tiny_dnn 1.0.0
A header only, dependency-free deep learning framework in C++11
Loading...
Searching...
No Matches
partial_connected_layer.h
1/*
2 Copyright (c) 2013, Taiga Nomi
3 All rights reserved.
4
5 Redistribution and use in source and binary forms, with or without
6 modification, are permitted provided that the following conditions are met:
7 * Redistributions of source code must retain the above copyright
8 notice, this list of conditions and the following disclaimer.
9 * Redistributions in binary form must reproduce the above copyright
10 notice, this list of conditions and the following disclaimer in the
11 documentation and/or other materials provided with the distribution.
12 * Neither the name of the <organization> nor the
13 names of its contributors may be used to endorse or promote products
14 derived from this software without specific prior written permission.
15
16 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY
17 EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19 DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY
20 DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
22 LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
23 ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26*/
27#pragma once
28#include "tiny_dnn/util/util.h"
29#include "tiny_dnn/layers/layer.h"
30
31namespace tiny_dnn {
32
33template<typename Activation>
34class partial_connected_layer : public feedforward_layer<Activation> {
35public:
36 CNN_USE_LAYER_MEMBERS;
37
38 typedef std::vector<std::pair<serial_size_t, serial_size_t> > io_connections;
39 typedef std::vector<std::pair<serial_size_t, serial_size_t> > wi_connections;
40 typedef std::vector<std::pair<serial_size_t, serial_size_t> > wo_connections;
42
43 partial_connected_layer(serial_size_t in_dim,
44 serial_size_t out_dim,
45 size_t weight_dim,
46 size_t bias_dim,
48 : Base(std_input_order(bias_dim > 0)),
49 weight2io_(weight_dim),
50 out2wi_(out_dim),
51 in2wo_(in_dim),
52 bias2out_(bias_dim),
53 out2bias_(out_dim),
54 scale_factor_(scale_factor){}
55
56 size_t param_size() const {
57 size_t total_param = 0;
58 for (auto w : weight2io_)
59 if (w.size() > 0) total_param++;
60 for (auto b : bias2out_)
61 if (b.size() > 0) total_param++;
62 return total_param;
63 }
64
65 serial_size_t fan_in_size() const override {
66 return max_size(out2wi_);
67 }
68
69 serial_size_t fan_out_size() const override {
70 return max_size(in2wo_);
71 }
72
73 void connect_weight(serial_size_t input_index, serial_size_t output_index, serial_size_t weight_index) {
74 weight2io_[weight_index].emplace_back(input_index, output_index);
75 out2wi_[output_index].emplace_back(weight_index, input_index);
76 in2wo_[input_index].emplace_back(weight_index, output_index);
77 }
78
79 void connect_bias(serial_size_t bias_index, serial_size_t output_index) {
80 out2bias_[output_index] = bias_index;
81 bias2out_[bias_index].push_back(output_index);
82 }
83
84 void forward_propagation(const std::vector<tensor_t*>& in_data,
85 std::vector<tensor_t*>& out_data) override {
86 const tensor_t& in = *in_data[0];
87 const vec_t& W = (*in_data[1])[0];
88 const vec_t& b = (*in_data[2])[0];
89 tensor_t& a = *out_data[1];
90
91 // @todo revise the parallelism strategy
92 for (serial_size_t sample = 0, sample_count = static_cast<serial_size_t>(in.size()); sample < sample_count; ++sample) {
93 vec_t& a_sample = a[sample];
94
95 for_i(parallelize_, out2wi_.size(), [&](int i) {
96 const wi_connections& connections = out2wi_[i];
97
98 float_t& a_element = a_sample[i];
99
100 a_element = float_t(0);
101
102 for (auto connection : connections)// 13.1%
103 a_element += W[connection.first] * in[sample][connection.second]; // 3.2%
104
105 a_element *= scale_factor_;
106 a_element += b[out2bias_[i]];
107 });
108 }
109
110 this->forward_activation(*out_data[0], *out_data[1]);
111 }
112
113 void back_propagation(const std::vector<tensor_t*>& in_data,
114 const std::vector<tensor_t*>& out_data,
115 std::vector<tensor_t*>& out_grad,
116 std::vector<tensor_t*>& in_grad) override {
117 const tensor_t& prev_out = *in_data[0];
118 const vec_t& W = (*in_data[1])[0];
119 vec_t& dW = (*in_grad[1])[0];
120 vec_t& db = (*in_grad[2])[0];
121 tensor_t& prev_delta = *in_grad[0];
122 tensor_t& curr_delta = *out_grad[0];
123
124 this->backward_activation(*out_grad[0], *out_data[0], curr_delta);
125
126 // @todo revise the parallelism strategy
127 for (serial_size_t sample = 0, sample_count = static_cast<serial_size_t>(prev_out.size()); sample < sample_count; ++sample) {
128 for_(parallelize_, 0, in2wo_.size(), [&](const blocked_range& r) {
129 for (int i = r.begin(); i != r.end(); i++) {
130 const wo_connections& connections = in2wo_[i];
131 float_t delta = float_t(0);
132
133 for (auto connection : connections)
134 delta += W[connection.first] * curr_delta[sample][connection.second]; // 40.6%
135
136 prev_delta[sample][i] = delta * scale_factor_; // 2.1%
137 }
138 });
139
140 for_(parallelize_, 0, weight2io_.size(), [&](const blocked_range& r) {
141 for (int i = r.begin(); i < r.end(); i++) {
142 const io_connections& connections = weight2io_[i];
143 float_t diff = float_t(0);
144
145 for (auto connection : connections) // 11.9%
146 diff += prev_out[sample][connection.first] * curr_delta[sample][connection.second];
147
148 dW[i] += diff * scale_factor_;
149 }
150 });
151
152 for (size_t i = 0; i < bias2out_.size(); i++) {
153 const std::vector<serial_size_t>& outs = bias2out_[i];
154 float_t diff = float_t(0);
155
156 for (auto o : outs)
157 diff += curr_delta[sample][o];
158
159 db[i] += diff;
160 }
161 }
162 }
163
164protected:
165 std::vector<io_connections> weight2io_; // weight_id -> [(in_id, out_id)]
166 std::vector<wi_connections> out2wi_; // out_id -> [(weight_id, in_id)]
167 std::vector<wo_connections> in2wo_; // in_id -> [(weight_id, out_id)]
168 std::vector<std::vector<serial_size_t> > bias2out_;
169 std::vector<size_t> out2bias_;
170 float_t scale_factor_;
171};
172
173} // namespace tiny_dnn
single-input, single-output network with activation function
Definition feedforward_layer.h:37
Simple image utility class.
Definition image.h:94
bool parallelize_
Flag indicating whether the layer/node operations ara paralellized.
Definition layer.h:696
Definition partial_connected_layer.h:34
serial_size_t fan_in_size() const override
number of incoming connections for each output unit used only for weight/bias initialization methods ...
Definition partial_connected_layer.h:65
void back_propagation(const std::vector< tensor_t * > &in_data, const std::vector< tensor_t * > &out_data, std::vector< tensor_t * > &out_grad, std::vector< tensor_t * > &in_grad) override
return delta of previous layer (delta=\frac{dE}{da}, a=wx in fully-connected layer)
Definition partial_connected_layer.h:113
serial_size_t fan_out_size() const override
number of outgoing connections for each input unit used only for weight/bias initialization methods w...
Definition partial_connected_layer.h:69
void forward_propagation(const std::vector< tensor_t * > &in_data, std::vector< tensor_t * > &out_data) override
Definition partial_connected_layer.h:84
Definition parallel_for.h:70