Hi! That looks interesting. I'm curious how big a network are you intending to experiment with? (ie, # of layers, size of layers?).
Carson On Nov 11, 8:17 am, "Eric Schulte" <schulte.e...@gmail.com> wrote: > Hi Saul, > > Saul Hazledine <shaz...@gmail.com> writes: > > On Nov 10, 11:20 pm, "Eric Schulte" <schulte.e...@gmail.com> wrote: > >> Hi, > > >> Inspired by cgrand's regexp example [1], I've implemented a simple DSL > >> for specifying neural networks using Clojure data types. > > > This is really clear. The web page documentation is awesome. > > Great, Thanks, the web page is produced using Emacs Org-mode [1], the > raw text source file is in the repository at [2] > > > > > > >> Construction of this simple language involved a number of choices as to > >> where to place complexity (into the DSL or into user land), generally I > >> erred on the side of leaving complexity out of the DSL resulting in > >> potentially more complex usage, but in increased generality. I'd love > >> to hear any feedback on how this could be improved, simplified, or made > >> more idiomatic. > > > Its probably not a direction you want to go in, but when using a Lisp > > to specify the network architecture it opens up the possibility of > > using genetic programming to design a network to fit a particular > > problem. The map format doesn't fit this but, because you use > > protocols in the implementation, a list format for the DSL would allow > > genetic programming. > > >http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.100.4432 > > Thanks for the paper link, I know GA has been used to design neural > networks, but have not yet seen any specific examples (much less one in > lisp and by Koza himself). I have some experience using Clojure for > evolutionary computation, and this is something I have been considering > since the beginning of this NN implementation. Now with this paper I > have no excuse not to follow through on this impulse. > > > > > Obviously though, their DSL is nowhere as clear and clean as yours. > > :) flattery, I doubt they had access to a lisp with protocols. > > Thanks -- Eric > > > > > Saul > > Footnotes: > [1] http://orgmode.org/ > > [2] http://cs.unm.edu/~eschulte/src/neural-net/neural-net.org -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
