Search found 293 matches

by mmt
Sat Oct 17, 2020 10:23 pm
Forum: Computer Chess Club: General Topics
Topic: Busted French line?
Replies: 11
Views: 700

Re: Busted French line?

I have the results of a long analysis with Oct 5 SF dev, nn-baeb9ef2d183.nnue, 450 GB of cache, most important 7-piece EGTBs. The TCEC game that SF lost was: [Event "TCEC Season 19 - Superfinal"] [Site "https://tcec-chess.com"] [Date "2020.10.01"] [Round "5.1"] [White "LCZero v0.26.3-rc1_T60.SV.JH.9...
by mmt
Sat Oct 10, 2020 10:55 pm
Forum: Computer Chess Club: General Topics
Topic: Slight change to rules of standard chess
Replies: 11
Views: 506

Re: Slight change to rules of standard chess

Things such as opening preparation down to move 35 is only going to happen on the very highest levels of the game. The impact of play at the top level is huge. That's what people watch and that's where there is at least some money involved. Look at other competitions - if Nadal vs Djokovic had just...
by mmt
Fri Oct 09, 2020 6:29 pm
Forum: Computer Chess Club: General Topics
Topic: Determining evaluation by a genetic algorithm
Replies: 5
Views: 261

Re: Determining evaluation by a genetic algorithm

I think that after many years of classical eval chess programs and with competition between them and with tuning frameworks, manually crafted features like BISHOP_PAIR_BONUS have their values pretty well tuned, so they will be hard to beat. The advantage of NNs is that they can come up with new func...
by mmt
Fri Oct 09, 2020 4:23 pm
Forum: Computer Chess Club: General Topics
Topic: 1.g4 opening is losing?
Replies: 449
Views: 54519

Re: 1.g4 opening is losing?

jp wrote:
Fri Oct 09, 2020 7:58 am
Do you feel ready now for a re-match CC game with Ovyron?
I think we should wait for the equivalent of contempt for SF NNUE or something else to prevent shuffling before trying this. But it would be fun to do another one :)
by mmt
Fri Oct 09, 2020 4:19 pm
Forum: Computer Chess Club: General Topics
Topic: 1.g4 opening is losing?
Replies: 449
Views: 54519

Re: 1.g4 opening is losing?

I have all 6 men and important 7 men. But it is possible I am missing some 7 man file that helps. All the same, I had hundreds of millions of tablebase hits (I have tablebase files on PCIE4 super high speed gumstick SSDs) I think that with enough RAM, all these TBs should be cached by the OS. What ...
by mmt
Fri Oct 09, 2020 7:14 am
Forum: Computer Chess Club: General Topics
Topic: 1.g4 opening is losing?
Replies: 449
Views: 54519

Re: 1.g4 opening is losing?

It also did to me until this fail low. I'm running it with 7-men TBs so our depth won't correspond unless you have them also. I've seen SF do this kind of a thing multiple times already. It seemingly only sees shuffling but gets it when pushed into a line where something substantial happens.
by mmt
Fri Oct 09, 2020 4:18 am
Forum: Computer Chess Club: General Topics
Topic: 1.g4 opening is losing?
Replies: 449
Views: 54519

Re: 1.g4 opening is losing?

When it hangs like that it is either a fail high or a fail low. So it found either something really good or something really bad. Yes, fail low. But I didn't have the patience anymore. Probably better to play out some games starting in this position instead until SF better handles shuffling pieces.
by mmt
Fri Oct 09, 2020 3:27 am
Forum: Computer Chess Club: General Topics
Topic: 1.g4 opening is losing?
Replies: 449
Views: 54519

Re: 1.g4 opening is losing?

Something unusual: the eval was -1.26 from depth 21 to depth 82 and it took just 34 minutes to get there. But now it's been searching depth 83 for a very long time: Just to update I gave up on waiting after a couple of days but I will re-evaluate once there is a newer version of SF that behaves dif...
by mmt
Wed Oct 07, 2020 10:33 pm
Forum: Computer Chess Club: General Topics
Topic: A Crossroad in Computer Chess; Or Desperate Flailing for Relevance
Replies: 152
Views: 9235

Re: A Crossroad in Computer Chess; Or Desperate Flailing for Relevance

Actually, one of the advantages of the ReLU activation is that it maps all the negative inputs to the same zero output, so if an input was negative and remains negative after an incremental update, then the output remains zero, and fewer recalculations are needed in further layers. Zeros would help...