Peter Berger wrote:This is much less obvious to me than it is to you.
If i had to design a tournament strategy for Junior in an imminent ICGA event in case Stockfish participated I'd think of a draw in our game as a *very* desirable result, and in case this was achieved we would have to see who is able to score better against the opposition. I'd try to prepare a very drawish opening book against Stockfish and try as hard as I can to win every other game.
Stockfish would likely draw some other game too against one of the weaker opponents ( just as it doesn't try too hard to win in general) , and so it could be outdone.
Would Stockfish still be the favourite to win the tournament? It would be indeed IMHO. But it still is a chess entity with apparemt weaknesses despite its enourmous playing strength, and as it isn't aware of them ( yet ??) - they could also be used against it IMHO.
Anyway - as all of this is clearly none of my business I'll not engage in this dicussion further
Peter over and out.
If there are 11 engines, 10 games each engine, and say 5 engines 200 Elo points weaker than SF at LTC and equal hardware, 4 engines 400 Elo weaker, then taking into account the hardware fluctuations and book fluctuations, SF has LESS than 50% to win WCCC.
Draws with so few games played are possible. Stockfish really would win relatively easy though. There are a number of fantastically strong opening books fine tuned.
Exploiting Stockfish isnt that easy man. Stronger is stronger. There is no magic. If I entered my 20 core with my Stockfish and opening book I would win this tournament.... there isn't a shred of doubt. patzers are patzers. It would be like taking candy from a baby.
Peter Berger wrote:This is much less obvious to me than it is to you.
If i had to design a tournament strategy for Junior in an imminent ICGA event in case Stockfish participated I'd think of a draw in our game as a *very* desirable result, and in case this was achieved we would have to see who is able to score better against the opposition. I'd try to prepare a very drawish opening book against Stockfish and try as hard as I can to win every other game.
Stockfish would likely draw some other game too against one of the weaker opponents ( just as it doesn't try too hard to win in general) , and so it could be outdone.
Would Stockfish still be the favourite to win the tournament? It would be indeed IMHO. But it still is a chess entity with apparemt weaknesses despite its enourmous playing strength, and as it isn't aware of them ( yet ??) - they could also be used against it IMHO.
Anyway - as all of this is clearly none of my business I'll not engage in this dicussion further
Peter over and out.
If there are 11 engines, 10 games each engine, and say 5 engines 200 Elo points weaker than SF at LTC and equal hardware, 4 engines 400 Elo weaker, then taking into account the hardware fluctuations and book fluctuations, SF has LESS than 50% to win WCCC.
This is totally wrong, probabilities of each individual engine winning tournament are not additive, so even if the sum of all other 10 engines tournament winning probabilities are above 50%, SF chance of winning tournament is still well beyond 80%.
Peter Berger wrote:This is much less obvious to me than it is to you.
If i had to design a tournament strategy for Junior in an imminent ICGA event in case Stockfish participated I'd think of a draw in our game as a *very* desirable result, and in case this was achieved we would have to see who is able to score better against the opposition. I'd try to prepare a very drawish opening book against Stockfish and try as hard as I can to win every other game.
Stockfish would likely draw some other game too against one of the weaker opponents ( just as it doesn't try too hard to win in general) , and so it could be outdone.
Would Stockfish still be the favourite to win the tournament? It would be indeed IMHO. But it still is a chess entity with apparemt weaknesses despite its enourmous playing strength, and as it isn't aware of them ( yet ??) - they could also be used against it IMHO.
Anyway - as all of this is clearly none of my business I'll not engage in this dicussion further
Peter over and out.
If there are 11 engines, 10 games each engine, and say 5 engines 200 Elo points weaker than SF at LTC and equal hardware, 4 engines 400 Elo weaker, then taking into account the hardware fluctuations and book fluctuations, SF has LESS than 50% to win WCCC.
This is totally wrong, probabilities of each individual engine winning tournament are not additive, so even if the sum of all other 10 engines tournament winning probabilities are above 50%, SF chance of winning tournament is still well beyond 80%.
Sure I did not add them. Do simulations with variability on hardware (say in +/- 100 Elo range, even larger for things like Jonny on 2400 cores) and books (say in +/- 50 Elo range).
Laskos wrote:Sure I did not add them. Do simulations with variability on hardware (say in +/- 100 Elo range, even larger for things like Jonny on 2400 cores) and books (say in +/- 50 Elo range).
Variability of +/- 100 Elo range in hardware is highly improbable. If SF showed it would be at least on TCEC hardware, assuming diminishing returns there should be an engine within 200Elo bracket from SF that could actually scale well on 256 cores or more to get +100 Elo which is more than highly improbable. I believe Jonny's number of cores is nothing but a gimmick and its actual strength increase compared to Jonny in TCEC is not even 50Elo.
Also there is plenty of good books available so even +/- 50 Elo range on books is improbable.
Your assumptions are simply grossly exaggerated.
Laskos wrote:Sure I did not add them. Do simulations with variability on hardware (say in +/- 100 Elo range, even larger for things like Jonny on 2400 cores) and books (say in +/- 50 Elo range).
Variability of +/- 100 Elo range in hardware is highly improbable. If SF showed it would be at least on TCEC hardware, assuming diminishing returns there should be an engine within 200Elo bracket from SF that could actually scale well on 256 cores or more to get +100 Elo which is more than highly improbable. I believe Jonny's number of cores is nothing but a gimmick and its actual strength increase compared to Jonny in TCEC is not even 50Elo.
Also there is plenty of good books available so even +/- 50 Elo range on books is improbable.
Your assumptions are simply grossly exaggerated.
You obviously underestimate all these variables. Recently I tested some not very good and not too bad bin books with SF at 1'+1'', the difference between those better and those not so good is in excess of 60 Elo points, and I didn't test those excellent CTG books. If SF would participate with its own "book.bin", then it might have a pretty bumpy road in WCCC.
Laskos wrote:You obviously underestimate all these variables. Recently I tested some not very good and not too bad bin books with SF at 1'+1'', the difference between those better and those not so good is in excess of 60 Elo points, and I didn't test those excellent CTG books. If SF would participate with its own "book.bin", then it might have a pretty bumpy road in WCCC.
The books can make a lot of difference that is certainly true, but what I guess you are missing is why would SF ever compete (in already totally hypothetical case it decided to participate) in a way that intentionally weakens it?
There is plenty of ppl who could and would immediately provide strong book to SF team or operator.
It is totally unrealistic expecting for SF to participate with its own crappy book.
Assuming conditions as such is just for theoretical discussion with little to no practical merit.
Laskos wrote:You obviously underestimate all these variables. Recently I tested some not very good and not too bad bin books with SF at 1'+1'', the difference between those better and those not so good is in excess of 60 Elo points, and I didn't test those excellent CTG books. If SF would participate with its own "book.bin", then it might have a pretty bumpy road in WCCC.
The books can make a lot of difference that is certainly true, but what I guess you are missing is why would SF ever compete (in already totally hypothetical case it decided to participate) in a way that intentionally weakens it?
There is plenty of ppl who could and would immediately provide strong book to SF team or operator.
It is totally unrealistic expecting for SF to participate with its own crappy book.
Assuming conditions as such is just for theoretical discussion with little to no practical merit.
Books are hugely important and there are two very strong ones that I can think of. But honestly Komodo on say a 16 core would destroy the opponents in this ICGA tournament just like Stockfish would. Even Rybka 4.1 could probably win. It is still stronger than all of the opponents (say 100 Elo), and Rybka cluster of course exists.
Laskos wrote:You obviously underestimate all these variables. Recently I tested some not very good and not too bad bin books with SF at 1'+1'', the difference between those better and those not so good is in excess of 60 Elo points, and I didn't test those excellent CTG books. If SF would participate with its own "book.bin", then it might have a pretty bumpy road in WCCC.
The books can make a lot of difference that is certainly true, but what I guess you are missing is why would SF ever compete (in already totally hypothetical case it decided to participate) in a way that intentionally weakens it?
There is plenty of ppl who could and would immediately provide strong book to SF team or operator.
It is totally unrealistic expecting for SF to participate with its own crappy book.
Assuming conditions as such is just for theoretical discussion with little to no practical merit.
Books are hugely important and there are two very strong ones that I can think of. But honestly Komodo on say a 16 core would destroy the opponents in this ICGA tournament just like Stockfish would. Even Rybka 4.1 could probably win. It is still stronger than all of the opponents (say 100 Elo), and Rybka cluster of course exists.
You and Milos seem to not realize what 100 and 200 Elo points difference means. 200 means that in a direct match it's about 55% win, 40% draw, 5% loss.
Anyway, I did simulations, having 5 engines 200 Elo points weaker than Stockfish, with a total of 11 participating engines, 10 games each engine:
Without hardware and book variables: Stockfish has 60% to win WCCC.
With hardware and book variables: Stockfish has about 50% to win WCCC.
lucasart wrote:even without book/tb and running on a cell phone, sf (or komodo) could win this pathetic event…
Sure "could", with some probability (I guess below 30%).
Generally, take it this way: there are only 10 games. If 5 games are against opponents which are only 200 Elo weaker (say the rest of the games are against gopo), then the number of expected (draws + losses) is about 2-3, and 3 draws or 1 loss + 1 draw might place SF on the second spot.
Junior team speculates very well WCCC conditions, they often have book wins and avoid draws against weaker opposition, by both book and some sort of contempt scheme.
Then, if Jonny comes with substantially improved parallelization on 2400 cores, even the favorite is unclear.