Page 1 of 3

It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Mon Jun 24, 2019 12:21 pm
by Nay Lin Tun
With rapid improvement of Allie, she stepped into the big 3 terriority of TCEC.

And TCEC S16 final would be any of these big three!!

Most possible finalists for S16 are

1. Lco vs Stockfish or

2. Lco vs Allie!!

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Tue Jun 25, 2019 2:16 pm
by Nay Lin Tun
Now Top 3 engines tied in TCEC!

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Tue Jun 25, 2019 4:35 pm
by mwyoung
Nay Lin Tun wrote:
Tue Jun 25, 2019 2:16 pm
Now Top 3 engines tied in TCEC!
The best engines are all very active in improving.
And all the best engines are free.
Komodo is the only commercial engine at the top that is still active.
And why must we watch Houdini 6 from 2017 lose again. Tcec needs a rule to not play engines season after season that are no longer active.

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Tue Jun 25, 2019 4:56 pm
by Laskos
Damn, this AllieStein could mess up the results in a 5-engine round-robin. It is weaker head to head against Lc0, and weaker against SF than Lc0 is against SF head to head. If Lc0 and SF qualify to the superfinal, the outcome will be a clear Lc0 win. But if AllieStein qualifies, I am less certain about the winner.

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Tue Jun 25, 2019 5:24 pm
by mwyoung
Laskos wrote:
Tue Jun 25, 2019 4:56 pm
Damn, this AllieStein could mess up the results in a 5-engine round-robin. It is weaker head to head against Lc0, and weaker against SF than Lc0 is against SF head to head. If Lc0 and SF qualify to the superfinal, the outcome will be a clear Lc0 win. But if AllieStein qualifies, I am less certain about the winner.
Right now if Lc0 is in the final. Lc0 will win again. Lc0 is a formidable match player. Lc0 may not crush you, but it refuses to lose.

In TCEC right now Lc0 has played 61 games without a loss. And this is no Fluke!

In my latest test match Lc0 played Stockfish 52 games without a LOSS!!

DESKTOP-CORSAIR, 5m/40+5m/40+5m/40 0

1 Lc0 v0.21.2 +20 +3/=49/-0 52.88% 27.5/52
2 Stockfish 210619 64 POPCNT -20 +0/=49/-3 47.12% 24.5/52

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Tue Jun 25, 2019 9:25 pm
by Ovyron
What is the deal with this "Allie"? "AllieStein"? What's its story? Have the rating lists become so irrelevant that they are missing the two strongest machines in the top three?

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Tue Jun 25, 2019 9:49 pm
by mwyoung
Ovyron wrote:
Tue Jun 25, 2019 9:25 pm
What is the deal with this "Allie"? "AllieStein"? What's its story? Have the rating lists become so irrelevant that they are missing the two strongest machines in the top three?
Not irrelevant, but too slow. In the day of the NN engines. Improvements came fast and furious. It is non stop just for me to keep testing the best 2 engines. Before more improvements come and change the results.

Before NN engines Stockfish had the best method for improving the Stockfish engine. But this still required humans to improve the code. Now with NN engines. The improvement come much faster with self training. And using other databases of games already played training Alliestein.

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Tue Jun 25, 2019 11:03 pm
by Leo
Unfortunately the only thing that interests me in the TCEC is to see if lcz can increase its margin of victory.

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Wed Jun 26, 2019 9:51 am
by whereagles
SF still kicks in round-robins, but competition is getting stronger.

Re: It looks like the big 3 in TCEC are LC0, SF and Allie now!

Posted: Wed Jun 26, 2019 9:42 pm
by Ovyron
mwyoung wrote:
Tue Jun 25, 2019 9:49 pm
Not irrelevant, but too slow. In the day of the NN engines. Improvements came fast and furious. It is non stop just for me to keep testing the best 2 engines. Before more improvements come and change the results.
Again, I'd like to question why are humans and machines dealt with differently. Humans have chess ratings, and their ratings is some kind of shadow that approaches their strength in a give time period, and that is good enough.

With engines, they want a fixed strength opponent, for some reason, like a human that has its memories wiped after each game so it's the same person before their previous game.

Just, test the latest thing, and upgrade as new versions appear, then compute the rating from the last 1000 games, or something. That way, games that are from old versions we don't care about leave the sample size eventually, but we have an idea of the strength of the machine.

NNs being missing from the rating lists makes them irrelevant, just like the best engines being absent from the WCCC makes it irrelevant.