mlpack
recurrent.hpp
Go to the documentation of this file.
1 
12 #ifndef MLPACK_METHODS_ANN_LAYER_RECURRENT_HPP
13 #define MLPACK_METHODS_ANN_LAYER_RECURRENT_HPP
14 
15 #include <mlpack/core.hpp>
16 
17 #include "../visitor/delete_visitor.hpp"
18 #include "../visitor/delta_visitor.hpp"
19 #include "../visitor/copy_visitor.hpp"
20 #include "../visitor/output_parameter_visitor.hpp"
21 #include "../visitor/input_shape_visitor.hpp"
22 
23 #include "layer_types.hpp"
24 #include "add_merge.hpp"
25 #include "sequential.hpp"
26 
27 namespace mlpack {
28 namespace ann {
29 
39 template <
40  typename InputDataType = arma::mat,
41  typename OutputDataType = arma::mat,
42  typename... CustomLayers
43 >
44 class Recurrent
45 {
46  public:
51  Recurrent();
52 
54  Recurrent(const Recurrent&);
55 
65  template<typename StartModuleType,
66  typename InputModuleType,
67  typename FeedbackModuleType,
68  typename TransferModuleType>
69  Recurrent(const StartModuleType& start,
70  const InputModuleType& input,
71  const FeedbackModuleType& feedback,
72  const TransferModuleType& transfer,
73  const size_t rho);
74 
82  template<typename eT>
83  void Forward(const arma::Mat<eT>& input, arma::Mat<eT>& output);
84 
94  template<typename eT>
95  void Backward(const arma::Mat<eT>& /* input */,
96  const arma::Mat<eT>& gy,
97  arma::Mat<eT>& g);
98 
99  /*
100  * Calculate the gradient using the output delta and the input activation.
101  *
102  * @param input The input parameter used for calculating the gradient.
103  * @param error The calculated error.
104  * @param gradient The calculated gradient.
105  */
106  template<typename eT>
107  void Gradient(const arma::Mat<eT>& input,
108  const arma::Mat<eT>& error,
109  arma::Mat<eT>& /* gradient */);
110 
112  std::vector<LayerTypes<CustomLayers...> >& Model() { return network; }
113 
115  bool Deterministic() const { return deterministic; }
117  bool& Deterministic() { return deterministic; }
118 
120  OutputDataType const& Parameters() const { return parameters; }
122  OutputDataType& Parameters() { return parameters; }
123 
125  OutputDataType const& OutputParameter() const { return outputParameter; }
127  OutputDataType& OutputParameter() { return outputParameter; }
128 
130  OutputDataType const& Delta() const { return delta; }
132  OutputDataType& Delta() { return delta; }
133 
135  OutputDataType const& Gradient() const { return gradient; }
137  OutputDataType& Gradient() { return gradient; }
138 
140  size_t const& Rho() const { return rho; }
141 
143  size_t InputShape() const;
144 
148  template<typename Archive>
149  void serialize(Archive& ar, const uint32_t /* version */);
150 
151  private:
153  DeleteVisitor deleteVisitor;
154 
156  CopyVisitor<CustomLayers...> copyVisitor;
157 
159  LayerTypes<CustomLayers...> startModule;
160 
162  LayerTypes<CustomLayers...> inputModule;
163 
165  LayerTypes<CustomLayers...> feedbackModule;
166 
168  LayerTypes<CustomLayers...> transferModule;
169 
171  size_t rho;
172 
174  size_t forwardStep;
175 
177  size_t backwardStep;
178 
180  size_t gradientStep;
181 
183  bool deterministic;
184 
187  bool ownsLayer;
188 
190  OutputDataType parameters;
191 
193  LayerTypes<CustomLayers...> initialModule;
194 
196  LayerTypes<CustomLayers...> recurrentModule;
197 
199  std::vector<LayerTypes<CustomLayers...> > network;
200 
202  LayerTypes<CustomLayers...> mergeModule;
203 
205  DeltaVisitor deltaVisitor;
206 
208  OutputParameterVisitor outputParameterVisitor;
209 
211  std::vector<arma::mat> feedbackOutputParameter;
212 
214  OutputDataType delta;
215 
217  OutputDataType gradient;
218 
220  OutputDataType outputParameter;
221 
223  arma::mat recurrentError;
224 }; // class Recurrent
225 
226 } // namespace ann
227 } // namespace mlpack
228 
229 // Include implementation.
230 #include "recurrent_impl.hpp"
231 
232 #endif
bool Deterministic() const
The value of the deterministic parameter.
Definition: recurrent.hpp:115
DeleteVisitor executes the destructor of the instantiated object.
Definition: delete_visitor.hpp:27
OutputDataType const & Gradient() const
Get the gradient.
Definition: recurrent.hpp:135
size_t InputShape() const
Get the shape of the input.
Definition: recurrent_impl.hpp:132
OutputDataType const & OutputParameter() const
Get the output parameter.
Definition: recurrent.hpp:125
Linear algebra utility functions, generally performed on matrices or vectors.
Definition: cv.hpp:1
This visitor is to support copy constructor for neural network module.
Definition: copy_visitor.hpp:26
bool & Deterministic()
Modify the value of the deterministic parameter.
Definition: recurrent.hpp:117
void Backward(const arma::Mat< eT > &, const arma::Mat< eT > &gy, arma::Mat< eT > &g)
Ordinary feed backward pass of a neural network, calculating the function f(x) by propagating x backw...
Definition: recurrent_impl.hpp:227
OutputDataType & Gradient()
Modify the gradient.
Definition: recurrent.hpp:137
Recurrent()
Default constructor—this will create a Recurrent object that can&#39;t be used, so be careful! Make sure...
Definition: recurrent_impl.hpp:29
OutputParameterVisitor exposes the output parameter of the given module.
Definition: output_parameter_visitor.hpp:27
void serialize(Archive &ar, const uint32_t)
Serialize the layer.
Definition: recurrent_impl.hpp:308
OutputDataType const & Delta() const
Get the delta.
Definition: recurrent.hpp:130
std::vector< LayerTypes< CustomLayers... > > & Model()
Get the model modules.
Definition: recurrent.hpp:112
OutputDataType const & Parameters() const
Get the parameters.
Definition: recurrent.hpp:120
Include all of the base components required to write mlpack methods, and the main mlpack Doxygen docu...
size_t const & Rho() const
Get the number of steps to backpropagate through time.
Definition: recurrent.hpp:140
OutputDataType & OutputParameter()
Modify the output parameter.
Definition: recurrent.hpp:127
DeltaVisitor exposes the delta parameter of the given module.
Definition: delta_visitor.hpp:27
OutputDataType & Delta()
Modify the delta.
Definition: recurrent.hpp:132
OutputDataType & Parameters()
Modify the parameters.
Definition: recurrent.hpp:122
void Forward(const arma::Mat< eT > &input, arma::Mat< eT > &output)
Ordinary feed forward pass of a neural network, evaluating the function f(x) by propagating the activ...
Definition: recurrent_impl.hpp:182