Search found 29 matches

by luke_g
Sun Nov 09, 2008 10:29 am
Forum: General Discussion
Topic: Tree exploration with uncertainty
Replies: 4
Views: 3871

Re: Tree exploration with uncertainty

Perfect, thank you!
by luke_g
Wed Nov 05, 2008 9:16 am
Forum: General Discussion
Topic: Tree exploration with uncertainty
Replies: 4
Views: 3871

Tree exploration with uncertainty

I've been thinking about (best-first) exploration of trees, taking into account uncertainties in the evaluation function. Basically, choosing a node to expand based not only on its evaluation, but also the reliability or uncertainty of the evaluation. A couple examples: One is in UCT where less-expl...
by luke_g
Thu Aug 21, 2008 3:32 am
Forum: General Discussion
Topic: Al Zimmermann's Programming Contests
Replies: 16
Views: 10944

Re: Al Zimmermann's Programming Contests

I am stuck at this point, can't solve n=8. Just one point short. There are only small fractions left on my n=9 to n=14 scores. Have you had time to work on the contest Rémi or are you done? Luke? Can anyone beat Hugo and Hermann? :) They seem to be doing very well, I hope they will share their algo...
by luke_g
Thu Jul 31, 2008 9:04 am
Forum: General Discussion
Topic: Al Zimmermann's Programming Contests
Replies: 16
Views: 10944

Re: Al Zimmermann's Programming Contests

The contest has started!
by luke_g
Wed Jun 11, 2008 3:45 am
Forum: General Discussion
Topic: Al Zimmermann's Programming Contests
Replies: 16
Views: 10944

Re: Al Zimmermann's Programming Contests

Looking forward to it! I've never done a programming contest before. I also gave SameGame my top choice.

I hope primal sums isn't picked, though, as I don't think that would be nearly as interesting as it might sound...
by luke_g
Mon May 26, 2008 11:24 pm
Forum: General Discussion
Topic: Sokoban
Replies: 6
Views: 5789

Sokoban

Here's an interesting game I ran into which is hard for computers. It's a 1-player puzzle game, and humans can solve many problems that computers cannot. There doesn't seem to be a lot published recently (in English, at least), but I thought it was a good read. Here's a PhD thesis about the game and...
by luke_g
Mon May 26, 2008 8:42 pm
Forum: General Discussion
Topic: numbrosia
Replies: 47
Views: 29543

Re: numbrosia

Looking at your code Remi, not quite sure of the difference between a breadth search and a beam search. I think "beam search" is basically a breadth-first search, but discarding nodes beyond a limit. I am not sure how these mod tables are designed Luke. There is something to be said for mechanics a...
by luke_g
Wed May 21, 2008 11:40 pm
Forum: General Discussion
Topic: numbrosia
Replies: 47
Views: 29543

Re: numbrosia

I've uploaded my code. It's messy (in particular I unrolled many loops to help performance), but maybe it'll be useful :) http://www.luke-g.com/numbrosia/ You will need the 5 .rar files to run the program--these uncompress to 2+GB and contain the lookup tables. I don't think I'll upload the code use...
by luke_g
Wed May 14, 2008 2:54 am
Forum: General Discussion
Topic: numbrosia
Replies: 47
Views: 29543

Re: numbrosia

Isn't a 2^25-entry table of minimum number of rotations indexed by tile parities enough ? I believe you suggested this idea yourself some time ago. Nope, although it does help a lot in practice, it still comes short on a lot of simple positions. In fact I'm using a 4^16-entry table for mod 4, but t...
by luke_g
Mon May 12, 2008 3:54 am
Forum: MCTS, RL, and neural networks
Topic: Applying go methods to other games
Replies: 22
Views: 20876

Re: Applying go methods to other games

Oh, so actually it looks like the website's scoring system penalizes you more than the scoring system presented in the paper. In the paper it says that you lose (n-2)^2 points for each color (where n is the number of blocks of that color remaining), while on the website you lose (N-2)^2 points (wher...