-
-
Notifications
You must be signed in to change notification settings - Fork 1k
/
NeuralLinearLayer.h
156 lines (141 loc) · 5.88 KB
/
NeuralLinearLayer.h
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
149
150
151
152
153
154
155
156
/*
* Copyright (c) 2014, Shogun Toolbox Foundation
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from this
* software without specific prior written permission.
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
* Written (W) 2014 Khaled Nasr
*/
#ifndef __NEURALLINEARLAYER_H__
#define __NEURALLINEARLAYER_H__
#include <shogun/lib/common.h>
#include <shogun/neuralnets/NeuralLayer.h>
namespace shogun
{
/** @brief Neural layer with linear neurons, with an identity activation
* function. can be used as a hidden layer or an output layer
*
* When used as an output layer, a squared error measure is used
*/
class CNeuralLinearLayer : public CNeuralLayer
{
public:
/** default constructor */
CNeuralLinearLayer();
/** Constuctor
*
* @param num_neurons Number of neurons in this layer
*/
CNeuralLinearLayer(int32_t num_neurons);
virtual ~CNeuralLinearLayer() {}
/** Gets the number of parameters (weights and biases) needed for this
* layer
*
* @return number of parameters (weights and biases) needed for this layer
*/
virtual int32_t get_num_parameters();
/** Initializes the layer's parameters. The layer should fill the given
* arrays with the initial value for its parameters
*
* @param parameters preallocated array of size get_num_parameters()
*
* @param parameter_regularizable preallocated array of size
* get_num_parameters(). This controls which of the layer's parameter are
* subject to regularization, i.e to turn off regularization for parameter
* i, set parameter_regularizable[i] = false. This is usally used to turn
* off regularization for bias parameters.
*
* @param sigma standard deviation of the gaussian used to random the
* parameters
*/
virtual void initialize_parameters(float64_t* parameters,
bool* parameter_regularizable,
float64_t sigma = 0.01f);
/** Computes the activations of the neurons in this layer, results should
* be stored in m_activations
*
* @param parameters pointer to the layer's parameters, array of size
* get_num_parameters()
*
* @param previous_layer_activations activations of the neurons in the
* previous layer, matrix of size previous_layer_num_neurons * batch_size
*/
virtual void compute_activations(float64_t* parameters,
float64_t* previous_layer_activations);
/** Computes the gradients that are relevent to this layer:
* - The gradients of the error with respect to the layer's parameters
* - The gradients of the error with respect to the layer's inputs
*
* The input gradients are stored in m_input_gradients
*
* @param parameters pointer to the layer's parameters, array of size
* get_num_parameters()
*
* @param is_output specifies if the layer is used as an output layer or a
* hidden layer
*
* @param p a matrix of size num_neurons*batch_size. If is_output is true,p
* is the desired values for the layer's activations, else it is the
* gradients of the error with respect to this layer's activations (the
* input gradients of the next layer).
*
* @param previous_layer_activations activations of the neurons in the
* previous layer, matrix of size previous_layer_num_neurons * batch_size
*
* @param parameter_gradients preallocated array of size
* get_num_parameters(), to be filled with the parameter gradients of this
* layer
*/
virtual void compute_gradients(float64_t* parameters,
bool is_output,
float64_t* p,
float64_t* previous_layer_activations,
float64_t* parameter_gradients);
/** Computes the error between the layer's current activations and the given
* target activations. Should only be used with output layers
*
* @param targets desired values for the layer's activations, matrix of size
* num_neurons*batch_size
*/
virtual float64_t computer_error(float64_t* targets);
/** Computes the gradients of the error with respect to this layer's
* activations. Results are stored in m_local_gradients.
*
* This is used by compute_gradients() and can be overriden to implement
* layers with different activation functions
*
* @param is_output specifies if the layer is used as an output layer or a
* hidden layer
*
* @param p a matrix of size num_neurons*batch_size. If is_output is true,p
* is the desired values for the layer's activations, else it is the
* gradients of the error with respect to this layer's activations (the
* input gradients of the next layer).
*/
virtual void compute_local_gradients(bool is_output, float64_t* p);
virtual const char* get_name() const { return "NeuralLinearLayer"; }
};
}
#endif