tiny_dnn  1.0.0
A header only, dependency-free deep learning framework in C++11
quantized_fully_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/layers/layer.h"
29 #include "tiny_dnn/util/product.h"
30 
31 namespace tiny_dnn {
32 
36 template<typename Activation>
38 public:
40  CNN_USE_LAYER_MEMBERS;
41 
47  quantized_fully_connected_layer(serial_size_t in_dim,
48  serial_size_t out_dim,
49  bool has_bias = true,
50  backend_t backend_type = core::backend_t::internal)
51  : Base(std_input_order(has_bias)) {
52  set_params(in_dim, out_dim, has_bias);
53  init_backend(backend_type);
54  }
55 
56  // move constructor
58  : Base(std::move(other))
59  , params_(std::move(other.params_)) {
60  init_backend(core::backend_t::internal);
61  }
62 
63  serial_size_t fan_in_size() const override {
64  return params_.in_size_;
65  }
66 
67  serial_size_t fan_out_size() const override {
68  return params_.out_size_;
69  }
70 
71  std::vector<index3d<serial_size_t>> in_shape() const override {
72  if (params_.has_bias_) {
73  return { index3d<serial_size_t>(params_.in_size_, 1, 1),
74  index3d<serial_size_t>(params_.in_size_,
75  params_.out_size_, 1),
76  index3d<serial_size_t>(params_.out_size_, 1, 1) };
77  } else {
78  return { index3d<serial_size_t>(params_.in_size_, 1, 1),
79  index3d<serial_size_t>(params_.in_size_,
80  params_.out_size_, 1) };
81  }
82  }
83 
84  std::vector<index3d<serial_size_t>> out_shape() const override {
85  return { index3d<serial_size_t>(params_.out_size_, 1, 1),
86  index3d<serial_size_t>(params_.out_size_, 1, 1) };
87  }
88 
89  void forward_propagation(const std::vector<tensor_t*>& in_data,
90  std::vector<tensor_t*>& out_data) override {
91  if (in_data.size() == 2 || in_data.size() == 3) {
92  Base::backend_->fully_q(in_data, out_data);
93 
94  // activations
95  this->forward_activation(*out_data[0], *out_data[1]);
96  } else if (in_data.size() == 4 || in_data.size() == 6) {
97  Base::backend_->fully_eq(in_data, out_data);
98  }
99  }
100 
101  void back_propagation(const std::vector<tensor_t*>& in_data,
102  const std::vector<tensor_t*>& out_data,
103  std::vector<tensor_t*>& out_grad,
104  std::vector<tensor_t*>& in_grad) override {
105  Base::backend_->fully_q(in_data, out_data, out_grad, in_grad);
106  }
107 
108  std::string layer_type() const override { return "q_fully-connected"; }
109 
110 protected:
111  fully_params params_;
112 
113  void set_params(const serial_size_t in_size,
114  const serial_size_t out_size,
115  bool has_bias) {
116  params_.in_size_ = in_size;
117  params_.out_size_ = out_size;
118  params_.has_bias_ = has_bias;
119  }
120 
121  void init_backend(backend_t backend_type) {
122  std::shared_ptr<core::backend> backend = nullptr;
123 
124  // allocate new backend
125  if (backend_type == backend_t::internal) {
126  backend = std::make_shared<core::tiny_backend>(&params_,
127  [this](const tensor_t& p_delta,
128  const tensor_t& out, tensor_t& c_delta) {
129  return Base::backward_activation(p_delta, out, c_delta);
130  });
131  } else if (backend_type == backend_t::nnpack) {
132  backend = std::make_shared<core::nnp_backend>(&params_);
133  } else if (backend_type == backend_t::libdnn) {
134  backend = std::make_shared<core::dnn_backend>();
135  } else {
136  throw nn_error("Not supported backend type.");
137  }
138 
139  if (backend) {
140  Base::set_backend(backend);
141  Base::set_backend_type(backend_type);
142  Base::backend_->set_layer(this);
143  } else {
144  throw nn_error("Could not allocate the backend.");
145  }
146  }
147 };
148 
149 } // namespace tiny_dnn
single-input, single-output network with activation function
Definition: feedforward_layer.h:37
std::shared_ptr< core::backend > backend_
The backend instance (deprecated)
Definition: layer.h:708
serial_size_t out_size() const
!
Definition: layer.h:181
serial_size_t in_size() const
!
Definition: layer.h:176
std::shared_ptr< core::backend > backend()
number of incoming edges in this layer
Definition: layer.h:143
compute fully-connected(matmul) operation
Definition: quantized_fully_connected_layer.h:37
std::string layer_type() const override
name of layer, should be unique for each concrete class
Definition: quantized_fully_connected_layer.h:108
serial_size_t fan_in_size() const override
number of incoming connections for each output unit used only for weight/bias initialization methods ...
Definition: quantized_fully_connected_layer.h:63
quantized_fully_connected_layer(serial_size_t in_dim, serial_size_t out_dim, bool has_bias=true, backend_t backend_type=core::backend_t::internal)
Definition: quantized_fully_connected_layer.h:47
std::vector< index3d< serial_size_t > > out_shape() const override
array of output shapes (width x height x depth)
Definition: quantized_fully_connected_layer.h:84
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: quantized_fully_connected_layer.h:67
std::vector< index3d< serial_size_t > > in_shape() const override
array of input shapes (width x height x depth)
Definition: quantized_fully_connected_layer.h:71
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: quantized_fully_connected_layer.h:101
void forward_propagation(const std::vector< tensor_t * > &in_data, std::vector< tensor_t * > &out_data) override
Definition: quantized_fully_connected_layer.h:89