TCEC Question

[M ANSARI]

I think TCEC should be looked at more for the entertainment value rather than a test to see which engine is stronger. Of course if one engine blows the other out of the water, then that can give a good idea of one engine's superiority over another. But since entertainment value is more important here, the openings used might favor one engine over another as one engine can play a certain type of position better than another ... and if those types of positions are forced by opening choice then it might skew the results in favor to one side. I think it is impressive how strong SF has become and it seems to not fall for many of the strong positional plans that Lc0 used to get easy wins with. Maybe it is just the increase in core count, but most likely it is improvement in the evaluation. Lc0 does seem to still have the annoying trolling intact and maybe not as sane in the endgame. I do think that SF has just improved more than Lc0 in the last year or so as Lc0 improvements seem to have stagnated while SF just improves methodically.

[Leo]

Why is SF doing so well against LCO in the latest TCEC?

Maybe simply because Lc0 net isn't the best around. It seems to be 75 elo lower than other big nets. Have a look here: http://talkchess.com/forum3/viewtopic.p ... 29#p849247

It is on the 30th rank.

I wonder how the TCEC LCO net made it to the superfinal. What net won the SF last season?

[MMarco]

Why is SF doing so well against LCO in the latest TCEC?

Maybe simply because Lc0 net isn't the best around. It seems to be 75 elo lower than other big nets. Have a look here: http://talkchess.com/forum3/viewtopic.p ... 29#p849247

It is on the 30th rank.

I wonder how the TCEC LCO net made it to the superfinal. What net won the SF last season?

The net that won TCEC 17 was 384x30-T60-3010 I believe (with Lc0 24). From what I read, players are allowed to submit a new net before the superfinal. Apparently, Lc0 team had several new nets that looked promising at low nodes games, but they couldn't test them thouroughly before the submission deadline. Unfortunately for them, they didn't take the best choice. Also Stockfish did really improve since the last superfinal. If I'm not mistaken, net 3972_20k played only the superfinal, T60-3010 played the qualifiers.

[MMarco]

Why is SF doing so well against LCO in the latest TCEC?

Maybe simply because Lc0 net isn't the best around. It seems to be 75 elo lower than other big nets. Have a look here: http://talkchess.com/forum3/viewtopic.p ... 29#p849247

It is on the 30th rank.

I wonder how the TCEC LCO net made it to the superfinal. What net won the SF last season?

The net that won TCEC 17 was 384x30-T60-3010 I believe (with Lc0 24). From what I read, players are allowed to submit a new net before the superfinal. Apparently, Lc0 team had several new nets that looked promising at low nodes games, but they couldn't test them thouroughly before the submission deadline. Unfortunately for them, they didn't take the best choice. Also Stockfish did really improve since the last superfinal. If I'm not mistaken, net 3972_20k played only the superfinal, T60-3010 played the qualifiers.

Maybe it is not the end of the story. The test will be rerun: http://talkchess.com/forum3/viewtopic.p ... 29#p849379

[Leo]

Much to my surprise LCO is really getting beat up. AB engines are not dead. Not even close.

[ernest]

Much to my surprise LCO is really getting beat up. AB engines are not dead. Not even close.

I do repeat my question : what about the LeelaRatio of the current TCEC configuration ?
Maybe this is a clue to the match outcome...

[Alayan]

Leela ratio is useless.

[Cornfed]

I think TCEC should be looked at more for the entertainment value rather than a test to see which engine is stronger. Of course if one engine blows the other out of the water, then that can give a good idea of one engine's superiority over another. But since entertainment value is more important here, the openings used might favor one engine over another as one engine can play a certain type of position better than another ... and if those types of positions are forced by opening choice then it might skew the results in favor to one side. I think it is impressive how strong SF has become and it seems to not fall for many of the strong positional plans that Lc0 used to get easy wins with. Maybe it is just the increase in core count, but most likely it is improvement in the evaluation. Lc0 does seem to still have the annoying trolling intact and maybe not as sane in the endgame. I do think that SF has just improved more than Lc0 in the last year or so as Lc0 improvements seem to have stagnated while SF just improves methodically.

ALL sporting events are for 'entertainment value'...but someone has to win and it comes down to whomever does whatever is needed to win...more than the other guy. Same here.

Arguably you could find a bunch of starting positions that favor Lc0 by nature...run things again and get a different result. That also plays a result- not sure how the organizers decide on the starting positions.

[SheikhYerbouti]

Could be that Leela's MLH doesn't fare well at TCEC time control. It may not interact well with the WDL head and the high draw score. Definitely noticeable compared to last SuFi, Leela's eval score is low even when she's in a winning position.

[SheikhYerbouti]

Also possible SF improved so much, that Leela's eval only appears to be conservative. In other words, SF sees the clear win for Leela before she does.