int main() {
Tensor * X_train = load_X("../data/train-images.idx3-ubyte", TRAIN_SIZE);
float ** Y_train = load_Y("../data/train-labels.idx1-ubyte", TRAIN_SIZE);
Tensor * X_test = load_X("../data/t10k-images.idx3-ubyte", TEST_SIZE);
float ** Y_test = load_Y("../data/t10k-labels.idx1-ubyte", TEST_SIZE);
const int num_layers = 2;
Layer ** layers = new Layer*[num_layers];
layers[0] = new FullyConnectedLayer(100, 784, SIGMOID);
layers[1] = new FullyConnectedLayer(10, 100, SIGMOID);
// Train neural network
ConvNet net = ConvNet(layers, num_layers, X_train, Y_train);
net.train(0.01, 100, TRAIN_SIZE / 10, 10, TRAIN_SIZE);
}
static int pvalues(int ii)
{
int i;
char x[LLENGTH];
double *freq;
long l;
double a;
strcpy(x,spb[ii]);
i=load_X(x); if (i<0) return(-1);
/* Rprintf("\ndim=%d,%d",mX,nX); getch(); */
i=data_open(word[1],&d); if (i<0) return(-1);
v=(int *)muste_malloc(mX*sizeof(int));
if (v==NULL) { ei_tilaa(); return(-1); }
i=nrot();
mT=mX; nT=2; rlabT=rlabX ;
i=mat_alloc_lab(&T,mT,nT,NULL,&clabT);
freq=T+mT;
for (i=0; i<mX; ++i) freq[i]=0.0;
i=conditions(&d); if (i<0) return(-1);
n=0L; sur_print("\n");
for (l=d.l1; l<=d.l2; ++l)
{
if (unsuitable(&d)) continue;
sprintf(sbuf,"%ld ",l); sur_print(sbuf);
++n;
for (i=0; i<mX; ++i)
{
data_load(&d,l,v[i],&a);
if (a==MISSING8) continue;
if (a>X[i]) ++freq[i];
}
}
if (n==0L)
{
sur_print("\nNo active observations!"); WAIT; return(-1);
}
a=1.0/(double)n;
for (i=0; i<mX; ++i)
{
T[i]=X[i];
freq[i]*=a;
}
strncpy(clabT,"Value P ",16);
sprintf(exprT,"Tail_frequencies_in_data_%s_N=%d",word[1],n);
save_T("TAILFREQ.M");
return(1);
}
