Russian Qt Forum

C++ (Qt)
#include <iostream>
#include <vector>
#include <cmath>
#include <iomanip>
 
#include "maxnet.h"
 
static constexpr unsigned int period = 10; // число предыдущих шагов для определения вероятности следующего шага  
static constexpr size_t SIZE = 550; // Полный размер данных 
static constexpr size_t TRAINING_SIZE = 500; // Часть данных, для обучения сети
 
template <class V, class F>
void apply(V & data, F f)
{
    for (size_t i = 0; i < data.size(); ++i)
    {
        data[i] = f(i);
    }
}
 
int main()
{    
    std::vector<bool> data(SIZE);
 
    apply(data, [](size_t i)->bool { return (sin(i) + cos(4*i) >= 0); }); // периодическая функция
 
    vanda::maxnet<period, 0> maxnet; // Наша сеть)
 
    // Разделяем данные на две части:
    // для обучения и тестирования
    auto end = data.begin()+TRAINING_SIZE; 
 
    maxnet.training(data.begin(), end); // обучаем сеть
 
    auto beg = end;
    end += period;
    // Тестируем её на оставшейся части данных 
    while (end != data.end())
    {
        auto p = maxnet.prediction_for(beg++, end);
        std::cout   << "observed value = " << *end++
                    << "\t p(1) = " << p.probability_for_1
                    << "\t  p(0) = " << p.probability_for_0 << std::endl;
    }
 
    std::cout << std::endl << "number of groups = " << maxnet.num_groups() << std::endl;
 
    return 0;
}
 

C++ (Qt)
observed value = 1	 p(1) = 0.894737	  p(0) = 0.105263
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 1	 p(1) = 0.941176	  p(0) = 0.0588235
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 1	 p(1) = 0.47619	  p(0) = 0.52381
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0.454545	  p(0) = 0.545455
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 0	 p(1) = 0.5	  p(0) = 0.5
observed value = 1	 p(1) = 0.5	  p(0) = 0.5
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 0	 p(1) = 0.894737	  p(0) = 0.105263
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0.47619	  p(0) = 0.52381
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 1	 p(1) = 0.3	  p(0) = 0.7
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0.5	  p(0) = 0.5
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 0	 p(1) = 0	  p(0) = 1
 
number of groups = 59
 

C++ (Qt)
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 0	 p(1) = 0.921875	  p(0) = 0.078125
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0	  p(0) = 1
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 0	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 1	 p(1) = 0.0448878	  p(0) = 0.955112
observed value = 1	 p(1) = 0.944444	  p(0) = 0.0555556
observed value = 1	 p(1) = 0.941176	  p(0) = 0.0588235
observed value = 1	 p(1) = 1	  p(0) = 0
observed value = 1	 p(1) = 0.921875	  p(0) = 0.078125
 

C++ (Qt)
#include <iostream>
#include <vector>
#include <boost/format.hpp>
#include "tiny_dnn/tiny_dnn.h"
 
using namespace std;
using namespace tiny_dnn;
using boost::format;
 
static constexpr size_t IN_SIZE       = 30;   // число предыдущих шагов для определения вероятности следующего шага
static constexpr size_t OUT_SIZE      = 2;
static constexpr size_t SIZE          = 2500; // Полный размер данных
static constexpr size_t TRAINING_SIZE = 2000; // Часть данных, для обучения сети
static constexpr size_t BATCH_SIZE    = 10;
static constexpr size_t EPOCHS        = 1000;
 
inline float toValue( bool v )
{
  return v? 1. : 0.;
}
 
int main( int /*argc*/, char */*argv*/[] )
{
  network<sequential> net;
  net << fully_connected_layer( IN_SIZE, 30 )
      << tanh_layer()
      << fully_connected_layer( 30, 30 )
      << tanh_layer()
      << fully_connected_layer( 30, OUT_SIZE );
 
  cout << "Generate source..." << endl;
  vector<bool> source( SIZE );
  for( size_t i = 0; i < source.size(); ++i )
    source[ i ] = sin( i ) + cos( 4 * i ) >= 0;
 
  cout << "Prepare datas..." << endl;
  const size_t numData = TRAINING_SIZE - IN_SIZE;
 
  std::vector<vec_t> inData;
  inData.reserve( numData );
  std::vector<vec_t> outData;
  outData.reserve( numData );
  for( size_t i = IN_SIZE; i < TRAINING_SIZE; ++i )
  {
    vec_t vIn( IN_SIZE );
    for( size_t j = 0; j < IN_SIZE; ++j )
      vIn[ j ] = toValue( source[ i - IN_SIZE + j ] );
 
    vec_t vOut( OUT_SIZE );
    vOut[ 0 ] = toValue( !source[ i ] );
    vOut[ 1 ] = toValue( source[ i ] );
 
    inData.push_back( vIn );
    outData.push_back( vOut );
  }
 
  int iEpoch = 0;
  auto on_enumerate_epoch = [&](){
    if( ++iEpoch % 100 )
      return;
 
    const double loss = net.get_loss<mse>( inData, outData );
    cout << format( "epoch = %1%/%2% loss = %3%" ) % iEpoch % EPOCHS % loss << endl;
  };
 
  cout << "Training..." << endl;
  adamax opt;
  net.fit<mse>( opt, inData, outData, BATCH_SIZE, EPOCHS, [](){}, on_enumerate_epoch );
 
  cout << "Checking..." << endl;
  size_t errCnt = 0;
  for( size_t i = TRAINING_SIZE; i < SIZE; ++i )
  {
    vec_t vIn( IN_SIZE );
    for( size_t j = 0; j < IN_SIZE; ++j )
      vIn[ j ] = toValue( source[ i - IN_SIZE + j ] );
 
    const vec_t vOut = net.predict( vIn );
    assert( vOut.size() == 2 );
 
    const bool observed = source[ i ];
    const bool predicted = vOut[ 0 ] < vOut[ 1 ];
    cout << format( "i = %1%\t[%2%]\tobserved = %3%\tpredicted = %4%\t[0] = %5%\t[1] = %6%" )
            % i
            % ((observed != predicted)? '!' : ' ')
            % observed
            % predicted
            % vOut[ 0 ]
            % vOut[ 1 ]
          << std::endl;
 
    if( observed != predicted )
      ++errCnt;
  }
 
  cout << format( "Error = %1%/%2% %3%%%" ) % errCnt % (SIZE - TRAINING_SIZE) % (float( errCnt ) / float( SIZE - TRAINING_SIZE ) * 100.) << std::endl;
 
  return 0;
}
 

C++ (Qt)
sin( i ) + cos( 4 * i ) + sin( 8 * i )
 

C++ (Qt)
sin(x+pi*i/4.0)
 

Bash
tiny_dnn:
=================================
Total p = 0.887292 +- 0.104493
Training time = 149152 ms
 
Total p = 0.867292 +- 0.119214
Training time = 148662 ms
 
Total p = 0.878542 +- 0.110621
Training time = 147957 ms
 
Total p = 0.876667 +- 0.112411
Training time = 148513 ms
 
Total p = 0.86875 +- 0.117555
Training time = 148443 ms
 
vanda:
=================================
Total p = 0.875 +- 0.113129
Total groups = 16
Training time = 0 ms
 
Total p = 0.878958 +- 0.110667
Total groups = 20
Training time = 0 ms
 
Total p = 0.861667 +- 0.125094
Total groups = 20
Training time = 0 ms
 
Total p = 0.870417 +- 0.117708
Total groups = 22
Training time = 0 ms
 
Total p = 0.873125 +- 0.114757
Total groups = 8
Training time = 0 ms