Mattias, Great idea! I would like to try that out. I am not sure how it fits into our plan right now but I will try to get it in at some point. ken ... to be...
... Ok a bit more on this one. The short answer is no, I haven't experimented with phased searching on the double pole problem. Double pole is (or should be)...
Colin, Thanks for the response. ... Okay, that makes sense. ... <snip> ... Hm...where did you find that info? Did you run it with Ken's code and get solutions...
While we continue to work on ANJI, Philip and I thought it would be good to implement another benchmark to ensure code integrity and the basic functioning of...
... I haven't really seen any simple control tasks in the literature that would /require/ many hidden nodes -- maybe Ken can answer this better. This seems to...
Hi, Apologies for the dupe, but for anyone not on the GP list/group the following article is now available: "A Neural Network Controller for the Physical...
Hi Thomas, I think Karl still reads this list, he and I discussed python NEAT a while ago. His version was pretty much feature-complete as of last summer,...
Hi Derek, I think that 'private' email was to the group BTW :) Anyway I'll try and respond in full tomorrow, but 'off the top of my head' I would say the 3,6...
Yes, I would be interested in taking a look at the code. At least for now to play with and to see what I can do. I'll keep this group posted on any progress...
Give me a day or so to locate a machine that has my code on it and I'll post a link to it. I'm not sure it was exactly feature complete but you're welcome to ...
Hey Ian, Any chance we can quote something you said on the NERO forums in a press release? If so please send me an email to my personal address (on my...
This is mainly for Thomas, but if anyone else is interested please take a look. ... I've put up a release of my python-based extensible reusable etc genetic ...
Hi Derek & everyone, I think this issue of building a set of benchmarking problems is an important one and thinking about it there's potentially a whole load...
... <snip> Allow me to suggest my SailChallenge. This is fully described at http://sourceforge.net/projects/annevolve. (click the FILES link) All the code you...
... Yes, this becomes very time-consuming, and is precisely the reason why I switched over to learning a controller for a board game instead. The game domain...
... Why not just have a complete set of test cases, and use a random subset of those for each evaluation? Evolution will be noisier, but should still converge...
Thanks Karl Lewin also responded so I have a bit of Python to look through. I'll let the group know how it goes. Tom ... take a ... genetic ... contains ... ...
At 02:20 07/07/2005, you wrote:
>--- Derek James <djames@...> wrote:
>
> > While we continue to work on ANJI, Philip and I
> > thought it would be
> >...
Okay, so we've implemented Double Pole Balancing and DPB with no velocities input into the network. And ANJI is solving them fine, with both Ken's mutation...
First off, we went back and looked at Ken's code to compare it with Colin's port, and we did notice some code for randomizing the initial state of the system: ...
Actually it is generally started with the poles in the position you stated, and not from random initial states. There's no particular reason for doing it that...
... Yeh this was pointed out recently on the list. It was probably a typo but it doesn't matter all that much so long as the starting angle isn't too big. I...
... Indeed, and you might expect this given that a large number of these states must be handled in order to be able to balance the poles from any one given...
... I suspect the same problem is still present in game playing since you can only evaluate a small proportion of the possible paths in the game - usually...
Here are some nice 3D graphs of test functions for a GA toolkit. There are some very simple shapes and some more complex, now image a network were each output...
... I don't think so. States 2 and 4 are in radians, which means they're between -PI and PI (unless I'm reading the code wrong). So dividing by 0.52 yields...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
