Search found 1075 matches

by Gian-Carlo Pascutto
Tue Dec 11, 2018 3:12 pm
Forum: Computer Chess Club: General Topics
Topic: Alphazero news
Replies: 248
Views: 13270

Re: Alphazero news

Matthew, I have one last question regarding the implementation details that aren't explicit in the paper. Normally, an MCTS search would do "tree reuse" from move to move, carrying forward the subtree that was actually chosen. But during the training, there is noise added to the root. If one does tr...
by Gian-Carlo Pascutto
Tue Dec 11, 2018 3:05 pm
Forum: Computer Chess Club: General Topics
Topic: CB15 bug
Replies: 16
Views: 1196

Re: CB15 bug

A small bug I notice is that it keeps deleting the .ico file of my database when I update the games.
by Gian-Carlo Pascutto
Fri Dec 07, 2018 9:25 pm
Forum: Computer Chess Club: General Topics
Topic: Alphazero news
Replies: 248
Views: 13270

Re: Alphazero news

Were other settings ever considered, notably 0.5 or parent? Yes and 0 seems to work best. Assumption being that most positions have 1 or at most a few good moves. All other moves are akin to passing or worse. In most equal-ish positions, passing will give the opponent a big advantage. I think this ...
by Gian-Carlo Pascutto
Fri Dec 07, 2018 8:26 pm
Forum: Computer Chess Club: General Topics
Topic: Alphazero news
Replies: 248
Views: 13270

Re: Alphazero news

How? One of Leela's problems is thinking theoretically drawn endgames can be won. This happens because during the training there is an intentional non-zero chance of "blundering" and in such endgames eventually a blunder will cause the side with the advantage to win. The blundering was implemented ...
by Gian-Carlo Pascutto
Fri Dec 07, 2018 7:48 pm
Forum: Computer Chess Club: General Topics
Topic: Alphazero news
Replies: 248
Views: 13270

Re: Alphazero news

During training, we do softmax sampling by visit count up to move 30. There is no value cutoff. Temperature is 1. This is a rather important difference and will explain a lot about Leela Chess Zero's endgame problems. Thanks for clarifying some of these things. The 0..1 vs -1..1 range thing is a bi...
by Gian-Carlo Pascutto
Wed Nov 21, 2018 4:11 pm
Forum: Computer Chess Club: General Topics
Topic: Previous amazing Alpha Zero position is not amazing anymore!
Replies: 26
Views: 2383

Re: Previous amazing Alpha Zero position is not amazing anymore!

Again, Losers Whiners. Let me ask you: who is the loser if AlphaZero never makes it to the public? Is it the other engine programmers? Can't be no? They should be happy about that. So who is the loser here? Just ask yourself that question. As for quibbles like Time Control Tuning and latest Dev Ver...
by Gian-Carlo Pascutto
Wed Nov 21, 2018 4:05 pm
Forum: Computer Chess Club: General Topics
Topic: My non-OC RTX 2070 is very fast with Lc0
Replies: 59
Views: 4380

Re: My non-OC RTX 2070 is very fast with Lc0

With alpha beta the reasons for search inefficiency are well known, is there a similar principle for Lc0? It's essentially the same: the core algorithm is sequential, all the parallelism is caused by speculatively evaluating nodes.The more parallelism you try to get, the more positions you will end...
by Gian-Carlo Pascutto
Wed Nov 21, 2018 8:23 am
Forum: Computer Chess Club: General Topics
Topic: My non-OC RTX 2070 is very fast with Lc0
Replies: 59
Views: 4380

Re: My non-OC RTX 2070 is very fast with Lc0

Just looking at NPS doesn't say anything about engine strength if you tweak the search settings.

Evaluating the same position over and over is really fast but also not very useful.
by Gian-Carlo Pascutto
Wed Nov 21, 2018 8:21 am
Forum: Computer Chess Club: General Topics
Topic: Previous amazing Alpha Zero position is not amazing anymore!
Replies: 26
Views: 2383

Re: Previous amazing Alpha Zero position is not amazing anymore!

Disappointingly, Google showed no interest in Commercializing AlphaZero They just announced a book with 2000 extra games to be published by New In Chess. You can go buy it and gasp at the games then. Plain and simple truth is that AlphaZero made a Mockery of the alpha-beta Computer Chess Engines ov...
by Gian-Carlo Pascutto
Mon Nov 19, 2018 5:31 pm
Forum: Computer Chess Club: General Topics
Topic: My non-OC RTX 2070 is very fast with Lc0
Replies: 59
Views: 4380

Re: My non-OC RTX 2070 is very fast with Lc0

chrisw wrote:
Mon Nov 19, 2018 4:34 pm
Useful to know. Thanks.
Do you happen to know the ratio of MCTS nodes to positions sent to GPU?
It should be 1-1 in the algorithmic sense but the NNCache (sortof like a transposition table) muddies it.