Search found 1141 matches

by Gian-Carlo Pascutto
Thu Mar 12, 2009 9:46 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

What's the difference between UCT with a tiny exploration term (=current top go programs) and an alpha-beta search with very heavy LMR (=current top chess programs), in terms of nodes explored? Not quite the same thing IMHO. While we are certainly reducing, we are not doing any sort of MC analysis ...
by Gian-Carlo Pascutto
Wed Mar 11, 2009 11:17 pm
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

I personally don't think there's that much potential in it, since you would always want to be searching the best evaluated (by UCB) move, not the "best out of the set of moves that we chose for this node at iteration 1 so we can have some hash table consistency". Correct. I do agree with Bob that j...
by Gian-Carlo Pascutto
Wed Mar 11, 2009 11:07 pm
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

Well it actually can if you use what GCP suggested. nodes = 1 << depth; // rough approximation based on ebf=2 prob = 1 / (1 + pow(10, -score/4)); // floating point score, see http://chessprogramming.wikispaces.com/Pawn+Advantage%2C+Win+Percentage%2C+and+ELO ucb = score + k*sqrt(log&...
by Gian-Carlo Pascutto
Wed Mar 11, 2009 10:58 pm
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

UCT is an interesting idea, (...) Extending that to chess, however, is a different issue entirely and whether it would work here or not is not a given. However, in the current context, it is irrelevant to the current discussion anyway. If you claim that, then you can certainly answer the following ...
by Gian-Carlo Pascutto
Wed Mar 11, 2009 10:22 pm
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

In a worst-case on my cluster here, you can have 560 broadcasts per hash store since every processor will need to broadcast when it updates the hash. I don't understand how you get to 560. I do agree keeping system overhead low is a problem. At some point you see the OS networking stack using CPU.....
by Gian-Carlo Pascutto
Wed Mar 11, 2009 2:45 pm
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

The lack of shared memory imposes new constraints on your search and changes the things you need to do. In particular, without shared memory, you need to split the search tree into chunks which will remain relatively stable from iteration to iteration. Vas I don't get why you guys are not doing sha...
by Gian-Carlo Pascutto
Wed Mar 11, 2009 11:27 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

4. Calculate the depth for this node, calculated from the reported depth of every move and a depth offset depending on the difference in score between each move and the best move. 5. Select and make the move that causes the current depth to be limited. If you can convert depth to nodes (you can, ap...
by Gian-Carlo Pascutto
Wed Mar 11, 2009 11:25 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Re: An idea for a new WCCC format - what do you think?

- What can happen is that the best move scores 1 pawn above move 2, thererfore searched to fe 21 ply, then fails low and move 2 becomes best but was only seached (yet) to 19 ply. This sounds so familiar :) Example with UCT (Monte Carlo Tree search): move a: 10 000 simulations 55% winning rate move ...
by Gian-Carlo Pascutto
Wed Mar 11, 2009 11:14 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23616

Results from UCT parallelization

Hi guys, as you know I spent a lot of time on Go programs and part of this involved working with UCT and parallelizing it. I think few people realize it because there's not many people working on both chess and go, but a strong implementation of UCT and alpha-beta with heavy pruning and LMR are much...
by Gian-Carlo Pascutto
Mon Mar 09, 2009 4:52 pm
Forum: Computer Chess Club: General Topics
Topic: An idea for a new WCCC format - what do you think?
Replies: 220
Views: 32268

Re: An idea for a new WCCC format - what do you think?

IIRC they collapse each entry to about 1.5 bits. most 4-5 piece files are 8 bit entries, some go tto 16 and the 6's are mainly 16 bit entries. Which gives roughly 10x size reduction before compression. Yes, but they compress a lot better, and nobody's using uncompressed 6-man tables... 5 men is alr...