example_fcnn_training_smiles.cpp

Programa para el testeo de las funciones Fully Connected Layer.Para compilar o código example_fcnn_training_smiles.cpp:

g++ -static -o example_fcnn_training_smiles example_fcnn_training_smiles.cpp -lpdsramm -lpdsnnmm 

Para executar o programa:

./example_fcnn_training_smiles

Retornando por consola:

Ninputs:2
Noutputs:1024
Test[  0]  Accuracy %: 92.8711
Test[  1]  Accuracy %: 93.3984
Test[  2]  Accuracy %: 93.8477
Test[  3]  Accuracy %: 94.3945
Test[  4]  Accuracy %: 94.3945
Test[  5]  Accuracy %: 95.2344
Test[  6]  Accuracy %: 95.1758
...
Test[ 34]  Accuracy %: 100
Test[ 35]  Accuracy %: 100
Test[ 36]  Accuracy %: 100
Test[ 37]  Accuracy %: 100
Test[ 38]  Accuracy %: 100
Test[ 39]  Accuracy %: 100

output in sequence

Código example_fcnn_training_smiles.cpp:

 
#include <Pds/Ra>
#include <Pds/Nn>
#include <iomanip>
 
int main(void)
{
    unsigned int M=40;
    double Acc=0;
 
    // Loading Y data
    Pds::Matrix IMGA1=Pds::Matrix::ImportBmpFileAsGray("../manual/imagenes/happy.bmp");
    Pds::Matrix IMGA2=Pds::Matrix::ImportBmpFileAsGray("../manual/imagenes/sad.bmp");
    Pds::Matrix IMGB1=Pds::Matrix::ImportBmpFileAsGray("../manual/imagenes/self-confidence.bmp");
    Pds::Matrix IMGB2=Pds::Matrix::ImportBmpFileAsGray("../manual/imagenes/fear.bmp");
    Pds::Matrix IMGC0=Pds::Matrix::ImportBmpFileAsGray("../manual/imagenes/pokerface.bmp");
    
    unsigned int Nlin=IMGA1.Nlin();
    unsigned int Ncol=IMGA1.Ncol();
    
    Pds::Matrix Y(5,Nlin*Ncol);
    Y.SetRowVector(0,Pds::Vector(IMGA1));
    Y.SetRowVector(1,Pds::Vector(IMGA2));
    Y.SetRowVector(2,Pds::Vector(IMGB1));
    Y.SetRowVector(3,Pds::Vector(IMGB2));
    Y.SetRowVector(4,Pds::Vector(IMGC0));
    Y=Y.Geq(127);
    
    Pds::Matrix Ys(Y.Size());
    
    // Loading X data
    Pds::Matrix X=  " 0.0  1.0\n" // happy
                    " 0.0 -1.0\n" // sad
                    " 1.0  0.0\n" // self-confidence
                    "-1.0  0.0\n" // fear
                    " 0.0  0.0";  // poker face
    
    std::cout<<"Ninputs:"<<X.Ncol()<<std::endl;
    std::cout<<"Noutputs:"<<Y.Ncol()<<std::endl;
    
    // training
    std::vector<unsigned int> N={X.Ncol(),4096,2048,Y.Ncol()};
    Pds::FCNn NN0=Pds::FCNn(N,Pds::Sigmoid,Pds::DSigmoid);
    
    for(unsigned int k=0;(k<M)&&(Acc<1.0);k++)
    {
        NN0.Training(X,Y,0.1,0.01);
        
        NN0.Predict(X,Ys);
        
        Acc=Pds::Accuracy(Ys.Geq(0.5),Y);
        std::cout<<"Test["<<std::setw(3)<<k<<"]  ";
        std::cout<<"Accuracy %: "<<100.0*Acc<<std::endl;
    }
    
    // testing
    X=Pds::Matrix(1,X.Ncol()); Ys=Pds::Matrix(1,Y.Ncol());
    std::vector<Pds::Matrix> Block;
    Pds::Matrix IMG;
    
    for(unsigned int k=0;k<360;k=k+5)
    {
        X.Set( 0, cos(k*Pds::Ra::Pi/180.0) );
        X.Set( 1, sin(k*Pds::Ra::Pi/180.0) );
        NN0.Predict(X,Ys);
        
        IMG=Ys.Remodel(Nlin,Ncol)*255;
        Block.push_back(IMG);
    }
 
    Pds::Matrix::ExportAsGif(Block,"smiles.gif",20);
            
    return 0;
}