Dave,
Thanks for the interesting reply.
" Basically, I train everything for 12 hours before I even look at it. I have trained certain nets for as long as 150 hours, but I'm
completely unconvinced that it either helped or was necessary. "
In the past, I let some nets train for over 40 hours without seeing any improvement over what I saw at 15 hours. My best net, a very good looking net on COMS was trained 13 hours. I'm finding that if they don't look good, by that I mean catching the bulk of the major trends, after 3 or 4 hours, they aren't going to get to what I'd consider a tradable state.
But is it really correct to compare hours trained when we're all using different CPUs? I'm using a P200.
"If you look at one of the *.neu files with a text editor, you will see all your input info in the beginning"
Among the parameters I see in the file is in nodes and out nodes, knowing that and the number of neurons can the number of weights be calculated? Looking at one of my .neu files I saw 2200 numerical values (weights ?). I've read one book and several articles on NNets but am lost on calculating the number of neurons and weights. It also covered "early stopping" but that topic went totally over my head. (please don't feel obligated to try explaining any further unless you really want to).
BTW, my COMS net called the turn around today perfectly.
CL |
