Hi Jim, Did you get my direct email on friday?
There's a couple of problems here: 1) You have a network of 3 layers, with a single layer in each node. A neural net works by changing the weights between arrays of nodes. Having a single node in each layer means you have something like this: o->o->o feeding activation into the left, it will flow through to the right, but when you adjust the weights, you only have 1 pathway to choose from, so the network won't learn. I've posted a very simple intro to how neural nets work at: http://www.g0n.net/nnflex/SimpleNNIntro.pdf There's also an animated gif at http://www.g0n.net/nnflex/x.gif that might help to make the structure of the net clear. 2) The second problem is that you are feeding analogue numbers into the net: [0],[0], [1/3],[1/9], [2/3],[4/9], [9/3],[9/9] You need to have a number of input units (see point 1), and translate your data into a binary format that a neural net can understand, like: [0,0,0,0],[0,0,0,0], [0,0,0,1],[0,0,0,1], [0,0,1,0],[0,1,0,0] [0,1,0,1],[1,0,0,1] (These are 0->0, 1->1, 2->4 & 3->9) etc. This assumes you have 4 input & 4 output nodes (and you'll probably need 4 hidden nodes as well), which means you can represent numbers up to 15. I'm not sure how well a simple backprop net will generalise x squared, it may learn the examples you give it, but be unable to square a number it hasn't seen before, although it might be possible to encode the data in a way that will allow a certain amount of generalisation. I'll have to have a think about that one and give it a try. charles. use strict; use AI::NNFlex::momentum; use AI::NNFlex::Dataset; # Create the network my $network = AI::NNFlex::momentum->new( learningrate=>.1, bias=>1, momentum=>0.6, round=>1); $network->add_layer( nodes=>1, activationfunction=>"tanh"); $network->add_layer( nodes=>1, activationfunction=>"tanh"); $network->add_layer( nodes=>1, activationfunction=>"linear"); $network->init(); my $dataset = AI::NNFlex::Dataset->new([ [0],[0], [1/3],[1/9], [2/3],[4/9], [9/3],[9/9]]); my $counter=0; my $err = 10; while ($err >1.001) { $err = $dataset->learn($network); print "Epoch $counter: Error = $err\n"; $counter++; } foreach (@{$dataset->run($network)}) { foreach (@$_){print $_} print "\n"; __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
