file)
      : SGDSolver<Dtype>(param_file) { AdaDeltaPreSolve(); }
  virtual inline const char* type() const { return "AdaDelta"; }

 protected:
  void AdaDeltaPreSolve();
  virtual void ComputeUpdateValue(int param_id, Dtype rate);

  DISABLE_COPY_AND_ASSIGN(AdaDeltaSolver);
};

/**
 * @brief AdamSolver, an algorithm for first-order gradient-based optimization
 *        of stochastic objective functions, based on adaptive estimates of
 *        lower-order moments. Described in [1].
 *
 * [1] D. P. Kingma and J. L. Ba, "ADAM: A Method for Stochastic Optimization."
 *     arXiv preprint arXiv:1412.6980v8 (2014).
 */
template <typename Dtype>
class AdamSolver : public SGDSolver<Dtype> {
 public:
  explicit AdamSolver(const SolverParameter& param)
      : SGDSolver<Dtype>(param) { AdamPreSolve();}
  explicit AdamSolver(const string& param_file)
      : SGDSolver<Dtype>(param_file) { AdamPreSolve(); }
  virtual inline const char* type() const { return "Adam"; }

 protected:
  void AdamPreSolve();
  virtual void ComputeUpdateValue(int param_id, Dtype rate);

  DISABLE_COPY_AND_ASSIGN(AdamSolver);
};

}  // namespace caffe

#endif  // CAFFE_SGD_SOLVERS_HPP_

APT Browse - Built by Thomas Orozco.