Create Cpp_IA_Vere_002

[danyveresh]

include

include

include

include

include

define layers 3

define N 784

define P 70

define Q 10

//#define R 5

using namespace std;

ifstream fTrain;

void readTraining(double y[], double y0[]) { printf("TRaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa\n"); int temp; fTrain >> temp; //cout << temp << endl; for (int i = 0; i < Q; ++i) { y[i] = 0; } y[temp] = 1.0; //cout << y[temp] << endl; for (int i = 0; i < Q; ++i) { //printf("%1.4lf - ", y[i]); } //cout << endl; //cout << y[5] << endl; for (int i = 0; i < N; ++i) { fTrain >> y0[i]; //printf("%.4lf ", y0[i]); } //fTrain >> temp; } double sigmoid(double x) { return 1 / (1 + exp(-x)); }

int main(void) { fTrain.open("train.txt");

unsigned int nr_noduri[layers] = { N,P,Q };
double alfa = 0.3;
double weights01[N][P], weights12[P][Q];
double deltaWeights01[N][P], deltaWeights12[P][Q];
double bias01[P], bias12[Q];
double deltaBias01[P], deltaBias12[Q];
double y0[N], y1[P], y2[Q], y[Q];
double erori[Q];
double gradient01[P], gradient12[Q];
double sigma = 1;
unsigned short int count = 0;
unsigned long int generatii = 0;

//randomize wights and biases
//works: 5
//not work: 14
//srand(5);
srand(time(0));
for (int i = 0; i < N; ++i)
{
    for (int j = 0; j < P; ++j)
    {
        weights01[i][j] = 1. / N;// double(rand() % 1000) / 500 * 2.4 / nr_noduri[0] - 2.4 / nr_noduri[0];
        //printf("%+0.6lf ", weights01[i][j]);
    }
    //printf("\n");
}
//printf("\n");

for (int i = 0; i < P; ++i)
{
    for (int j = 0; j < Q; ++j)
    {
        weights12[i][j] = 1. / P;// double(rand() % 1000) / 500 * 2.4 / nr_noduri[0] - 2.4 / nr_noduri[0];
        //printf("%+0.6lf ", weights12[i][j]);
    }
    //printf("\n");
}
//printf("\n");

for (int i = 0; i < P; ++i)
{
    bias01[i] = double(rand() % 1000) / 500 * 2.4 / nr_noduri[0] - 2.4 / nr_noduri[0];
    //printf("%+0.6lf ", bias01[i]);
}
//printf("\n\n");

for (int i = 0; i < Q; ++i)
{
    bias12[i] = double(rand() % 1000) / 500 * 2.4 / nr_noduri[0] - 2.4 / nr_noduri[0];
    //printf("%+0.6lf ", bias12[i]);
}
//printf("\n");
//printf("1\n");
sigma = 10000;
while (sigma > 0.001)
{
    //printf("2\n");
    //sigma = 0;
    readTraining(y, y0);
    --sigma;

    for (int aa = 0; aa < 1; ++aa)
    {
        //printf("Da?\n");
        //printf("wwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwww\t\taa = %d   wwwwwwwwwwwwwwwwwwwwwwwwwwwwwww\n", aa);
        //read inputs -> expected outputs

        //calculate y1
        //printf("\n");
        for (int i = 0; i < P; ++i)
        {
            y1[i] = -bias01[i];
            //printf("%y1[%d]=%lf \n", i, y1[i]);
            for (int j = 0; j < N; ++j)
            {
                y1[i] = y1[i] + y0[j] * weights01[j][i];
                //printf("   %0.4lf %0.4lf %0.4lf \n", y1[i], y0[j], weights01[j][i]);
            }
            //printf("%y1[%d]=%.10lf \n", i, y1[i]);
            y1[i] = sigmoid(y1[i]);
            //printf("%y1[%d]=%.10lf \n", i, y1[i]);
        }
        //printf("\n");
        //printf("\n");

        //calculate y2
        for (int j = 0; j < Q; ++j)
        {
            y2[j] = -bias12[j];
            //printf("y2[%d]=%lf \n", j, y2[j]);
            for (int k = 0; k < P; ++k)
            {
                y2[j] = y2[j] + y1[k] * weights12[k][j];
                //printf("    %0.4lf %0.4lf %0.4lf \n", y2[j], y1[k], weights12[j][k]);
            }
            //printf("%y2[%d]=%.10lf \n", j, y2[j]);
            y2[j] = sigmoid(y2[j]);
            //printf("%y2[%d]=%.10lf \n", j, y2[j]);
        }
        //printf("\n");

        //printf("--%lf %lf %.10lf --\n", y0[0], y0[1], y2[0]);

        //printf("-----------------------------------------------\n");
        for (int k = 0; k < Q; ++k)
        {
            //calculam erorile pentru iesire
            erori[k] = y[k] - y2[k];
            //calculam gradientul pentru nuronul/neuronii respectivi
            gradient12[k] = y2[k] * (1 - y2[k]) * erori[k];
            //printf("gradient12 -----> %.10lf\n", gradient12[k]);
            //cout << k << " " << erori[k] << " OOOO gradient12" << gradient12[k] << endl;
        }

        //caculam delta pentru stratul de iesire
        for (int j = 0; j < P; ++j)
        {
            for (int k = 0; k < Q; ++k)
            {
                deltaWeights12[j][k] = alfa * y1[j] * gradient12[k];
                //printf("dw[%d][%d]=%.10lf   ", j, k, deltaWeights12[j][k]);
                //cout << deltaWeights12[j][k] << " ---" << endl;
            }
            //printf("\n");
        }
        //calculam delta pentru bias
        //printf("___ deltaBias12\n");
        for (int k = 0; k < Q; ++k)
        {
            deltaBias12[k] = -alfa * gradient12[k];
            //printf("%.10lf ", deltaBias12[k]);
        }

        //printf("\n-----------------------------------------------\n");

        for (int j = 0; j < P; ++j)
        {
            double suma = 0;
            //calculam suma/sumele 
            for (int k = 0; k < Q; ++k)
            {
                suma = suma + gradient12[k] * weights12[j][k];
                //printf("\tgr12[%d] = %.10lf   ---   we[%d][%d] = %.10lf\n",k, gradient12[k],j,k,weights12[j][k]);
            }
            //calculam gradientul pentru nuronul/neuronii respectivi
            gradient01[j] = y1[j] * (1 - y1[j]) * suma;
            //cout << j << "  " << suma << " " << gradient01[j] << endl;
            //printf("gr01[%d] = %.10lf * (1 - %.10lf) * %.10lf = %.10lf \n", j, y1[j], y1[j], suma, gradient01[j]);
        }

        //calculam delta pentru neuronii ascunsi
        for (int i = 0; i < N; ++i)
        {
            //cout << "y0[" << i << "]=" << y0[i] << endl;
            for (int j = 0; j < P; ++j)
            {
                deltaWeights01[i][j] = alfa * y0[i] * gradient01[j];
                //printf("dw[%d][%d]=%.10lf   ", i, j, deltaWeights01[i][j]);
                //cout << deltaWeights01[i][j] << " __ ";
            }
            //printf("\n");
        }

        //printf("___ deltaBias01\n");
        for (int j = 0; j < P; ++j)
        {
            deltaBias01[j] = -alfa * gradient01[j];
            //printf("%.10lf ", deltaBias01[j]);
        }
        //printf("\n");

        for (int i = 0; i < N; ++i)
        {
            for (int j = 0; j < P; ++j)
            {
                //printf("%.10lf_%.10lf ", weights01[i][j], deltaWeights01[i][j]);
            }
            //printf("\n");
        }

        //printf("\n weights01 \n");
        for (int i = 0; i < N; ++i)
        {
            for (int j = 0; j < P; ++j)
            {
                weights01[i][j] = weights01[i][j] + deltaWeights01[i][j];
            }
            //printf("\n");
        }

        //printf("\n weights12 \n");
        for (int j = 0; j < P; ++j)
        {
            for (int k = 0; k < Q; ++k)
            {
                weights12[j][k] = weights12[j][k] + deltaWeights12[j][k];
            }
            //printf("\n");
        }

        //noul bias01
        //printf("%.10lf %.10lf --------\n", deltaBias01[0], deltaBias01[1]);
        for (int j = 0; j < P; ++j)
        {
            //printf("bias01[%d]= %.10lf + %.10lf =  ", j, bias01[j], deltaBias01[j]);
            bias01[j] = bias01[j] + deltaBias01[j];
            //printf("%.10lf \n", bias01[j]);
        }
        //printf("%.10lf %.10lf --------\n", deltaBias01[0], deltaBias01[1]);
        //printf("%.10lf %.10lf --------\n", bias01[0], bias01[1]);
        //printf("\n");
        //noul bias12
        for (int k = 0; k < Q; ++k)
        {
            //printf("bias12[%d]= %.10lf + %.10lf =  ", k, bias12[k], deltaBias12[k]);
            bias12[k] = bias12[k] + deltaBias12[k];
            //printf(" %.10lf", bias12[k]);
        }

        //printf("\n weights01 \n");
        for (int i = 0; i < N; ++i)
        {
            for (int j = 0; j < P; ++j)
            {
                //printf("%.10lf ", weights01[i][j]);
            }
            //printf("\n");
        }

        //printf("\n weights12 \n");
        for (int j = 0; j < P; ++j)
        {
            for (int k = 0; k < Q; ++k)
            {
                //printf("%.10lf ", weights12[j][k]);
            }
            //printf("\n");
        }

        for (int a1 = 0; a1 < Q; ++a1)
        {
            //sigma = sigma + (y[a1] - y2[a1]) * (y[a1] - y2[a1]);
        }
        printf("s=%lf\n", sigma);
    }
    for (int p = 0; p < 4; ++p)
    {
        for (int k = 0; k < Q; ++k)
        {
            //sigma = sigma + (y[0] - y2[0]) * (y[0] - y2[0]);
        }
    }
    if (count == 0)
    {
        printf("sigma %.8lf  %d\n", sigma, generatii);
    }
    ++count;
    ++generatii;
    if (generatii > 3000000)
    {
        break;
    }
}
//printf("n\n");
// ------------------------------------------------------------------------------------------------

//noul bias01
for (int j = 0; j < P; ++j)
{
    printf("bias01[%d]= %.10lf \n", j, bias01[j]);
}
//printf("\n");
//noul bias12
for (int k = 0; k < Q; ++k)
{
    printf("bias12[%d]= %.10lf \n", k, bias12[k]);
}

//printf("\n weights01 \n");
for (int i = 0; i < N; ++i)
{
    for (int j = 0; j < P; ++j)
    {
        //printf("%.10lf ", weights01[i][j]);
    }
    //printf("\n");
}

//printf("\n weights12 \n");
for (int j = 0; j < P; ++j)
{
    for (int k = 0; k < Q; ++k)
    {
        //printf("%.10lf ", weights12[j][k]);
    }
    //printf("\n");
}

for (int qq = 0; qq < 5; ++qq)
{
    //readInputs(y0);
    //readOutputs(y, y0);
    readTraining(y, y0);
    for (int i = 0; i < P; ++i)
    {
        y1[i] = -bias01[i];
        //printf("%y1[%d]=%lf \n", i, y1[i]);
        for (int j = 0; j < N; ++j)
        {
            y1[i] = y1[i] + y0[j] * weights01[j][i];
            //printf("   %0.4lf %0.4lf %0.4lf \n", y1[i], y0[j], weights01[i][j]);
        }
        //printf("%y1[%d]=%lf \n", i, y1[i]);
        y1[i] = sigmoid(y1[i]);
        //printf("%y1[%d]=%lf \n", i, y1[i]);
    }
    //printf("\n");
    //printf("\n");

    //calculate y2
    for (int j = 0; j < Q; ++j)
    {
        y2[j] = -bias12[j];
        //printf("y2[%d]=%lf \n", j, y2[j]);
        for (int k = 0; k < P; ++k)
        {
            y2[j] = y2[j] + y1[k] * weights12[k][j];
            //printf("    %0.4lf %0.4lf %0.4lf \n", y2[j], y1[k], weights12[j][k]);
        }
        //printf("%y2[%d]=%lf \n", j, y2[j]);
        y2[j] = sigmoid(y2[j]);
        //printf("%y2[%d]=%lf \n", j, y2[j]);
    }

    //printf("--%lf %lf %lf %lf %lf %lf --\n", y0[0], y0[1], y2[0], y2[1], y2[2], y2[3]);
    for (int hh = 0; hh < Q; ++hh)
    {
        printf("%.4lf ", y2[hh]);
    }
    printf("\n");
    for (int hh = 0; hh < Q; ++hh)
    {
        printf("%.4lf ", y[hh]);
    }
    printf("\n");
}

printf("err = %.6lf \n", sigma);
printf("nr_generatii = %u\n", generatii);

fTrain.close();
return 0;

}