Page 8 of 18

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 4:24 am
by lkaufman
Uri Blass wrote:
Fri Jul 05, 2019 3:06 am
lkaufman wrote:
Fri Jul 05, 2019 2:17 am
Vinvin wrote:
Thu Jul 04, 2019 11:30 pm
carldaman wrote:
Thu Jul 04, 2019 10:46 pm
CM9000 shows up on the CCRL 40/40 list, you can check it out.

Code: Select all

Chessmaster 11 Aiglos II	2700	 
Chessmaster 11	2664	 
CM9000 Default	2639	
CM10th Default	2632	
Thanks, I wrongly assumed that if it wasn't on the 40/4 list, it wouldn't be on the 40/40. So if I'm right about adding 100 for FIDE comparisons at standard time limit, only the world's top twenty would be favored against it. Of course we didn't specify the processor or opening book, these things matter. Anyway it is about a class weaker than the Arasan 18.0 which I've been using in my handicap matches. Probably this means that Komodo vs CM 9000 both on one thread would be a close match at knight odds at some rapid (not blitz) time control. But it doesn't follow that we could give knight odds at that time control to world number 20 human.
I think that the only way to know is simply by playing.
It may be interesting to have knight odd rating list for chess engines.

It is possible that there are engines that are better in converting a knight advantage to a win and not better in normal chess.

CM is a class weaker than arasan18 in normal chess.
We do not know if it is a class weaker than arasan18 in knight odd games.
In my experience any diffences between how engines play with handicaps or with certain openings etc. are usually fairly small. The higher rated engine always does better, unless they are very close. It is possible that CM has extensive knowledge about how to play when a piece ahead, but extremely unlikely. One exception is that engines have trouble giving weaker or crippled versions of themselves as much handicap as you would expect based on ratings.

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 4:38 am
by Chessqueen
lkaufman wrote:
Fri Jul 05, 2019 4:24 am
Uri Blass wrote:
Fri Jul 05, 2019 3:06 am
lkaufman wrote:
Fri Jul 05, 2019 2:17 am
Vinvin wrote:
Thu Jul 04, 2019 11:30 pm
carldaman wrote:
Thu Jul 04, 2019 10:46 pm
CM9000 shows up on the CCRL 40/40 list, you can check it out.

Code: Select all

Chessmaster 11 Aiglos II	2700	 
Chessmaster 11	2664	 
CM9000 Default	2639	
CM10th Default	2632	
Thanks, I wrongly assumed that if it wasn't on the 40/4 list, it wouldn't be on the 40/40. So if I'm right about adding 100 for FIDE comparisons at standard time limit, only the world's top twenty would be favored against it. Of course we didn't specify the processor or opening book, these things matter. Anyway it is about a class weaker than the Arasan 18.0 which I've been using in my handicap matches. Probably this means that Komodo vs CM 9000 both on one thread would be a close match at knight odds at some rapid (not blitz) time control. But it doesn't follow that we could give knight odds at that time control to world number 20 human.
I think that the only way to know is simply by playing.
It may be interesting to have knight odd rating list for chess engines.

It is possible that there are engines that are better in converting a knight advantage to a win and not better in normal chess.

CM is a class weaker than arasan18 in normal chess.
We do not know if it is a class weaker than arasan18 in knight odd games.
In my experience any diffences between how engines play with handicaps or with certain openings etc. are usually fairly small. The higher rated engine always does better, unless they are very close. It is possible that CM has extensive knowledge about how to play when a piece ahead, but extremely unlikely. One exception is that engines have trouble giving weaker or crippled versions of themselves as much handicap as you would expect based on ratings.
I might be wrong to assume that Komodo is the program that performs best giving handicap odds, Am I correct, or there is another program like LC0 ?

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 9:44 am
by S.Taylor
Chessqueen wrote:
Fri Jun 28, 2019 4:47 am
There should be more computer with 2 pawns or Knight vs pawn odds that will bring back the challenge of Humans GMs vs top Engines. :shock:
I believe that top GMs like Nakamura or MVL can bring back the excitement of Computer Vs Humans.
Because, when does small material count?
Mainly at the end of the end game, when there is almost left on the board. It's off-putting to always have to wait around.
What about earlier in the game?
Well, there can be so many tactics and maneuvers, in fact that's what chess is all about!
So what's so special about slightly imballanced material distribution? It's almost just like an alternative variant of standard chess.
The main thing earlier on in the game is how you are going to put more of your material into use. Players are constantly prepared to sacrifice material for more activity anyway. So why complicate it with a pawn difference in overall value with slightly altered position from the start? It's only a variant!

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 4:14 pm
by lkaufman
Chessqueen wrote:
Fri Jul 05, 2019 4:38 am
lkaufman wrote:
Fri Jul 05, 2019 4:24 am
Uri Blass wrote:
Fri Jul 05, 2019 3:06 am
lkaufman wrote:
Fri Jul 05, 2019 2:17 am
Vinvin wrote:
Thu Jul 04, 2019 11:30 pm
carldaman wrote:
Thu Jul 04, 2019 10:46 pm
CM9000 shows up on the CCRL 40/40 list, you can check it out.

Code: Select all

Chessmaster 11 Aiglos II	2700	 
Chessmaster 11	2664	 
CM9000 Default	2639	
CM10th Default	2632	
Thanks, I wrongly assumed that if it wasn't on the 40/4 list, it wouldn't be on the 40/40. So if I'm right about adding 100 for FIDE comparisons at standard time limit, only the world's top twenty would be favored against it. Of course we didn't specify the processor or opening book, these things matter. Anyway it is about a class weaker than the Arasan 18.0 which I've been using in my handicap matches. Probably this means that Komodo vs CM 9000 both on one thread would be a close match at knight odds at some rapid (not blitz) time control. But it doesn't follow that we could give knight odds at that time control to world number 20 human.
I think that the only way to know is simply by playing.
It may be interesting to have knight odd rating list for chess engines.

It is possible that there are engines that are better in converting a knight advantage to a win and not better in normal chess.

CM is a class weaker than arasan18 in normal chess.
We do not know if it is a class weaker than arasan18 in knight odd games.
In my experience any diffences between how engines play with handicaps or with certain openings etc. are usually fairly small. The higher rated engine always does better, unless they are very close. It is possible that CM has extensive knowledge about how to play when a piece ahead, but extremely unlikely. One exception is that engines have trouble giving weaker or crippled versions of themselves as much handicap as you would expect based on ratings.
I might be wrong to assume that Komodo is the program that performs best giving handicap odds, Am I correct, or there is another program like LC0 ?
I ran Stockfish 10 against Arasan 18 at knight odds with same Contempt and time limit as Komodo. Stockfish lost by 14 elo less than Komodo did (57 elo vs 71), with the margin of error on each test a bit over 30. But Stockfish is rated 80 elo above Komodo 13.01 CCRL 40/4 1 cpu and 110 above on CEGT 40/4 1 cpu. So Komodo is much better than Stockfish at giving knight odds compared to its strength in even games, which is logical as it has special code about winning material advantages. As for Lc0, their first series of networks, culminating in 11248, is very good at giving handicaps; it scored about 70% in blitz games giving handicaps averaging around knight odds to GM Naroditsky, an excellent blitz player. However more recent networks are just awful at giving knight odds, they just make blunders. Lc0 just plays awful chess when the winning expectancy gets down below 2% or so. For the old networks that only happens around queen odds, but for the new ones it happens even below knight odds.

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 7:40 pm
by Uri Blass
lkaufman wrote:
Fri Jul 05, 2019 4:14 pm
Chessqueen wrote:
Fri Jul 05, 2019 4:38 am
lkaufman wrote:
Fri Jul 05, 2019 4:24 am
Uri Blass wrote:
Fri Jul 05, 2019 3:06 am
lkaufman wrote:
Fri Jul 05, 2019 2:17 am
Vinvin wrote:
Thu Jul 04, 2019 11:30 pm
carldaman wrote:
Thu Jul 04, 2019 10:46 pm
CM9000 shows up on the CCRL 40/40 list, you can check it out.

Code: Select all

Chessmaster 11 Aiglos II	2700	 
Chessmaster 11	2664	 
CM9000 Default	2639	
CM10th Default	2632	
Thanks, I wrongly assumed that if it wasn't on the 40/4 list, it wouldn't be on the 40/40. So if I'm right about adding 100 for FIDE comparisons at standard time limit, only the world's top twenty would be favored against it. Of course we didn't specify the processor or opening book, these things matter. Anyway it is about a class weaker than the Arasan 18.0 which I've been using in my handicap matches. Probably this means that Komodo vs CM 9000 both on one thread would be a close match at knight odds at some rapid (not blitz) time control. But it doesn't follow that we could give knight odds at that time control to world number 20 human.
I think that the only way to know is simply by playing.
It may be interesting to have knight odd rating list for chess engines.

It is possible that there are engines that are better in converting a knight advantage to a win and not better in normal chess.

CM is a class weaker than arasan18 in normal chess.
We do not know if it is a class weaker than arasan18 in knight odd games.
In my experience any diffences between how engines play with handicaps or with certain openings etc. are usually fairly small. The higher rated engine always does better, unless they are very close. It is possible that CM has extensive knowledge about how to play when a piece ahead, but extremely unlikely. One exception is that engines have trouble giving weaker or crippled versions of themselves as much handicap as you would expect based on ratings.
I might be wrong to assume that Komodo is the program that performs best giving handicap odds, Am I correct, or there is another program like LC0 ?
I ran Stockfish 10 against Arasan 18 at knight odds with same Contempt and time limit as Komodo. Stockfish lost by 14 elo less than Komodo did (57 elo vs 71), with the margin of error on each test a bit over 30. But Stockfish is rated 80 elo above Komodo 13.01 CCRL 40/4 1 cpu and 110 above on CEGT 40/4 1 cpu. So Komodo is much better than Stockfish at giving knight odds compared to its strength in even games, which is logical as it has special code about winning material advantages. As for Lc0, their first series of networks, culminating in 11248, is very good at giving handicaps; it scored about 70% in blitz games giving handicaps averaging around knight odds to GM Naroditsky, an excellent blitz player. However more recent networks are just awful at giving knight odds, they just make blunders. Lc0 just plays awful chess when the winning expectancy gets down below 2% or so. For the old networks that only happens around queen odds, but for the new ones it happens even below knight odds.
I think that the same elo advantage in normal games is not translated to the same elo advantage with knight odds.

The right comparison is of programs with the same playing strength
so you can compare stockfish10 1 cpu with komodo 4 cpu that are almost the same strength and see if komodo does better with knight odds and how much better.

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 7:48 pm
by Chessqueen
Uri Blass wrote:
Fri Jul 05, 2019 7:40 pm
lkaufman wrote:
Fri Jul 05, 2019 4:14 pm
Chessqueen wrote:
Fri Jul 05, 2019 4:38 am
lkaufman wrote:
Fri Jul 05, 2019 4:24 am
Uri Blass wrote:
Fri Jul 05, 2019 3:06 am
lkaufman wrote:
Fri Jul 05, 2019 2:17 am
Vinvin wrote:
Thu Jul 04, 2019 11:30 pm
carldaman wrote:
Thu Jul 04, 2019 10:46 pm
CM9000 shows up on the CCRL 40/40 list, you can check it out.

Code: Select all

Chessmaster 11 Aiglos II	2700	 
Chessmaster 11	2664	 
CM9000 Default	2639	
CM10th Default	2632	
Thanks, I wrongly assumed that if it wasn't on the 40/4 list, it wouldn't be on the 40/40. So if I'm right about adding 100 for FIDE comparisons at standard time limit, only the world's top twenty would be favored against it. Of course we didn't specify the processor or opening book, these things matter. Anyway it is about a class weaker than the Arasan 18.0 which I've been using in my handicap matches. Probably this means that Komodo vs CM 9000 both on one thread would be a close match at knight odds at some rapid (not blitz) time control. But it doesn't follow that we could give knight odds at that time control to world number 20 human.
I think that the only way to know is simply by playing.
It may be interesting to have knight odd rating list for chess engines.

It is possible that there are engines that are better in converting a knight advantage to a win and not better in normal chess.

CM is a class weaker than arasan18 in normal chess.
We do not know if it is a class weaker than arasan18 in knight odd games.
In my experience any diffences between how engines play with handicaps or with certain openings etc. are usually fairly small. The higher rated engine always does better, unless they are very close. It is possible that CM has extensive knowledge about how to play when a piece ahead, but extremely unlikely. One exception is that engines have trouble giving weaker or crippled versions of themselves as much handicap as you would expect based on ratings.
I might be wrong to assume that Komodo is the program that performs best giving handicap odds, Am I correct, or there is another program like LC0 ?
I ran Stockfish 10 against Arasan 18 at knight odds with same Contempt and time limit as Komodo. Stockfish lost by 14 elo less than Komodo did (57 elo vs 71), with the margin of error on each test a bit over 30. But Stockfish is rated 80 elo above Komodo 13.01 CCRL 40/4 1 cpu and 110 above on CEGT 40/4 1 cpu. So Komodo is much better than Stockfish at giving knight odds compared to its strength in even games, which is logical as it has special code about winning material advantages. As for Lc0, their first series of networks, culminating in 11248, is very good at giving handicaps; it scored about 70% in blitz games giving handicaps averaging around knight odds to GM Naroditsky, an excellent blitz player. However more recent networks are just awful at giving knight odds, they just make blunders. Lc0 just plays awful chess when the winning expectancy gets down below 2% or so. For the old networks that only happens around queen odds, but for the new ones it happens even below knight odds.
I think that the same elo advantage in normal games is not translated to the same elo advantage with knight odds.

The right comparison is of programs with the same playing strength
so you can compare Stockfish10 1 cpu with Komodo 4 cpu that are almost the same strength and see if Komodo does better with knight odds and how much better.
I agree with Uri, you need to compare both programs when they become equal in strength even if you have to increase the number of cpu on Komodo and decrease it to Stockfish. I also wonder if somebody can test LCO against Arasan 18, the only problem would be to find out what GPU and hardware is comparable with Komodo latest odds.

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 11:31 pm
by Chessqueen
carldaman wrote:
Thu Jul 04, 2019 10:46 pm
CM9000 shows up on the CCRL 40/40 list, you can check it out.
Since the testing of Chessmaster 11 Aiglos II 2700 was done, CCRL had an adjustment in rating, therefore, that rating could be much lower if Chessmaster 11 Aiglos II 2700 is tested again Nowadays. I believe that Arasan 18.0 is way stronger than Chessmaster 11 Aiglos II 2700 :mrgreen:

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Fri Jul 05, 2019 11:36 pm
by Graham Banks
I might be able to play a top engine with rook odds against the following player - https://en.wikipedia.org/wiki/Martin_Dreyer.

In a six game match, how do you think he'd fare?

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Sat Jul 06, 2019 12:47 am
by Chessqueen
Graham Banks wrote:
Fri Jul 05, 2019 11:36 pm
I might be able to play a top engine with rook odds against the following player - https://en.wikipedia.org/wiki/Martin_Dreyer.

In a six game match, how do you think he'd fare?
In the year 1993 he was rated 2345, but depending how active he still is, a players over time lose somewhere between 50 to 100 rating points and in 26 years it could be very different than was he used to be. I would say based on that point of view that he might get one or two win versus Komodo with a knight odds, but with a Rook odds Komodo might still get a few draws if he is still around 2250 to 2275 Elo and that also depends on how many cpu you will use with Komodo :mrgreen:

Re: Why there is no interest in Computer with odds Vs Humans match?

Posted: Sat Jul 06, 2019 2:36 am
by Chessqueen
Chessqueen wrote:
Sat Jul 06, 2019 12:47 am
Graham Banks wrote:
Fri Jul 05, 2019 11:36 pm
I might be able to play a top engine with rook odds against the following player - https://en.wikipedia.org/wiki/Martin_Dreyer.

In a six game match, how do you think he'd fare?
In the year 1993 he was rated 2345, but depending how active he still is, a players over time lose somewhere between 50 to 100 rating points and in 26 years it could be very different than was he used to be. I would say based on that point of view that he might get one or two win versus Komodo with a knight odds, but with a Rook odds Komodo might still get a few draws if he is still around 2250 to 2275 Elo and that also depends on how many cpu you will use with Komodo :mrgreen:
I would also let him play against one of these programs to see how is is standing head to head without odds ===>

283‑284 Joker 1.1.14 2306 +13 −13 46.4% +25.6 20.8% 2461
52.7%
283‑284 Kingfisher 1.0 64-bit 2306 +17 −17 51.3% −8.4 36.5% 1166
55.4%
285 Tytan 9.32 64-bit 2304 +13 −13 46.4% +26.2 23.8% 2225
85.6%
286 Resp 0.19 64-bit Nobook 2296 +10 −10 49.8% −2.5 21.5% 3757
72.9%
287 Googleplex Starthinker 1.4 64-bit 2289 +18 −17 62.7% −97.1 20.5% 1303
58.1%
288 Ayito 0.2.994 2287 +11 −11 53.1% −25.5 19.1% 3519
54.6%
289 Chaturanga 2.4.3 2285 +27 −27 53.0% −21.7 19.8% 499
59.4%
290 Matilde 2008 64-bit 2282 +14 −13 53.2% −25.5 21.0% 2137
59.2%
291 Fischerle 0.9.80 SE 64-bit 2279 +16 −16 51.7% −11.7 23.1% 1405
66.3%
292 Rival 1.0.1 64-bit