## Search found 30 matches

Tue Oct 13, 2020 7:25 pm
Forum: General Discussion
Topic: Wyps - a word game based on Y
Replies: 1
Views: 2038

### Re: Wyps - a word game based on Y

I've completed a fairly successful computer player, and will sketch the algorithm here. The overall algorithm is a simple static evaluation maximizer, without search. The evaluation has only a few components: the number of edges currently contacted (0-3) and the number of own-color tiles currently o...
Mon Oct 05, 2020 11:19 pm
Forum: General Discussion
Topic: Wyps - a word game based on Y
Replies: 1
Views: 2038

### Wyps - a word game based on Y

refer to https://boardgamegeek.com/boardgame/43096/wyps Its a word game, whose winning conditions are like the Game of Y - that is, connecting three sides of a triangular board. The interesting thing about this is that the winning condition has basically no relationship to the tactical game play, wh...
Sun Jan 28, 2018 1:34 am
Forum: MCTS, RL, and neural networks
Topic: there must be a better way
Replies: 1
Views: 8000

### there must be a better way

Regarding using neural nets for games. Doesn't it seem odd that we take a domain of perfectly discreet objects - black stones, white stones, squares on the board, and re-represent it as floating point numbers with thousands of parameters? It seems like there ought to be an approach that uses similar...
Wed Jan 24, 2018 9:09 pm
Forum: MCTS, RL, and neural networks
Topic: some newbie notes about Neural Net AIs
Replies: 1
Views: 3729

### Re: some newbie notes about Neural Net AIs

I've been using Hex on a 7x7 board as a test bed. My basic approach is to use the visit counts at top level of a MCTS search as the target; make the network learn what the visit count should be, and use that prediction to weight the random playout phase of new searches. I understand this is the basi...
Thu Jan 11, 2018 6:07 am
Forum: MCTS, RL, and neural networks
Topic: MCTS search visialization
Replies: 1
Views: 6588

### MCTS search visialization

Here's a visualization of how a MCTS search evolves.
Mon Jan 08, 2018 1:13 am
Forum: MCTS, RL, and neural networks
Replies: 0
Views: 12110

Leela Zero is a green-field effort to reproduce and explore the methods used by Alphago zero which I've been following. http://zero.sjeng.org/ It seems to be having some success, and it's open source. The general archecture it uses is a NN similar to the one described for Alphago zero, which guides ...
Sun Jan 07, 2018 10:29 pm
Forum: MCTS, RL, and neural networks
Topic: some newbie notes about Neural Net AIs
Replies: 1
Views: 3729

### some newbie notes about Neural Net AIs

Inspired by Alphago and Alphazero, I've started exploring NN methods for game AIs, and neural nets in general. For starters, I found that there are no good forums for general discussion, hence this Machine Learning section, and this post. I'm kind of a bottom-up guy, I like to understand the stuff I...
Wed Oct 18, 2017 12:23 am
Forum: MCTS, RL, and neural networks
Topic: MCTS AI development for "Tintas"
Replies: 0
Views: 11833

### MCTS AI development for "Tintas"

Tintas is an abstract strategy game involving collecting colored tokens. https://boardspace.net/tintas/english/tintas-rules.html If you were trying to design a game to be a perfect vehicle for MCTS, this would be it. There are a maximum of 49 moves in the game, and usually the branching factor is 6 ...
Fri Jun 30, 2017 7:19 pm
Forum: General Discussion
Replies: 3
Views: 7622

Thanks! FWIW, back in the stone age of computing I struggled to collect about 5K professional games.

One caveat about this data set; a lot of the file names include unicode characters, and a lot of the files contain UTF-8 text.
A lot of tools will behave badly because of this.
Fri Jun 30, 2017 7:15 pm
Forum: General Discussion
Topic: Insight: the best evaluator depends on the search depth
Replies: 2
Views: 6967

### Re: Insight: the best evaluator depends on the search depth

The depth is fixed, but maybe you missed the point. For the same pair of evaluators, the one tuned to win at 1-ply does much worse if the search is deeper. For the actual evaluators I was testing, the actual playing depth is 5 or 6 ply, and it just didn't work to use a 1-ply search as a proxy to sel...