You all have been a great help pointing me in the proper direction. I appreciate it. Is there any documentation on the actual source code of NEAT? Something...
4197
Jason Gauci
jgmath2000
Jul 7, 2008 11:53 am
The header files are individually documented but I will compile a doxygen pdf with the next release. I will have the new release done before friday. ... I...
4198
newtwist2000
Jul 8, 2008 12:04 am
I've just read through the paper Compositional Pattern Producing Networks: A Novel Abstraction of Development - very interesting stuff. Are there additional...
4199
petar_chervenski
petar_cherve...
Jul 8, 2008 12:39 am
HyperNEAT uses CPPNs to build the neural network phenotypes (and evolves CPPNs). I see this as an extension. There are about 3 or more papers on HyperNEAT. It...
4200
Kenneth Stanley
kenstanley01
Jul 8, 2008 8:06 am
I am not sure if you are more interested in NEAT or HyperNEAT, but for what it's worth my original NEAT C++ code does come with a tutorial pdf: ...
4201
Kenneth Stanley
kenstanley01
Jul 8, 2008 8:23 am
Kirk, what is your idea for applying CPPNs to pattern recognition? I am not sure I followed exactly how the CPPN would be connected to the neural network? As...
4202
newtwist2000
Jul 8, 2008 3:02 pm
I may be over-interpreting the CPPN process, but it stimulated a few ideas. I work in the field of Computational Chemistry and my in my day to day operations...
4203
petar_chervenski
petar_cherve...
Jul 8, 2008 8:37 pm
Hi all, It is an interesting idea to see what if the phenotypes produced by the CPPNs change over their lifetime. There are two approaches to this: the CPPN...
4204
Kenneth Stanley
kenstanley01
Jul 8, 2008 10:55 pm
Peter, I think re-querying the CPPN over time is a great idea. We've talked about it in our research group but we don't have any ongoing projects with it, and...
4205
Kenneth Stanley
kenstanley01
Jul 8, 2008 11:04 pm
Kirk, I think that's a very interesting potential application area. Let me make sure I'm understanding how it would work. You are saying that you might...
4206
newtwist2000
Jul 8, 2008 11:40 pm
You've captured the basic idea. Let me expand on it a bit. One of the currently leading problems in computational chemistry or molecular modeling is coming up...
4207
petar_chervenski
petar_cherve...
Jul 9, 2008 12:15 am
I agree that all the re-querying will get expensive. In fact running such an evolution, in the matter of computational effort, that would come as close as...
4208
Kenneth Stanley
kenstanley01
Jul 10, 2008 1:54 am
It seems like at first glance the simplest thing to do with a CPPN would be to search for the shape or comformation of some protein or molecule that would...
4209
Kenneth Stanley
kenstanley01
Jul 10, 2008 1:56 am
Peter, it's great to hear that you've found our work useful. We are always happy to receive positive feedback. We also like your demos and from what I've...
4210
newtwist2000
Jul 10, 2008 3:32 am
Actually it is probably the reverse problem - search for a small molecule (or shape) that is expected to bind best to a given protein binding site which is...
4211
petar_chervenski
petar_cherve...
Jul 10, 2008 12:21 pm
Well I thought about this many times but things are complicated. First the quality of education in Bulgaria is poor compared to the U.S., Canada and most...
4212
stig.bergestad
Jul 10, 2008 12:39 pm
Hello! I was wondering if the original NEAT(c++) used/uses threshold values for node activation? And if there are any advantages to using it in general? To be...
4213
petar_chervenski
petar_cherve...
Jul 10, 2008 1:45 pm
You refer to the bias value that is usually present in the NN nodes in most models. In NEAT there is a bias input which has a constant value of 1.0. There is...
4214
Correy Allen Kowall
coolbeanskowall
Jul 10, 2008 2:20 pm
Does anyone know of a NEAT or hyperNEAT plugin for the breve simulation engine? If not, I will write one, if someone points me toward the most solid and...
4215
Cesar G. Miguel
fdital
Jul 10, 2008 6:00 pm
Peter B. and I have been doing some experiments with Breve using the NEAT-Python implementation. It's not a plugin, though. You have to set up a fitness...
4216
Correy Allen Kowall
coolbeanskowall
Jul 10, 2008 8:46 pm
Yes, I would like the code! I have been stalling on making the upgrades past breve 2.5 i.e. not really using the new python capabilities, simply because Klein...
4217
Kenneth Stanley
kenstanley01
Jul 10, 2008 8:55 pm
... I'm not sure how lightweight you need for C++ implementations, but for HyperNEAT, Jason Gauci's C++ is probably the most tested. It's available from our...
4218
Correy Allen Kowall
coolbeanskowall
Jul 10, 2008 9:39 pm
How about Gaussian activation function?...
4219
petar_chervenski
petar_cherve...
Jul 10, 2008 9:55 pm
The Gaussian function fits just as well as all other possible activation functions. OK well think of the bias like a shift of the function. The graph doesn't...
4220
Correy Allen Kowall
coolbeanskowall
Jul 10, 2008 9:55 pm
Sorry, I just realized how unclear I was. I haven't updated breve recently because of legacy code. There might be more than one NEAT biped on youtube! I was...
4222
Cesar G. Miguel
fdital
Jul 11, 2008 2:49 am
... Yes, that's Peter's work! ... I'm going to contact him in order to send you the code. He has spent more time on that than me. Although he's on this group,...
4223
Correy Allen Kowall
coolbeanskowall
Jul 11, 2008 4:59 am
Suggest to Peter that a better (smoother) fitness function could be the sum of altitudes over a series of discrete time steps. I have been experimenting with...
4224
stig.bergestad
Jul 11, 2008 6:08 pm
Ok, I am familiar with bias nodes. So basically, if I understand you correctly, bias nodes which are "usually" given as a stable input to the network is the...
4225
stig.bergestad
Jul 11, 2008 6:10 pm
Pardon me, I forgot to thank everyone for taking their time to answer. Thank you! :) -Stig...
4226
petar_chervenski
petar_cherve...
Jul 11, 2008 9:50 pm
Yes. ActivationFunction( sum(W*I) - Treshold ) is equivalent to ActivationFunction( sum(W*I) + (Wb * 1.0) ) where Wb is the weight of the connection to the...
