tiny_dnn  1.0.0
A header only, dependency-free deep learning framework in C++11
power_layer.h
1 /*
2  Copyright (c) 2016, 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 #include <cmath>
31 
32 namespace tiny_dnn {
33 
34 
38 class power_layer : public layer {
39 public:
40  typedef layer Base;
41 
47  power_layer(const shape3d& in_shape, float_t factor, float_t scale=1.0f)
48  : layer({ vector_type::data }, { vector_type::data }),
49  in_shape_(in_shape), factor_(factor), scale_(scale) {
50  }
51 
57  power_layer(const layer& prev_layer, float_t factor, float_t scale=1.0f)
58  : layer({ vector_type::data }, { vector_type::data }),
59  in_shape_(prev_layer.out_shape()[0]), factor_(factor), scale_(scale) {
60  }
61 
62  std::string layer_type() const override {
63  return "power";
64  }
65 
66  std::vector<shape3d> in_shape() const override {
67  return {in_shape_};
68  }
69 
70  std::vector<shape3d> out_shape() const override {
71  return {in_shape_};
72  }
73 
74  void forward_propagation(const std::vector<tensor_t*>& in_data,
75  std::vector<tensor_t*>& out_data) override {
76  const tensor_t& x = *in_data[0];
77  tensor_t& y = *out_data[0];
78 
79  for (serial_size_t i = 0; i < x.size(); i++) {
80  std::transform(x[i].begin(), x[i].end(), y[i].begin(), [=](float_t x) {
81  return scale_*std::pow(x, factor_);
82  });
83  }
84  }
85 
86  void back_propagation(const std::vector<tensor_t*>& in_data,
87  const std::vector<tensor_t*>& out_data,
88  std::vector<tensor_t*>& out_grad,
89  std::vector<tensor_t*>& in_grad) override {
90  tensor_t& dx = *in_grad[0];
91  const tensor_t& dy = *out_grad[0];
92  const tensor_t& x = *in_data[0];
93  const tensor_t& y = *out_data[0];
94 
95  for (serial_size_t i = 0; i < x.size(); i++) {
96  for (serial_size_t j = 0; j < x[i].size(); j++) {
97  // f(x) = (scale*x)^factor
98  // ->
99  // dx = dy * df(x)
100  // = dy * scale * factor * (scale * x)^(factor - 1)
101  // = dy * scale * factor * (scale * x)^factor * (scale * x)^(-1)
102  // = dy * factor * y / x
103  if (std::abs(x[i][j]) > 1e-10) {
104  dx[i][j] = dy[i][j] * factor_ * y[i][j] / x[i][j];
105  }
106  else {
107  dx[i][j] = dy[i][j] * scale_ * factor_ * std::pow(x[i][j], factor_ - 1.0f);
108  }
109  }
110  }
111  }
112 
113  template <class Archive>
114  static void load_and_construct(Archive & ar, cereal::construct<power_layer> & construct) {
116  float_t factor;
117  float_t scale(1.0f);
118 
119  ar(cereal::make_nvp("in_size", in_shape), cereal::make_nvp("factor", factor), cereal::make_nvp("scale", scale));
120  construct(in_shape, factor, scale);
121  }
122 
123  template <class Archive>
124  void serialize(Archive & ar) {
125  layer::serialize_prolog(ar);
126  ar(cereal::make_nvp("in_size", in_shape_), cereal::make_nvp("factor", factor_), cereal::make_nvp("scale", scale_));
127  }
128 
129  float_t factor() const {
130  return factor_;
131  }
132 
133  float_t scale() const {
134  return scale_;
135  }
136 private:
137 
138  shape3d in_shape_;
139  float_t factor_;
140  float_t scale_;
141 };
142 
143 } // namespace tiny_dnn
base class of all kind of NN layers
Definition: layer.h:62
element-wise pow: y = scale*x^factor
Definition: power_layer.h:38
void forward_propagation(const std::vector< tensor_t * > &in_data, std::vector< tensor_t * > &out_data) override
Definition: power_layer.h:74
std::vector< shape3d > in_shape() const override
array of input shapes (width x height x depth)
Definition: power_layer.h:66
power_layer(const layer &prev_layer, float_t factor, float_t scale=1.0f)
Definition: power_layer.h:57
std::string layer_type() const override
name of layer, should be unique for each concrete class
Definition: power_layer.h:62
std::vector< shape3d > out_shape() const override
array of output shapes (width x height x depth)
Definition: power_layer.h:70
power_layer(const shape3d &in_shape, float_t factor, float_t scale=1.0f)
Definition: power_layer.h:47
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: power_layer.h:86