Greetings Perlfolk, ** What is this? (This version includes 3 new major features.) AI::NeuralNet::Mesh is an optimized, accurate neural network Mesh written in Perl. It was designed with accruacy and speed in mind. This better learning accuracy (by twenty-three percent), as well as a much faster learning rate and run speed over any other neural network simulater that I know of in Perl.(*) Included are fifteen seperate example scripts. In particular demonstration of the accuracy and speed of this network model, check out ex_add2.pl, ex_dow.pl, and ex_add.pl. As always, included is a cleaned, CSS-ed, HTML-format of the POD docs. ** What's new? >From the POD: This is version *0.31*, the second release of this module. In this version, I have included three major features. Also in this release I have included two minor fixes which increase the learning speed of networks. I also fixed a bug in the load_pcx() method which prevented it from loading the PCX::Loader module correctly. This version also has the ability to have negative weights in the network. The major features added are: LAYER SIZES Rodin Porrata once suggested it would be good to have control over each layer's node size. Well, Rodin, here you go. Each layer can have a custom number of nodes, which you can set in two ways, detailed in the new() constructor, below. Layer sizes are preserved across load() and save() calls. LAYER EXTENSION With the ability to have custom layer sizes, I have also included the ability to extend layer sizes after network construction. You can add nodes with extend() or extend_layer() after the network is constructed or loaded. CUSTOM NODE ACTIVATION Ahh, and another treat. You can choose from one of four activation functions and set the activation function by layer, or you can even set each individual node to a seperate activation function. Possible activation types are: 'linear' (simply transfer sum of inputs as output) 'sigmoid' (also called 'sigmoid_1') (0 or 1, threshold based) 'sigmoid_2' (-1,0,1, threshold based) user specified (passed as a CODE ref.) You can also customize threshold levels on a per-layer, or per-node basis. ** What do you think? Now I know you people are out there that are using the module... I can hear the fists hitting the keyboards in frustration. :-) Relieve some of that frustration by e-mailing me and letting me know what you think of the module and any suggestions you got. Use it, let me know what you all think. This is just a groud-up write of a neural network, no code stolen or anything else. Don't expect a classicist view of nerual networking here. I simply wrote from operating theory, not math theory. Any die-hard neural networking gurus out there? Let me know how far off I am with this code! :-) Regards, ~ Josiah Bryan, Latest Version: http://www.josiah.countystart.com/modules/get.pl?mesh:README (*) In regards to the speed claim: I only know of two neural networks in Perl, AI::NeuralNet::Mesh (this module) and AI::NeuralNet::BackProp, both of which I wrote. If someone else has written a neural-net in Perl I would love to see how or what techniques you have used. I am sorry if I have missed anybody's Perl neural-net. If I have, it was not an intentional error on my part.