99 lines
4.2 KiB
C++
99 lines
4.2 KiB
C++
/********************************************************
|
|
* ██████╗ ██████╗████████╗██╗
|
|
* ██╔════╝ ██╔════╝╚══██╔══╝██║
|
|
* ██║ ███╗██║ ██║ ██║
|
|
* ██║ ██║██║ ██║ ██║
|
|
* ╚██████╔╝╚██████╗ ██║ ███████╗
|
|
* ╚═════╝ ╚═════╝ ╚═╝ ╚══════╝
|
|
* Geophysical Computational Tools & Library (GCTL)
|
|
*
|
|
* Copyright (c) 2022 Yi Zhang (yizhang-geo@zju.edu.cn)
|
|
*
|
|
* GCTL is distributed under a dual licensing scheme. You can redistribute
|
|
* it and/or modify it under the terms of the GNU Lesser General Public
|
|
* License as published by the Free Software Foundation, either version 2
|
|
* of the License, or (at your option) any later version. You should have
|
|
* received a copy of the GNU Lesser General Public License along with this
|
|
* program. If not, see <http://www.gnu.org/licenses/>.
|
|
*
|
|
* If the terms and conditions of the LGPL v.2. would prevent you from using
|
|
* the GCTL, please consider the option to obtain a commercial license for a
|
|
* fee. These licenses are offered by the GCTL's original author. As a rule,
|
|
* licenses are provided "as-is", unlimited in time for a one time fee. Please
|
|
* send corresponding requests to: yizhang-geo@zju.edu.cn. Please do not forget
|
|
* to include some description of your company and the realm of its activities.
|
|
* Also add information on how to contact you by electronic and paper mail.
|
|
******************************************************/
|
|
|
|
#ifndef _GCTL_DNN_HLAYER_H
|
|
#define _GCTL_DNN_HLAYER_H
|
|
|
|
#include "activation_identity.h"
|
|
#include "activation_mish.h"
|
|
#include "activation_relu.h"
|
|
#include "activation_prelu.h"
|
|
#include "activation_sigmoid.h"
|
|
#include "activation_softmax.h"
|
|
#include "activation_tanh.h"
|
|
|
|
namespace gctl
|
|
{
|
|
enum hlayer_type_e
|
|
{
|
|
FullyConnected,
|
|
MaxPooling,
|
|
AvgPooling,
|
|
Convolution,
|
|
};
|
|
|
|
enum pad_type_e
|
|
{
|
|
Valid,
|
|
Same,
|
|
};
|
|
|
|
class dnn_hlayer
|
|
{
|
|
public:
|
|
dnn_hlayer();
|
|
virtual ~dnn_hlayer();
|
|
|
|
int in_size() const;
|
|
int out_size() const;
|
|
int obs_size() const;
|
|
|
|
const matrix<double> &forward_propagation_data();
|
|
const matrix<double> &backward_propagation_data();
|
|
const matrix<double> &get_linear_term();
|
|
const matrix<double> &get_linear_gradient();
|
|
|
|
activation_type_e get_activation_type() const;
|
|
std::string get_activation_name() const;
|
|
|
|
virtual void forward_propagation(const array<double> &all_weights, const matrix<double> &prev_layer_data) = 0;
|
|
virtual void backward_propagation(const array<double> &all_weights, const array<double> &all_ders,
|
|
const matrix<double> &prev_layer_data, const matrix<double> &next_layer_data) = 0;
|
|
|
|
virtual hlayer_type_e get_layer_type() const = 0;
|
|
virtual std::string get_layer_name() const = 0;
|
|
virtual std::string layer_info() const = 0;
|
|
virtual void save_layer_setup(std::ofstream &os) const = 0;
|
|
virtual void load_layer_setup(std::ifstream &is) = 0;
|
|
virtual void save_weights2text(const array<double> &all_weights, std::ofstream &os) const = 0;
|
|
|
|
protected:
|
|
void cal_valid_padding_idx(array<int> &idx, int i, int j, int i_rows, int i_cols, int p_rows, int p_cols, int s_rows, int s_cols);
|
|
void cal_same_padding_idx(array<int> &idx, int i, int j, int i_rows, int i_cols, int p_rows, int p_cols, int s_rows, int s_cols, int u_pad, int l_pad);
|
|
|
|
protected:
|
|
dnn_activation *activator_; // activation object
|
|
int w_is_, w_outs_; // weight start index, weight input size, weight output size
|
|
int o_is_; // observation input size
|
|
matrix<double> z_; // Linear term, z = W' * in + b
|
|
matrix<double> a_; // Output of this layer, a_ = act(z)
|
|
matrix<double> der_in_; // Derivative of the input of this layer, which is also the output of previous layer
|
|
matrix<double> der_z_; // Derivative of the linear term
|
|
};
|
|
}
|
|
|
|
#endif // _GCTL_DNN_HLAYER_H
|