-
-
Notifications
You must be signed in to change notification settings - Fork 1k
/
ConvolutionalFeatureMap.h
200 lines (181 loc) · 6.38 KB
/
ConvolutionalFeatureMap.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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
/*
* 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 __CONVOLUTIONALFEATUREMAP_H__
#define __CONVOLUTIONALFEATUREMAP_H__
#include <shogun/lib/common.h>
namespace shogun
{
enum EConvMapActivationFunction
{
CMAF_IDENTITY = 0,
CMAF_LOGISTIC = 1,
CMAF_RECTIFIED_LINEAR = 2
};
template <class T> class SGVector;
template <class T> class SGMatrix;
class CDynamicObjectArray;
/** @brief Handles convolution and gradient calculation for a single feature
* map in a convolutional neural network
*/
class CConvolutionalFeatureMap
{
public:
/** Constuctor
*
* @param width Width of the input
*
* @param height Height of the input
*
* @param radius_x Radius of the convolution filter on the x (width) axis
*
* @param radius_y Radius of the convolution filter on the y (height) axis
*
* @param index Index of this feature map in its layer. This affects which
* part of the activations/activation_gradients matrix the map will store
* its outputs in.
*/
CConvolutionalFeatureMap(int32_t width, int32_t height,
int32_t radius_x, int32_t radius_y, int32_t index=0,
EConvMapActivationFunction function = CMAF_IDENTITY);
/** Computes the activations of the feature map
*
* @param parameters Vector of parameters for the map. length
* width*height+(2*radius_x+1)+(2*radius_y+1)
*
* @param layers The layers array that forms the network in which the map
* is being used
*
* @param input_indices Indices of the layers that are connected to the map
* as input
*
* @param activations Matrix in which the activations are to be stored
*
* @param buffer Matrix of the same size as activations. Used as a buffer
* during computations
*/
void compute_activations(SGVector<float64_t> parameters,
CDynamicObjectArray* layers,
SGVector<int32_t> input_indices,
SGMatrix<float64_t> activations,
SGMatrix<float64_t> buffer);
/** Computes the gradients with respect to the parameters and the inputs to
* the map
*
* @param parameters Vector of parameters for the map. length
* width*height+(2*radius_x+1)+(2*radius_y+1)
*
* @param activations Activations of the map
*
* @param activation_gradients Gradients of the error with respect to the
* map's activations
*
* @param layers The layers array that forms the network in which the map
* is being used
*
* @param input_indices Indices of the layers that are connected to the map
* as input
*
* @param activations Matrix to store the activations in
*
* @param parameters Vector in which the parameters gradients are to be
* stored
*/
void compute_gradients(SGVector<float64_t> parameters,
SGMatrix<float64_t> activations,
SGMatrix<float64_t> activation_gradients,
CDynamicObjectArray* layers,
SGVector<int32_t> input_indices,
SGVector<float64_t> parameter_gradients);
protected:
/** Perfoms convolution
*
* @param inputs Inputs matrix. Each column in the matrix is treated as an
* image in column major format
*
* @param weights Convolution filter
*
* @param outputs Output matrix
*
* @param flip If true the weights are flipped, performing cross-correlation
* instead of convolution
*
* @param reset_output true the output is reset to zero before performing
* convolution
*
* @param inputs_row_offset Index of the row at which the input image starts
*
* @param outputs_row_offset Index of the row at which the output image starts
*/
void convolve(SGMatrix<float64_t> inputs,
SGMatrix<float64_t> weights,
SGMatrix<float64_t> outputs,
bool flip,
bool reset_output,
int32_t inputs_row_offset,
int32_t outputs_row_offset);
/** Computes the gradients of the error with respect to the weights, for a
* particular input matrix
*
* @param inputs Inputs matrix
*
* @param local_gradients Gradients with respect the map's pre-activations
*
* @param weight_gradients Matrix to store the gradients in
*
* @param inputs_row_offset Offset for accessing the rows of the inputs
* matrix
*
* @param local_gradients_row_offset Offset for accessing the rows of the
* local gradients matrix
*/
void compute_weight_gradients(SGMatrix<float64_t> inputs,
SGMatrix<float64_t> local_gradients,
SGMatrix<float64_t> weight_gradients,
int32_t inputs_row_offset,
int32_t local_gradients_row_offset);
protected:
/** Width of the input */
int32_t m_width;
/** Height of the input */
int32_t m_height;
/** Radius of the convolution filter on the x (width) axis */
int32_t m_radius_x;
/** Radius of the convolution filter on the y (height) axis */
int32_t m_radius_y;
/** Index of this feature map in its layer. This affects which
* part of the activations/activation_gradients matrix that map will use
*/
int32_t m_index;
/** The map's activation function */
EConvMapActivationFunction m_activation_function;
};
}
#endif