Practical Neural Network Recipies in C++

Appendix

463

copy weights - Copy the weights from one network to another

Note that this 1s NOT like a copy or assignment,

as 1t does not copy other parameters. In fact,

1t gets sizes from the calling Instance!

void LayerNet::copy weights ( LayerNet *dest , LayerNet *source )

{

1nt n ;

dest->neterr — source->neterr ;

1f (source->exe && dest->exe) // These may be Important too!

memcpy ( dest->confusi on . source->confusi on . (nout+1) * slzeofdnt) )

1f (nhldl — 0) { // No hidden layer

n - nout * (n1n+l) ;

memcpy ( dest->out coefs , source->out coefs , n * s1zeof(doubl e) ) ;

else 1f (nh1d2 — 0) { // One hidden layer

n - nhldl * (n1n+l) ;

memcp ...