Search found 73 matches

by Vasik Rajlich
Sun Mar 15, 2009 8:06 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

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

I _did_ an algorithm like that in 1983. I put it together in under 2 weeks and won the 1983 WCCC championship with it. And I _know_ how it works and how the PVs look. Actually, I'd be curious to hear more about it. How did you split the root moves? Which moves got PV searches and which ones got sco...
by Vasik Rajlich
Sun Mar 15, 2009 8:03 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

Re: Results from UCT parallelization

The so-called "root" algorithm in the paper you quoted has nothing to do with splitting at the root, as far as I can tell. In my view the similarity is because searching alternate root moves with a more open window is comparable in effect to running multiple UCT searches and relying on their random...
by Vasik Rajlich
Sat Mar 14, 2009 7:57 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

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

I _did_ an algorithm like that in 1983. I put it together in under 2 weeks and won the 1983 WCCC championship with it. And I _know_ how it works and how the PVs look. Actually, I'd be curious to hear more about it. How did you split the root moves? Which moves got PV searches and which ones got sco...
by Vasik Rajlich
Sat Mar 14, 2009 7:56 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

Re: Results from UCT parallelization

I think I should clarify this. I do believe it's useful to search alternates more deeply, and I do not have any problems comparing moves from different depths. But the question is what to do with the bounds. I mean, if you don't have an alpha value yet from your first move, what are you going to do...
by Vasik Rajlich
Sat Mar 14, 2009 7:52 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

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

The problem is that there is no good way to share the hash. Broadcasting a handful of high-depth entries is just a drop in the bucket. The algorithm needs to be designed around this issue. Pretty much, but I'd generalize it to latency in general. I have quite a few ideas on how to handle this, but ...
by Vasik Rajlich
Sat Mar 14, 2009 7:49 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

Re: Results from UCT parallelization

I've not seen you really refute Bob's point about the limits of speedups with your algorithm, yet your claims suggest a much higher strength gain than would be expected. I'm inclined to think the above result is an explanation for that, but you're the only one who knows for sure. It's also possible...
by Vasik Rajlich
Sat Mar 14, 2009 7:43 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

Re: Results from UCT parallelization

Are these figures in your view legit? I've not verified them, and they're in an academic paper, so they're almost certainly lies :) Seriously, I've not tried root parallelization yet because I had a very highly optimized version with local mutexes. I will try it though, if only because I have a clu...
by Vasik Rajlich
Sat Mar 14, 2009 7:37 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

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

Ok, I haven't worked on this, but intuitively I would think that the inherent randomness would really help with the scaling. Can't you just do X completely independent MC runs and sum up the data (somehow) at the end? Vas Look at the leaf parallelization results in the paper. It's a bit like trying...
by Vasik Rajlich
Fri Mar 13, 2009 6:29 am
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?

book authors tend to form clusters and work together, exchanging their books regularly inside their small group (...) I am not so sure this is common practice among top engine book authors ... Sure there are book exchanges and competitions, specially on rybka's forum, among amateurs who all use ryb...
by Vasik Rajlich
Fri Mar 13, 2009 6:26 am
Forum: Computer Chess Club: General Topics
Topic: Clustering etc. thread
Replies: 159
Views: 23617

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

No it isn't. Here's why. Speedup is the time for the sequential search divided by the time for the parallel search. In the first example above (and lets stop at a fixed depth to make the comparison easy) Stop right here. It makes the comparison easy, but it also makes it invalid, because you DONT H...