I guess even though the results tend to fall on a straight line, the calculations are still fundamentally geometric.
A 3000 could easily beat a 2000, but could a 4000 easily beat a 3000?
Sort:
congrandolor wrote:
llama wrote:
congrandolor wrote:
The 2000 is 100% better than the 1000, but the 4000 is only 33% better than the 3000.
It's utterly shameless and ignorant comments like this that keep this otherwise forgettable topic going.
It goes without saying you're wrong.
Unfortunately this topic is too ridiculous to be worth commenting on (at least regularly) for anyone who knows anything... so there's a proliferation of silly comments such as yours.
ouch, that hurts, bro. Why is my statement wrong?
I think his point is it's utterly shameless and ignorant and wrong to think that 4000 is 33% more than 3000. His point, probably is that you are way, way, way off. You aren't even close. In my opinion, the answer is 33 and a third percent. So for someone like him, it's just wrong on so many levels to even suggest that it could be 33%. Rounding it off is just shameless, I guess.
Or, it could just be that it's more a matter of which has more rating points, not how much "better" the higher rated one is.
Oct 18, 2020
0
#484
long_quach wrote:
Elroch wrote:
EndgameStudier wrote:
Not double but 60% of 60% from A to C, it's geometric, not linear, if it is transitive that way.
No, it is not geometric, exponential or linear. It is what the Elo formula for expected score says.
To be specific, here is the Elo calculation for the expected score for two players with rating RA and RB.
So if EA = 0.6 (i.e. 60%), RB-RA = 400*log10(1/0.6-1) = 70.43 points.
This means the rating difference of A and C = 70.43 + 70.43 = 140.86, so from the same formula, the expected score of A against C is 69.23%.
There is nothing obvious about this relationship between the expected scores, it just comes from the formula, and the formula has been found to be empirically quite accurate, so the rating system works.
Geometric = logarithmic.
"Geometric" is a term applied to a series where each term is a fixed multiple of the previous one. "Logarithmic" is a term most commonly used for a scale which has been transformed with a logarithm. A logarithmic transform makes the terms of a geometric series equally spaced. I think that is what you mean, but they are not interchangeable terms (and neither applies here).
Oct 18, 2020
0
#485
llama wrote:
Ebonyblackpawn wrote:
Hikaru recently said that Alpha 0 (rated 3500ish) would crush any grandmaster 10-0 in classical, blitz, or rapid/
That's an unremarkable claim for multiple reasons.
First of all, the topic is about 3000 vs 4000. GMs are far from 3000, and engines are far from 4000.
Second of all, engines have been able to beat GMs 10-0 in classical, blitz, or rapid games for years.
Thirdly, Alpha zero is years old and obsolete in spite of many casual chess fans (and other non-players) believing it to be the best chess playing entity in existence... no doubt Naka (called "Hikaru" by new ***s) only mentioned Alpha Zero because he knows his audience...
I can't see how you can try to justify the last claim. Alpha Zero appears to be quite a lot stronger than Stockfish and no other entity is.
Hardware matters of course: when Alpha Zero is sufficiently handicapped by speed, Stockfish has a chance, but it is quite a lot.
Leela Zero is highly competitive, but has never reached dominance over Stockfish (rather it is a close competitor in strength).
llama wrote:
Still, I have to correct you in so far as the Elo formula has been shown to yield inaccurate perditions when the rating disparity is as much as 1000 (or even half that much). There has been debate and articles written about what, if anything, should be done about this.
But yes, 1000 vs 2000 is the same as 2000 vs 3000.
Thanks for pointing out that the extrapolation of expectations is less accurate at extremes - I only really meant it was pretty accurate for most competitive games.
500 points is a huge rating difference (eg the venerable BCF rating system assumes 100% scores for differences of over 400!). 1000 still more so (almost zero chance on the Elo scale). Clearly at such extremes, the possibility of one player having a stroke or something becomes significant!
Oct 18, 2020
0
#486
llama wrote:
Still, I have to correct you in so far as the Elo formula has been shown to yield inaccurate perditions when the rating disparity is as much as 1000 (or even half that much). There has been debate and articles written about what, if anything, should be done about this.
But yes, 1000 vs 2000 is the same as 2000 vs 3000.
Thanks for pointing out that the extrapolation of expectations is less accurate at extremes. 500 points is a huge rating difference (the venerable BCF rating system assumes 100% scores for differences of over 400!). 1000 still more so (almost zero chance on the Elo scale). Clearly at such extremes, the possibility of one player having a stroke or something becomes significant!
Oct 18, 2020
0
#487
Elroch wrote:
llama wrote:
Ebonyblackpawn wrote:
Hikaru recently said that Alpha 0 (rated 3500ish) would crush any grandmaster 10-0 in classical, blitz, or rapid/
That's an unremarkable claim for multiple reasons.
First of all, the topic is about 3000 vs 4000. GMs are far from 3000, and engines are far from 4000.
Second of all, engines have been able to beat GMs 10-0 in classical, blitz, or rapid games for years.
Thirdly, Alpha zero is years old and obsolete in spite of many casual chess fans (and other non-players) believing it to be the best chess playing entity in existence... no doubt Naka (called "Hikaru" by new ***s) only mentioned Alpha Zero because he knows his audience...
I can't see how you can try to justify the last claim. Alpha Zero appears to be quite a lot stronger than Stockfish and no other entity is.
Hardware matters of course: when Alpha Zero is sufficiently handicapped by speed, Stockfish has a chance, but it is quite a lot.
Leela Zero is highly competitive, but has never reached dominance over Stockfish (rather it is a close competitor in strength).
llama wrote:
Still, I have to correct you in so far as the Elo formula has been shown to yield inaccurate perditions when the rating disparity is as much as 1000 (or even half that much). There has been debate and articles written about what, if anything, should be done about this.
But yes, 1000 vs 2000 is the same as 2000 vs 3000.
Thanks for pointing out that the extrapolation of expectations is less accurate at extremes - I only really meant it was pretty accurate for most competitive games.
500 points is a huge rating difference (eg the venerable BCF rating system assumes 100% scores for differences of over 400!). 1000 still more so (almost zero chance on the Elo scale). Clearly at such extremes, the possibility of one player having a stroke or something becomes significant!
A0 is ~150 rating points stronger than sf8, sf12 is ~300 rating points stronger than sf8.
Oct 18, 2020
0
#488
Elroch wrote:
I can't see how you can try to justify the last claim. Alpha Zero appears to be quite a lot stronger than Stockfish and no other entity is.
Maybe I'm unaware of more recent matches, but as far as I know the AZ vs SF match was 3 years ago and AZ's performance rating was 100 Elo higher. Since SF has continued its development over these last 3 years and gotten more than 100 Elo stronger it's reasonable to say the AZ of 2017 is no longer relevant in terms of best playing machine.
Leela is the NN engine torchbearer AFAIK, and Stockfish recently beat Leela in the TCEC Season 19 finals.
all say that a higher rated player can defeat a lower rated player but what if the higher rated player does a blunder in his game ??
Oct 18, 2020
0
#490
ooooeeeeooeeoe wrote:
all say that a higher rated player can defeat a lower rated player but what if the higher rated player does a blunder in his game ??
These sorts of questions miss the point. We're making at least two assumptions, I'll add a 3rd:
1) The player's ratings are accurate
2) The players preform as well as their ratings predict
3) Sufficiently many games are played such that the average result can be used to answer whether the higher rated player can "easily" win.
Oct 19, 2020
0
#491
llama wrote:
Elroch wrote:
I can't see how you can try to justify the last claim. Alpha Zero appears to be quite a lot stronger than Stockfish and no other entity is.
Maybe I'm unaware of more recent matches, but as far as I know the AZ vs SF match was 3 years ago and AZ's performance rating was 100 Elo higher. Since SF has continued its development over these last 3 years and gotten more than 100 Elo stronger it's reasonable to say the AZ of 2017 is no longer relevant in terms of best playing machine.
Leela is the NN engine torchbearer AFAIK, and Stockfish recently beat Leela in the TCEC Season 19 finals.
The second, 1000 game AlphaZero versus Stockfish match took place 18 months ago, not 3 years. AlphaZero won by 155 to 6 with 839 draws, indicating a 52 point difference at that time. The major neural network update to Stockfish (clearly inspired by its nemesis) has indeed yielded an 80 point increase in strength, which should be adequate to deal with the previous disadvantage. It is fair to say the new Stockfish is now a different species of fish!
Leelachess beat Stockfish in the quickplay match on chess.com before the NN upgrade but TCEC 19 was a good result for Stockfish, showing the power of its added NN.
llama wrote:
ooooeeeeooeeoe wrote:
all say that a higher rated player can defeat a lower rated player but what if the higher rated player does a blunder in his game ??
These sorts of questions miss the point. We're making at least two assumptions, I'll add a 3rd:
1) The player's ratings are accurate
2) The players preform as well as their ratings predict
3) Sufficiently many games are played such that the average result can be used to answer whether the higher rated player can "easily" win.
Hey, you didn't answer my question. You called me stupid for saying that 2000 is a 100% more than 1000, and 4000 is a 33% more than 3000. What is wrong with that?
Oct 19, 2020
0
#493
congrandolor wrote:
llama wrote:
ooooeeeeooeeoe wrote:
all say that a higher rated player can defeat a lower rated player but what if the higher rated player does a blunder in his game ??
These sorts of questions miss the point. We're making at least two assumptions, I'll add a 3rd:
1) The player's ratings are accurate
2) The players preform as well as their ratings predict
3) Sufficiently many games are played such that the average result can be used to answer whether the higher rated player can "easily" win.
Hey, you didn't answer my question. You called me stupid for saying that 2000 is a 100% more than 1000, and 4000 is a 33% more than 3000. What is wrong with that?
The rating size is completely arbitrary, what matters is the rating DIFFERENCE between the two players.
congrandolor wrote:
llama wrote:
ooooeeeeooeeoe wrote:
all say that a higher rated player can defeat a lower rated player but what if the higher rated player does a blunder in his game ??
These sorts of questions miss the point. We're making at least two assumptions, I'll add a 3rd:
1) The player's ratings are accurate
2) The players preform as well as their ratings predict
3) Sufficiently many games are played such that the average result can be used to answer whether the higher rated player can "easily" win.
Hey, you didn't answer my question. You called me stupid for saying that 2000 is a 100% more than 1000, and 4000 is a 33% more than 3000. What is wrong with that?
It's not 33 percent. It's 33 and a third percent. He may have a thing against rounding off.
guys.. i think the answer is yes because if ur a fan of Hikaru, he has done it before😂🤣😁
Hikaru is in the 3300s, for comparison, Stockfish in only 3390
But with neural networks(not talking about nnue) you can get very close
Neural networks are basically human minds amplified by the instant relays possible in a computer.
And what calculations prove white can checkmate black?
The 2000 is 100% better than the 1000, but the 4000 is only 33% better than the 3000.
It's utterly shameless and ignorant comments like this that keep this otherwise forgettable topic going.
It goes without saying you're wrong.
Unfortunately this topic is too ridiculous to be worth commenting on (at least regularly) for anyone who knows anything... so there's a proliferation of silly comments such as yours.
ouch, that hurts, bro. Why is my statement wrong?