vlsvrs 1.0.0

VLSV reader with C and Python bindings
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165

/* Author: Kostis Papadakis
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version 2
 * of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301,
 * USA.
 * */
#pragma once
#include "genericTsPool.h"
#include "matrix.h"

namespace TINYAI {

template <typename T, BACKEND Backend>
class BaseLayer {
public:
   size_t neurons = 0;
   T wmega = 1.0;
   GENERIC_TS_POOL::MemPool* _pool;
   NumericMatrix::Matrix<T, Backend> buffer, w, w_t, b, b_broadcasted, z, a, a_prime, a_t, dw, db, delta, delta_store;
   NumericMatrix::Matrix<T, Backend> m_w, m_b, v_w, v_b;
   virtual ~BaseLayer() = default;
   virtual void forward(const NumericMatrix::Matrix<T, Backend>& input, tinyAI_blasHandle_t* handle,
                        tinyAI_gpuStream_t stream) noexcept = 0;

   virtual void forward(const NumericMatrix::MatrixView<T>& input, tinyAI_blasHandle_t* handle,
                        tinyAI_gpuStream_t stream) noexcept = 0;

   virtual void forward(const NumericMatrix::ConstMatrixView<T>& input, tinyAI_blasHandle_t* handle,
                        tinyAI_gpuStream_t stream) noexcept = 0;

   virtual void setup(size_t neurons, size_t input, size_t batchSize, size_t layer_id) = 0;
   virtual void reset(size_t input, size_t layer_id) = 0;
   void print_stats(std::size_t layer_id){
      const std::size_t sz=this->w.size();
      NumericMatrix::HostMatrix<T> w_host(this->w);
      const auto [min,max] = std::minmax_element(w_host.begin(),w_host.end());
      const T sum = std::accumulate(w_host.begin(),w_host.end(),T(0));
      const T mu = sum/w_host.size();
      const T variance = std::accumulate(w_host.begin(), w_host.end(), T(0),
                                  [mu](T acc, T x) { return acc + (x - mu) * (x - mu); }) / w_host.size();
      const T sigma = std::sqrt(variance);
      printf("Layer {%zu}\n\t Min,Max,Mu,Sigma=[%f,%f,%f,%f]\n",layer_id,*min,*max,mu,sigma);
   }
};

template <typename T, ACTIVATION Activation, BACKEND Backend>
class LinearLayer : public BaseLayer<T, Backend> {
public:
   LinearLayer() { this->_pool = nullptr; }
   LinearLayer(GENERIC_TS_POOL::MemPool* p) { this->_pool = p; }
   LinearLayer(size_t n, GENERIC_TS_POOL::MemPool* p) {
      this->neurons = n;
      this->_pool = p;
   }

   void setup(size_t neurons, size_t input, size_t batchSize, size_t layer_id) override {
      assert(neurons > 0 && "This layer has 0 neurons!");
      this->w = NumericMatrix::Matrix<T, Backend>(input, neurons, this->_pool);
      this->m_w = NumericMatrix::Matrix<T, Backend>(input, neurons, this->_pool);
      this->v_w = NumericMatrix::Matrix<T, Backend>(input, neurons, this->_pool);
      this->b = NumericMatrix::Matrix<T, Backend>(1, neurons, this->_pool);
      this->m_b = NumericMatrix::Matrix<T, Backend>(1, neurons, this->_pool);
      this->v_b = NumericMatrix::Matrix<T, Backend>(1, neurons, this->_pool);
      this->z = NumericMatrix::Matrix<T, Backend>(batchSize, neurons, this->_pool);
      this->a = NumericMatrix::Matrix<T, Backend>(batchSize, neurons, this->_pool);
      this->b_broadcasted = NumericMatrix::Matrix<T, Backend>(batchSize, neurons, this->_pool);
      this->a_prime = NumericMatrix::Matrix<T, Backend>(batchSize, neurons, this->_pool);
      this->w_t = NumericMatrix::Matrix<T, Backend>(neurons, input, this->_pool);
      this->a_t = NumericMatrix::Matrix<T, Backend>(neurons, batchSize, this->_pool);
      this->delta = NumericMatrix::Matrix<T, Backend>(batchSize, neurons, this->_pool);
      this->delta_store = NumericMatrix::Matrix<T, Backend>(batchSize, neurons, this->_pool);
      this->buffer = NumericMatrix::Matrix<T, Backend>(batchSize, input, this->_pool);
      this->dw = NumericMatrix::Matrix<T, Backend>(input, neurons, this->_pool);
      this->db = NumericMatrix::Matrix<T, Backend>(1, neurons, this->_pool);

      const T fan_in = input;
      const T fan_out = neurons;
      T std = std::sqrt(2.0 / (fan_in));
      if constexpr (Activation == ACTIVATION::SIN) {
         if (layer_id == 0) {
            this->wmega = 10;
            std = std::sqrt(6.0f / (T)input);
         } else {
            this->wmega = 1;
            std = std::sqrt(6.0f / ((T)input));
         }
      }
      // printf("Layer init with std=%f\n",std);
      if constexpr (Backend == BACKEND::DEVICE) {
         NumericMatrix::HostMatrix<T> _w(this->w);
         NumericMatrix::mat_randomise(_w, std);
         NumericMatrix::get_from_host(this->w, _w);
      } else {
         NumericMatrix::mat_randomise(this->w, std);
      }
      this->b.zero_out();
   }

   void reset(size_t input, size_t layer_id) override {
      const T fan_in = input;
      const T fan_out = this->neurons;
      T std = std::sqrt(2.0 / (fan_in + fan_out));
      if constexpr (Activation == ACTIVATION::SIN) {
         if (layer_id == 0) {
            this->wmega = 10;
            std = std::sqrt(6.0f / (T)input);
         } else {
            this->wmega = 1;
            std = std::sqrt(6.0f / ((T)input));
         }
      }

      if constexpr (Backend == BACKEND::DEVICE) {
         NumericMatrix::HostMatrix<T> _w(this->w.nrows(), this->w.ncols());
         NumericMatrix::HostMatrix<T> _b(this->b.nrows(), this->b.ncols());
         NumericMatrix::export_to_host(this->w, _w);
         NumericMatrix::export_to_host(this->b, _b);
         NumericMatrix::mat_randomise(_w, std);
         NumericMatrix::mat_randomise(_b, std);
         NumericMatrix::get_from_host(this->w, _w);
         NumericMatrix::get_from_host(this->b, _b);
      } else {
         NumericMatrix::mat_randomise(this->w, std);
         NumericMatrix::mat_randomise(this->b, std);
      }
   }

   void forward(const NumericMatrix::Matrix<T, Backend>& input, tinyAI_blasHandle_t* handle,
                tinyAI_gpuStream_t stream) noexcept override {
      assert(this->neurons > 0 && "This layer has 0 neurons!");
      NumericMatrix::matmul(input, this->w, this->z, handle);
      NumericMatrix::matbroadcast(this->b, this->b_broadcasted, stream);
      NumericMatrix::matadd_and_activate<T, Activation>(this->z, this->b_broadcasted, this->z, this->a, this->wmega,
                                                        handle, stream);
   }

   void forward(const NumericMatrix::MatrixView<T>& input, tinyAI_blasHandle_t* handle,
                tinyAI_gpuStream_t stream) noexcept override {
      assert(this->neurons > 0 && "This layer has 0 neurons!");
      NumericMatrix::matmul(input, this->w, this->z, handle);
      NumericMatrix::matbroadcast(this->b, this->b_broadcasted, stream);
      NumericMatrix::matadd_and_activate<T, Activation>(this->z, this->b_broadcasted, this->z, this->a, this->wmega,
                                                        handle, stream);
   }

   void forward(const NumericMatrix::ConstMatrixView<T>& input, tinyAI_blasHandle_t* handle,
                tinyAI_gpuStream_t stream) noexcept override {
      assert(this->neurons > 0 && "This layer has 0 neurons!");
      NumericMatrix::matmul(input, this->w, this->z, handle);
      NumericMatrix::matbroadcast(this->b, this->b_broadcasted, stream);
      NumericMatrix::matadd_and_activate<T, Activation>(this->z, this->b_broadcasted, this->z, this->a, this->wmega,
                                                        handle, stream);
   }
};
} // namespace TINYAI