tiny_dnn  1.0.0
A header only, dependency-free deep learning framework in C++11
linear_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 <algorithm>
30 
31 
32 namespace tiny_dnn {
33 
37 template<typename Activation>
38 class linear_layer : public feedforward_layer<Activation> {
39 public:
40  CNN_USE_LAYER_MEMBERS;
41 
43 
49  explicit linear_layer(serial_size_t dim, float_t scale = float_t(1), float_t bias = float_t(0))
50  : Base({vector_type::data}),
51  dim_(dim), scale_(scale), bias_(bias) {}
52 
53  std::vector<shape3d> in_shape() const override {
54  return {shape3d(dim_, 1, 1) };
55  }
56 
57  std::vector<shape3d> out_shape() const override {
58  return{ shape3d(dim_, 1, 1), shape3d(dim_, 1, 1) };
59  }
60 
61  std::string layer_type() const override { return "linear"; }
62 
63  void forward_propagation(const std::vector<tensor_t*>& in_data,
64  std::vector<tensor_t*>& out_data) override {
65  const tensor_t& in = *in_data[0];
66  tensor_t& out = *out_data[0];
67  tensor_t& a = *out_data[1];
68 
69  // do nothing
70  CNN_UNREFERENCED_PARAMETER(out);
71 
72  // @todo revise the parallelism strategy
73  for_i(parallelize_, dim_, [&](int i) {
74  for (serial_size_t sample = 0, sample_count = static_cast<serial_size_t>(in.size()); sample < sample_count; ++sample)
75  a[sample][i] = scale_ * in[sample][i] + bias_;
76  });
77  this->forward_activation(*out_data[0], *out_data[1]);
78  }
79 
80  void back_propagation(const std::vector<tensor_t*>& in_data,
81  const std::vector<tensor_t*>& out_data,
82  std::vector<tensor_t*>& out_grad,
83  std::vector<tensor_t*>& in_grad) override {
84  tensor_t& prev_delta = *in_grad[0];
85  tensor_t& curr_delta = *out_grad[1];
86 
87  CNN_UNREFERENCED_PARAMETER(in_data);
88 
89  this->backward_activation(*out_grad[0], *out_data[0], curr_delta);
90 
91  // @todo revise parallelism strategy
92  for (serial_size_t sample = 0; sample < static_cast<serial_size_t>(prev_delta.size()); ++sample) {
93  for_i(parallelize_, dim_, [&](int i) {
94  prev_delta[sample][i] = curr_delta[sample][i] * scale_;
95  });
96  }
97  }
98 
99  template <class Archive>
100  static void load_and_construct(Archive & ar, cereal::construct<linear_layer> & construct) {
101  serial_size_t dim;
102  float_t scale, bias;
103 
104  ar(cereal::make_nvp("in_size", dim), cereal::make_nvp("scale", scale), cereal::make_nvp("bias", bias));
105 
106  construct(dim, scale, bias);
107  }
108 
109  template <class Archive>
110  void serialize(Archive & ar) {
111  layer::serialize_prolog(ar);
112  ar(cereal::make_nvp("in_size", dim_), cereal::make_nvp("scale", scale_), cereal::make_nvp("bias", bias_));
113  }
114 
115 protected:
116  serial_size_t dim_;
117  float_t scale_, bias_;
118 };
119 
120 } // namespace tiny_dnn
single-input, single-output network with activation function
Definition: feedforward_layer.h:37
bool parallelize_
Flag indicating whether the layer/node operations ara paralellized.
Definition: layer.h:696
element-wise operation: f(x) = h(scale*x+bias)
Definition: linear_layer.h:38
std::vector< shape3d > in_shape() const override
array of input shapes (width x height x depth)
Definition: linear_layer.h:53
linear_layer(serial_size_t dim, float_t scale=float_t(1), float_t bias=float_t(0))
Definition: linear_layer.h:49
std::vector< shape3d > out_shape() const override
array of output shapes (width x height x depth)
Definition: linear_layer.h:57
std::string layer_type() const override
name of layer, should be unique for each concrete class
Definition: linear_layer.h:61
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: linear_layer.h:80
void forward_propagation(const std::vector< tensor_t * > &in_data, std::vector< tensor_t * > &out_data) override
Definition: linear_layer.h:63