Merge pull request BVLC#3116 from ronghanghu/solver-refactor

Solver Refactor: Separate files and Change Solver's Type to String

docs/tutorial/solver.md

@@ -8,12 +8,12 @@ The responsibilities of learning are divided between the Solver for overseeing t
 
 The Caffe solvers are:
 
-- Stochastic Gradient Descent (`SGD`), 
-- AdaDelta (`ADADELTA`),
-- Adaptive Gradient (`ADAGRAD`),
-- Adam (`ADAM`),
-- Nesterov's Accelerated Gradient (`NESTEROV`) and
-- RMSprop (`RMSPROP`)
+- Stochastic Gradient Descent (`type: "SGD"`),
+- AdaDelta (`type: "AdaDelta"`),
+- Adaptive Gradient (`type: "AdaGrad"`),
+- Adam (`type: "Adam"`),
+- Nesterov's Accelerated Gradient (`type: "Nesterov"`) and
+- RMSprop (`type: "RMSProp"`)
 
 The solver
 
@@ -51,7 +51,7 @@ The parameter update $$\Delta W$$ is formed by the solver from the error gradien
 
 ### SGD
 
-**Stochastic gradient descent** (`solver_type: SGD`) updates the weights $$ W $$ by a linear combination of the negative gradient $$ \nabla L(W) $$ and the previous weight update $$ V_t $$.
+**Stochastic gradient descent** (`type: "SGD"`) updates the weights $$ W $$ by a linear combination of the negative gradient $$ \nabla L(W) $$ and the previous weight update $$ V_t $$.
 The **learning rate** $$ \alpha $$ is the weight of the negative gradient.
 The **momentum** $$ \mu $$ is the weight of the previous update.
 
@@ -113,7 +113,7 @@ If learning diverges (e.g., you start to see very large or `NaN` or `inf` loss v
 
 ### AdaDelta
 
-The **AdaDelta** (`solver_type: ADADELTA`) method (M. Zeiler [1]) is a "robust learning rate method". It is a gradient-based optimization method (like SGD). The update formulas are
+The **AdaDelta** (`type: "AdaDelta"`) method (M. Zeiler [1]) is a "robust learning rate method". It is a gradient-based optimization method (like SGD). The update formulas are
 
 $$
 \begin{align}
@@ -125,7 +125,7 @@ E[g^2]_t &= \delta{E[g^2]_{t-1} } + (1-\delta)g_{t}^2
 \end{align}
 $$
 
-and 
+and
 
 $$
 (W_{t+1})_i =
@@ -139,7 +139,7 @@ $$
 
 ### AdaGrad
 
-The **adaptive gradient** (`solver_type: ADAGRAD`) method (Duchi et al. [1]) is a gradient-based optimization method (like SGD) that attempts to "find needles in haystacks in the form of very predictive but rarely seen features," in Duchi et al.'s words.
+The **adaptive gradient** (`type: "AdaGrad"`) method (Duchi et al. [1]) is a gradient-based optimization method (like SGD) that attempts to "find needles in haystacks in the form of very predictive but rarely seen features," in Duchi et al.'s words.
 Given the update information from all previous iterations $$ \left( \nabla L(W) \right)_{t'} $$ for $$ t' \in \{1, 2, ..., t\} $$,
 the update formulas proposed by [1] are as follows, specified for each component $$i$$ of the weights $$W$$:
 
@@ -159,7 +159,7 @@ Note that in practice, for weights $$ W \in \mathcal{R}^d $$, AdaGrad implementa
 
 ### Adam
 
-The **Adam** (`solver_type: ADAM`), proposed in Kingma et al. [1], is a gradient-based optimization method (like SGD). This includes an "adaptive moment estimation" ($$m_t, v_t$$) and can be regarded as a generalization of AdaGrad. The update formulas are
+The **Adam** (`type: "Adam"`), proposed in Kingma et al. [1], is a gradient-based optimization method (like SGD). This includes an "adaptive moment estimation" ($$m_t, v_t$$) and can be regarded as a generalization of AdaGrad. The update formulas are
 
 $$
 (m_t)_i = \beta_1 (m_{t-1})_i + (1-\beta_1)(\nabla L(W_t))_i,\\
@@ -181,7 +181,7 @@ Kingma et al. [1] proposed to use $$\beta_1 = 0.9, \beta_2 = 0.999, \varepsilon
 
 ### NAG
 
-**Nesterov's accelerated gradient** (`solver_type: NESTEROV`) was proposed by Nesterov [1] as an "optimal" method of convex optimization, achieving a convergence rate of $$ \mathcal{O}(1/t^2) $$ rather than the $$ \mathcal{O}(1/t) $$.
+**Nesterov's accelerated gradient** (`type: "Nesterov"`) was proposed by Nesterov [1] as an "optimal" method of convex optimization, achieving a convergence rate of $$ \mathcal{O}(1/t^2) $$ rather than the $$ \mathcal{O}(1/t) $$.
 Though the required assumptions to achieve the $$ \mathcal{O}(1/t^2) $$ convergence typically will not hold for deep networks trained with Caffe (e.g., due to non-smoothness and non-convexity), in practice NAG can be a very effective method for optimizing certain types of deep learning architectures, as demonstrated for deep MNIST autoencoders by Sutskever et al. [2].
 
 The weight update formulas look very similar to the SGD updates given above:
@@ -206,10 +206,10 @@ What distinguishes the method from SGD is the weight setting $$ W $$ on which we
 
 ### RMSprop
 
-The **RMSprop** (`solver_type: RMSPROP`), suggested by Tieleman in a Coursera course lecture, is a gradient-based optimization method (like SGD). The update formulas are
+The **RMSprop** (`type: "RMSProp"`), suggested by Tieleman in a Coursera course lecture, is a gradient-based optimization method (like SGD). The update formulas are
 
 $$
-(v_t)_i = 
+(v_t)_i =
 \begin{cases}
 (v_{t-1})_i + \delta, &(\nabla L(W_t))_i(\nabla L(W_{t-1}))_i > 0\\
 (v_{t-1})_i \cdot (1-\delta), & \text{else}

examples/mnist/lenet_adadelta_solver.prototxt

@@ -20,5 +20,5 @@ snapshot: 5000
 snapshot_prefix: "examples/mnist/lenet_adadelta"
 # solver mode: CPU or GPU
 solver_mode: GPU
-solver_type: ADADELTA
+type: "AdaDelta"
 delta: 1e-6

examples/mnist/lenet_solver_adam.prototxt

@@ -22,5 +22,5 @@ max_iter: 10000
 snapshot: 5000
 snapshot_prefix: "examples/mnist/lenet"
 # solver mode: CPU or GPU
-solver_type: ADAM
+type: "Adam"
 solver_mode: GPU

examples/mnist/lenet_solver_rmsprop.prototxt

@@ -23,5 +23,5 @@ snapshot: 5000
 snapshot_prefix: "examples/mnist/lenet_rmsprop"
 # solver mode: CPU or GPU
 solver_mode: GPU
-solver_type: RMSPROP
+type: "RMSProp"
 rms_decay: 0.98

examples/mnist/mnist_autoencoder_solver_adadelta.prototxt

@@ -16,4 +16,4 @@ snapshot: 10000
 snapshot_prefix: "examples/mnist/mnist_autoencoder_adadelta_train"
 # solver mode: CPU or GPU
 solver_mode: GPU
-solver_type: ADADELTA
+type: "AdaDelta"

examples/mnist/mnist_autoencoder_solver_adagrad.prototxt

@@ -14,4 +14,4 @@ snapshot: 10000
 snapshot_prefix: "examples/mnist/mnist_autoencoder_adagrad_train"
 # solver mode: CPU or GPU
 solver_mode: GPU
-solver_type: ADAGRAD
+type: "AdaGrad"

examples/mnist/mnist_autoencoder_solver_nesterov.prototxt

@@ -17,4 +17,4 @@ snapshot_prefix: "examples/mnist/mnist_autoencoder_nesterov_train"
 momentum: 0.95
 # solver mode: CPU or GPU
 solver_mode: GPU
-solver_type: NESTEROV
+type: "Nesterov"

include/caffe/caffe.hpp

@@ -13,8 +13,10 @@
 #include "caffe/parallel.hpp"
 #include "caffe/proto/caffe.pb.h"
 #include "caffe/solver.hpp"
+#include "caffe/solver_factory.hpp"
 #include "caffe/util/benchmark.hpp"
 #include "caffe/util/io.hpp"
+#include "caffe/util/upgrade_proto.hpp"
 #include "caffe/vision_layers.hpp"
 
 #endif  // CAFFE_CAFFE_HPP_

include/caffe/sgd_solvers.hpp

@@ -0,0 +1,148 @@
+#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_