How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
lkaufman
Posts: 4524
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by lkaufman » Mon Feb 15, 2021 11:47 pm

Chessqueen wrote:
Mon Feb 15, 2021 11:29 pm
lkaufman wrote:
Sun Feb 14, 2021 5:56 pm
jefk wrote:
Sun Feb 14, 2021 10:56 am
mr LK wrote:
Is there someplace on the Web that lists all the handicaps you propose as "Fair-Chess"? All I found was a general description. If not perhaps you can list them here. In particular, what handicaps do you list between f7 odds and b1 odds?
the Fair-hess (FC) system is found in the download here
(with a zillions (*) prototype)
https://sourceforge.net/projects/fairchess/

but it needs to be updated, as i see it know, working with
rating ratio, it could be something like this:
The higher rated player will play with White just like
in simultaneous exhibitions; at equal rating, White's f2
pawn is removed, and Black's c7 pawn (i checked such
a setup with two Nnue engines and this handicap combination
seems to compensate the White advantage pretty well.

As for your question regarding a gap between f7 and b1 handicap,
well imho we can use different combinations of pawns/pieces
Below at various higher levels or rating for White a description
of handicaps vs rating ratio;s as i would currently suggest:
(should be worked out later in a bit more detail)
so currently i have something like :
---------------------------------------------------------
White higher rating :
a) White 1.15x higher rating only f2 removed (not the c7 pawn)
b) White approx 1.3 higher rating, f2 and c3 removed
c) White approx 1.45 higher rating knight b1 removed
d) White .. 1.6 higher rating knight b1 and f2 removed
e) .. .. 1.75 .. .. knt b1 f2 and c3 and f2 removed
f) .. .. 1.9 .. .. rook a1, and f2 removed
g) .. .. >2.1 .. .. knight b1, c2, and f1 (bishop) removed
------------------------------------------------------
NB this is tentative, and although i intend to update
my Zillions rule-file for FC in such a way, i don't claim it's
accurate but it's a first trial, and my intention is to update
it in several steps, possibly with your cooperation and/or
suggestions (especially regarding the handicaps

As for time odds, or move odds, i prefer to keep it simple,
as i now am going to explain to 'cornfed' in another thread...

(*) the zillions program seems to be abandonware (**), although
there's still a demo for download available
(**) considering the discussion regarding Fritz532, i guess i
shouldn't include a (my) registration nr for Zillions in the
download although it would make sense for people who would like
to test it (for longer than the zillions trial period ..)
Zillions is much easier for developing testing slight deviations of
a chess variant than eg, winboard so i some way it's a pity
that it's not sold anymore; maybe they simply should make
it freeware, just a little hint :)
You give one handicap between pawn and knight, namely c2 +f2 (you wrote c3 + f2 but that's an obvious typo). That's fine, but not nearly enough steps for serious play. We found that at Rapid the c2 + f2 handicap was not enough even for Hikaru Nakamura vs. Komodo Dragon; he scored 1/2 out of 2 at that, and just 1.5 out of 8 at similar two pawn handicaps in total. But at the same Rapid time control (15' + 10"), I am at least even with Dragon at knight odds. The gap between me and Hikaru in Rapid play is probably about 700 elo points. So the difference between these two handicaps is more than that (at this level), maybe 900 elo or so. Clearly we need steps in between for use in matches between Dragon and human GMs. The obvious ones are various two Black pawn handicaps and various three White pawn handicaps, and/or knight for pawn handicaps. But taking too many pawns off really changes the game, much more so than removing a knight. Not sure what is best.
I was wondering if Komodo can give a Knight to GM Smeldon at TC 3'+1" and get at least a Draw
https://www.youtube.com/watch?v=j1cZEFh ... inFinegold
I would bet on the Komodo Dragon to win a match at Knight odds vs GM Smerdon or another similarly rated GM at 3' + 1". Actually Komodo was about even in blitz at knight odds with similarly rated GMs a few years ago, so the Dragon should be a clear favorite. Maybe I should try it vs. GM Lenderman, who is around 2640 FIDE at Rapid (a bit more at blitz). He beat regular Komodo in blitz pretty easily last year, but maybe Dragon is strong enough.
Komodo rules!

Chessqueen
Posts: 1440
Joined: Wed Sep 05, 2018 12:16 am
Full name: George Pichard

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by Chessqueen » Mon Feb 15, 2021 11:51 pm

lkaufman wrote:
Mon Feb 15, 2021 11:47 pm
Chessqueen wrote:
Mon Feb 15, 2021 11:29 pm
lkaufman wrote:
Sun Feb 14, 2021 5:56 pm
jefk wrote:
Sun Feb 14, 2021 10:56 am
mr LK wrote:
Is there someplace on the Web that lists all the handicaps you propose as "Fair-Chess"? All I found was a general description. If not perhaps you can list them here. In particular, what handicaps do you list between f7 odds and b1 odds?
the Fair-hess (FC) system is found in the download here
(with a zillions (*) prototype)
https://sourceforge.net/projects/fairchess/

but it needs to be updated, as i see it know, working with
rating ratio, it could be something like this:
The higher rated player will play with White just like
in simultaneous exhibitions; at equal rating, White's f2
pawn is removed, and Black's c7 pawn (i checked such
a setup with two Nnue engines and this handicap combination
seems to compensate the White advantage pretty well.

As for your question regarding a gap between f7 and b1 handicap,
well imho we can use different combinations of pawns/pieces
Below at various higher levels or rating for White a description
of handicaps vs rating ratio;s as i would currently suggest:
(should be worked out later in a bit more detail)
so currently i have something like :
---------------------------------------------------------
White higher rating :
a) White 1.15x higher rating only f2 removed (not the c7 pawn)
b) White approx 1.3 higher rating, f2 and c3 removed
c) White approx 1.45 higher rating knight b1 removed
d) White .. 1.6 higher rating knight b1 and f2 removed
e) .. .. 1.75 .. .. knt b1 f2 and c3 and f2 removed
f) .. .. 1.9 .. .. rook a1, and f2 removed
g) .. .. >2.1 .. .. knight b1, c2, and f1 (bishop) removed
------------------------------------------------------
NB this is tentative, and although i intend to update
my Zillions rule-file for FC in such a way, i don't claim it's
accurate but it's a first trial, and my intention is to update
it in several steps, possibly with your cooperation and/or
suggestions (especially regarding the handicaps

As for time odds, or move odds, i prefer to keep it simple,
as i now am going to explain to 'cornfed' in another thread...

(*) the zillions program seems to be abandonware (**), although
there's still a demo for download available
(**) considering the discussion regarding Fritz532, i guess i
shouldn't include a (my) registration nr for Zillions in the
download although it would make sense for people who would like
to test it (for longer than the zillions trial period ..)
Zillions is much easier for developing testing slight deviations of
a chess variant than eg, winboard so i some way it's a pity
that it's not sold anymore; maybe they simply should make
it freeware, just a little hint :)
You give one handicap between pawn and knight, namely c2 +f2 (you wrote c3 + f2 but that's an obvious typo). That's fine, but not nearly enough steps for serious play. We found that at Rapid the c2 + f2 handicap was not enough even for Hikaru Nakamura vs. Komodo Dragon; he scored 1/2 out of 2 at that, and just 1.5 out of 8 at similar two pawn handicaps in total. But at the same Rapid time control (15' + 10"), I am at least even with Dragon at knight odds. The gap between me and Hikaru in Rapid play is probably about 700 elo points. So the difference between these two handicaps is more than that (at this level), maybe 900 elo or so. Clearly we need steps in between for use in matches between Dragon and human GMs. The obvious ones are various two Black pawn handicaps and various three White pawn handicaps, and/or knight for pawn handicaps. But taking too many pawns off really changes the game, much more so than removing a knight. Not sure what is best.
I was wondering if Komodo can give a Knight to GM Smeldon at TC 3'+1" and get at least a Draw
I would bet on the Komodo Dragon to win a match at Knight odds vs GM Smerdon or another similarly rated GM at 3' + 1". Actually Komodo was about even in blitz at knight odds with similarly rated GMs a few years ago, so the Dragon should be a clear favorite. Maybe I should try it vs. GM Lenderman, who is around 2640 FIDE at Rapid (a bit more at blitz). He beat regular Komodo in blitz pretty easily last year, but maybe Dragon is strong enough.
Against GM Lenderman I believe Dragon is NOT ready at Rapid Yet, but it has a chance at TC 5'+1
https://www.youtube.com/watch?v=j1cZEFh ... inFinegold

lkaufman
Posts: 4524
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by lkaufman » Tue Feb 16, 2021 2:29 am

Chessqueen wrote:
Mon Feb 15, 2021 11:51 pm
lkaufman wrote:
Mon Feb 15, 2021 11:47 pm
Chessqueen wrote:
Mon Feb 15, 2021 11:29 pm
lkaufman wrote:
Sun Feb 14, 2021 5:56 pm
jefk wrote:
Sun Feb 14, 2021 10:56 am
mr LK wrote:
Is there someplace on the Web that lists all the handicaps you propose as "Fair-Chess"? All I found was a general description. If not perhaps you can list them here. In particular, what handicaps do you list between f7 odds and b1 odds?
the Fair-hess (FC) system is found in the download here
(with a zillions (*) prototype)
https://sourceforge.net/projects/fairchess/

but it needs to be updated, as i see it know, working with
rating ratio, it could be something like this:
The higher rated player will play with White just like
in simultaneous exhibitions; at equal rating, White's f2
pawn is removed, and Black's c7 pawn (i checked such
a setup with two Nnue engines and this handicap combination
seems to compensate the White advantage pretty well.

As for your question regarding a gap between f7 and b1 handicap,
well imho we can use different combinations of pawns/pieces
Below at various higher levels or rating for White a description
of handicaps vs rating ratio;s as i would currently suggest:
(should be worked out later in a bit more detail)
so currently i have something like :
---------------------------------------------------------
White higher rating :
a) White 1.15x higher rating only f2 removed (not the c7 pawn)
b) White approx 1.3 higher rating, f2 and c3 removed
c) White approx 1.45 higher rating knight b1 removed
d) White .. 1.6 higher rating knight b1 and f2 removed
e) .. .. 1.75 .. .. knt b1 f2 and c3 and f2 removed
f) .. .. 1.9 .. .. rook a1, and f2 removed
g) .. .. >2.1 .. .. knight b1, c2, and f1 (bishop) removed
------------------------------------------------------
NB this is tentative, and although i intend to update
my Zillions rule-file for FC in such a way, i don't claim it's
accurate but it's a first trial, and my intention is to update
it in several steps, possibly with your cooperation and/or
suggestions (especially regarding the handicaps

As for time odds, or move odds, i prefer to keep it simple,
as i now am going to explain to 'cornfed' in another thread...

(*) the zillions program seems to be abandonware (**), although
there's still a demo for download available
(**) considering the discussion regarding Fritz532, i guess i
shouldn't include a (my) registration nr for Zillions in the
download although it would make sense for people who would like
to test it (for longer than the zillions trial period ..)
Zillions is much easier for developing testing slight deviations of
a chess variant than eg, winboard so i some way it's a pity
that it's not sold anymore; maybe they simply should make
it freeware, just a little hint :)
You give one handicap between pawn and knight, namely c2 +f2 (you wrote c3 + f2 but that's an obvious typo). That's fine, but not nearly enough steps for serious play. We found that at Rapid the c2 + f2 handicap was not enough even for Hikaru Nakamura vs. Komodo Dragon; he scored 1/2 out of 2 at that, and just 1.5 out of 8 at similar two pawn handicaps in total. But at the same Rapid time control (15' + 10"), I am at least even with Dragon at knight odds. The gap between me and Hikaru in Rapid play is probably about 700 elo points. So the difference between these two handicaps is more than that (at this level), maybe 900 elo or so. Clearly we need steps in between for use in matches between Dragon and human GMs. The obvious ones are various two Black pawn handicaps and various three White pawn handicaps, and/or knight for pawn handicaps. But taking too many pawns off really changes the game, much more so than removing a knight. Not sure what is best.
I was wondering if Komodo can give a Knight to GM Smeldon at TC 3'+1" and get at least a Draw
I would bet on the Komodo Dragon to win a match at Knight odds vs GM Smerdon or another similarly rated GM at 3' + 1". Actually Komodo was about even in blitz at knight odds with similarly rated GMs a few years ago, so the Dragon should be a clear favorite. Maybe I should try it vs. GM Lenderman, who is around 2640 FIDE at Rapid (a bit more at blitz). He beat regular Komodo in blitz pretty easily last year, but maybe Dragon is strong enough.
Against GM Lenderman I believe Dragon is NOT ready at Rapid Yet, but it has a chance at TC 5'+1
https://www.youtube.com/watch?v=j1cZEFh ... inFinegold
I was talking about the time limit you mentioned, 3' + 1". Of course Dragon isn't strong enough for him at Rapid knight odds, since I'm about even at 10' +5" at knight odds and Lenderman is much stronger than I am now. Since the computer will normally just be playing for a draw at knight odds, Armageddon knight odds makes the most sense as a handicap, where Black has to win with the extra knight. Maybe then 5' + 1" might be fair. In my experience, it often happens that I retain the extra knight into the endgame, perhaps losing one pawn along the way, but with all the pawns remaining on one side many endings like that are rather difficult to win with short time against a computer.
Komodo rules!

Chessqueen
Posts: 1440
Joined: Wed Sep 05, 2018 12:16 am
Full name: George Pichard

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by Chessqueen » Tue Feb 16, 2021 2:47 pm

lkaufman wrote:
Tue Feb 16, 2021 2:29 am
Chessqueen wrote:
Mon Feb 15, 2021 11:51 pm
lkaufman wrote:
Mon Feb 15, 2021 11:47 pm
Chessqueen wrote:
Mon Feb 15, 2021 11:29 pm
lkaufman wrote:
Sun Feb 14, 2021 5:56 pm
jefk wrote:
Sun Feb 14, 2021 10:56 am
mr LK wrote:
Is there someplace on the Web that lists all the handicaps you propose as "Fair-Chess"? All I found was a general description. If not perhaps you can list them here. In particular, what handicaps do you list between f7 odds and b1 odds?
the Fair-hess (FC) system is found in the download here
(with a zillions (*) prototype)
https://sourceforge.net/projects/fairchess/

but it needs to be updated, as i see it know, working with
rating ratio, it could be something like this:
The higher rated player will play with White just like
in simultaneous exhibitions; at equal rating, White's f2
pawn is removed, and Black's c7 pawn (i checked such
a setup with two Nnue engines and this handicap combination
seems to compensate the White advantage pretty well.

As for your question regarding a gap between f7 and b1 handicap,
well imho we can use different combinations of pawns/pieces
Below at various higher levels or rating for White a description
of handicaps vs rating ratio;s as i would currently suggest:
(should be worked out later in a bit more detail)
so currently i have something like :
---------------------------------------------------------
White higher rating :
a) White 1.15x higher rating only f2 removed (not the c7 pawn)
b) White approx 1.3 higher rating, f2 and c3 removed
c) White approx 1.45 higher rating knight b1 removed
d) White .. 1.6 higher rating knight b1 and f2 removed
e) .. .. 1.75 .. .. knt b1 f2 and c3 and f2 removed
f) .. .. 1.9 .. .. rook a1, and f2 removed
g) .. .. >2.1 .. .. knight b1, c2, and f1 (bishop) removed
------------------------------------------------------
NB this is tentative, and although i intend to update
my Zillions rule-file for FC in such a way, i don't claim it's
accurate but it's a first trial, and my intention is to update
it in several steps, possibly with your cooperation and/or
suggestions (especially regarding the handicaps

As for time odds, or move odds, i prefer to keep it simple,
as i now am going to explain to 'cornfed' in another thread...

(*) the zillions program seems to be abandonware (**), although
there's still a demo for download available
(**) considering the discussion regarding Fritz532, i guess i
shouldn't include a (my) registration nr for Zillions in the
download although it would make sense for people who would like
to test it (for longer than the zillions trial period ..)
Zillions is much easier for developing testing slight deviations of
a chess variant than eg, winboard so i some way it's a pity
that it's not sold anymore; maybe they simply should make
it freeware, just a little hint :)
You give one handicap between pawn and knight, namely c2 +f2 (you wrote c3 + f2 but that's an obvious typo). That's fine, but not nearly enough steps for serious play. We found that at Rapid the c2 + f2 handicap was not enough even for Hikaru Nakamura vs. Komodo Dragon; he scored 1/2 out of 2 at that, and just 1.5 out of 8 at similar two pawn handicaps in total. But at the same Rapid time control (15' + 10"), I am at least even with Dragon at knight odds. The gap between me and Hikaru in Rapid play is probably about 700 elo points. So the difference between these two handicaps is more than that (at this level), maybe 900 elo or so. Clearly we need steps in between for use in matches between Dragon and human GMs. The obvious ones are various two Black pawn handicaps and various three White pawn handicaps, and/or knight for pawn handicaps. But taking too many pawns off really changes the game, much more so than removing a knight. Not sure what is best.
I was wondering if Komodo can give a Knight to GM Smeldon at TC 3'+1" and get at least a Draw
I would bet on the Komodo Dragon to win a match at Knight odds vs GM Smerdon or another similarly rated GM at 3' + 1". Actually Komodo was about even in blitz at knight odds with similarly rated GMs a few years ago, so the Dragon should be a clear favorite. Maybe I should try it vs. GM Lenderman, who is around 2640 FIDE at Rapid (a bit more at blitz). He beat regular Komodo in blitz pretty easily last year, but maybe Dragon is strong enough.
Against GM Lenderman I believe Dragon is NOT ready at Rapid Yet, but it has a chance at TC 5'+1
https://www.youtube.com/watch?v=j1cZEFh ... inFinegold
I was talking about the time limit you mentioned, 3' + 1". Of course Dragon isn't strong enough for him at Rapid knight odds, since I'm about even at 10' +5" at knight odds and Lenderman is much stronger than I am now. Since the computer will normally just be playing for a draw at knight odds, Armageddon knight odds makes the most sense as a handicap, where Black has to win with the extra knight. Maybe then 5' + 1" might be fair. In my experience, it often happens that I retain the extra knight into the endgame, perhaps losing one pawn along the way, but with all the pawns remaining on one side many endings like that are rather difficult to win with short time against a computer.
Mr. Kaufman, You are the person with most computer chess engine experience alive. You helped to write the Opening Book back in the 70's or a few years earlier of the chess engine or program Mac Hack, later on Socrates VI which was very primitive probably rated 1500 since it beat Ben Landin USCF rated 1510 back in the 70's. Please tell us how different Mac Hack VI was from Komodo as best as you can remember. :roll:

jefk
Posts: 502
Joined: Sun Jul 25, 2010 8:07 pm
Location: the Netherlands
Contact:

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by jefk » Tue Feb 16, 2021 5:11 pm

Larry (if i may call you so).
while i'm not going to mix into the little discussion
with Chessqueen here (whereby she apparently also indicates a
wish to have a system with equal chances at various levels between
top engine and Gm's super or not (fine with me, once we have such
-possibly preliminary- standards, it also would work at the
lowest levels, i hope_
:)

you wrote:
Clearly we need steps in between for use in matches between Dragon and human GMs. The obvious ones are various two Black pawn handicaps and various three White pawn handicaps, and/or knight for pawn handicaps.
My preference is for knight vs pawn handicap (eg. b7 for 'Nakamura'
(we can test this maybe with RodentIII with the naka settings :)
Reason: you wrote yourself the game will change a lot when there
are too much pawn handicaps>

As for the ratio's i proposed, i'm now thinking about levels
as 1.1, 1.2, 1.3, 1.45 etc. (and in between ofcourse).
It can be done, obviously not perfectly, but then
at least we get a more standard procedure than only
-ever changing- exhibition games or so

Meanwhile i think about it a bit further, and in a few
days revert with a more refined proposal. Meantime i suggest
we not only think about Nakamura but also other Gm's
and then the comps may come back in the game again
(well at least in some special, possibly informal tourns)
jefk

lkaufman
Posts: 4524
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by lkaufman » Tue Feb 16, 2021 8:00 pm

Chessqueen wrote:
Tue Feb 16, 2021 2:47 pm
lkaufman wrote:
Tue Feb 16, 2021 2:29 am
Chessqueen wrote:
Mon Feb 15, 2021 11:51 pm
lkaufman wrote:
Mon Feb 15, 2021 11:47 pm
Chessqueen wrote:
Mon Feb 15, 2021 11:29 pm
lkaufman wrote:
Sun Feb 14, 2021 5:56 pm
jefk wrote:
Sun Feb 14, 2021 10:56 am
mr LK wrote:
Is there someplace on the Web that lists all the handicaps you propose as "Fair-Chess"? All I found was a general description. If not perhaps you can list them here. In particular, what handicaps do you list between f7 odds and b1 odds?
the Fair-hess (FC) system is found in the download here
(with a zillions (*) prototype)
https://sourceforge.net/projects/fairchess/

but it needs to be updated, as i see it know, working with
rating ratio, it could be something like this:
The higher rated player will play with White just like
in simultaneous exhibitions; at equal rating, White's f2
pawn is removed, and Black's c7 pawn (i checked such
a setup with two Nnue engines and this handicap combination
seems to compensate the White advantage pretty well.

As for your question regarding a gap between f7 and b1 handicap,
well imho we can use different combinations of pawns/pieces
Below at various higher levels or rating for White a description
of handicaps vs rating ratio;s as i would currently suggest:
(should be worked out later in a bit more detail)
so currently i have something like :
---------------------------------------------------------
White higher rating :
a) White 1.15x higher rating only f2 removed (not the c7 pawn)
b) White approx 1.3 higher rating, f2 and c3 removed
c) White approx 1.45 higher rating knight b1 removed
d) White .. 1.6 higher rating knight b1 and f2 removed
e) .. .. 1.75 .. .. knt b1 f2 and c3 and f2 removed
f) .. .. 1.9 .. .. rook a1, and f2 removed
g) .. .. >2.1 .. .. knight b1, c2, and f1 (bishop) removed
------------------------------------------------------
NB this is tentative, and although i intend to update
my Zillions rule-file for FC in such a way, i don't claim it's
accurate but it's a first trial, and my intention is to update
it in several steps, possibly with your cooperation and/or
suggestions (especially regarding the handicaps

As for time odds, or move odds, i prefer to keep it simple,
as i now am going to explain to 'cornfed' in another thread...

(*) the zillions program seems to be abandonware (**), although
there's still a demo for download available
(**) considering the discussion regarding Fritz532, i guess i
shouldn't include a (my) registration nr for Zillions in the
download although it would make sense for people who would like
to test it (for longer than the zillions trial period ..)
Zillions is much easier for developing testing slight deviations of
a chess variant than eg, winboard so i some way it's a pity
that it's not sold anymore; maybe they simply should make
it freeware, just a little hint :)
You give one handicap between pawn and knight, namely c2 +f2 (you wrote c3 + f2 but that's an obvious typo). That's fine, but not nearly enough steps for serious play. We found that at Rapid the c2 + f2 handicap was not enough even for Hikaru Nakamura vs. Komodo Dragon; he scored 1/2 out of 2 at that, and just 1.5 out of 8 at similar two pawn handicaps in total. But at the same Rapid time control (15' + 10"), I am at least even with Dragon at knight odds. The gap between me and Hikaru in Rapid play is probably about 700 elo points. So the difference between these two handicaps is more than that (at this level), maybe 900 elo or so. Clearly we need steps in between for use in matches between Dragon and human GMs. The obvious ones are various two Black pawn handicaps and various three White pawn handicaps, and/or knight for pawn handicaps. But taking too many pawns off really changes the game, much more so than removing a knight. Not sure what is best.
I was wondering if Komodo can give a Knight to GM Smeldon at TC 3'+1" and get at least a Draw
I would bet on the Komodo Dragon to win a match at Knight odds vs GM Smerdon or another similarly rated GM at 3' + 1". Actually Komodo was about even in blitz at knight odds with similarly rated GMs a few years ago, so the Dragon should be a clear favorite. Maybe I should try it vs. GM Lenderman, who is around 2640 FIDE at Rapid (a bit more at blitz). He beat regular Komodo in blitz pretty easily last year, but maybe Dragon is strong enough.
Against GM Lenderman I believe Dragon is NOT ready at Rapid Yet, but it has a chance at TC 5'+1
https://www.youtube.com/watch?v=j1cZEFh ... inFinegold
I was talking about the time limit you mentioned, 3' + 1". Of course Dragon isn't strong enough for him at Rapid knight odds, since I'm about even at 10' +5" at knight odds and Lenderman is much stronger than I am now. Since the computer will normally just be playing for a draw at knight odds, Armageddon knight odds makes the most sense as a handicap, where Black has to win with the extra knight. Maybe then 5' + 1" might be fair. In my experience, it often happens that I retain the extra knight into the endgame, perhaps losing one pawn along the way, but with all the pawns remaining on one side many endings like that are rather difficult to win with short time against a computer.
Mr. Kaufman, You are the person with most computer chess engine experience alive. You helped to write the Opening Book back in the 70's or a few years earlier of the chess engine or program Mac Hack, later on Socrates VI which was very primitive probably rated 1500 since it beat Ben Landin USCF rated 1510 back in the 70's. Please tell us how different Mac Hack VI was from Komodo as best as you can remember. :roll:
MacHack was rated 1529 at the end of 1968, the year I graduated from M.I.T. and finished my work on the project. I think it typically did a five ply search at standard tournament time controls. But the chess knowledge was rather primitive, and I think the five ply search was not full width but tapered. Probably if you took regular Komodo 14 or 14.1, set it to a fixed depth of five ply, and turned Dynamism way down to something like 25, it would be a crude simulation of MacHack at 40 moves in two hours. But probably Komodo would still be a lot stronger than MacHack even so. Maybe Skill levels 13 or 14 might roughly simulate MacHack at 40 moves in two hours, maybe level 11 or so might simulate it at Rapid.
Komodo rules!

lkaufman
Posts: 4524
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by lkaufman » Tue Feb 16, 2021 8:13 pm

jefk wrote:
Tue Feb 16, 2021 5:11 pm
Larry (if i may call you so).
while i'm not going to mix into the little discussion
with Chessqueen here (whereby she apparently also indicates a
wish to have a system with equal chances at various levels between
top engine and Gm's super or not (fine with me, once we have such
-possibly preliminary- standards, it also would work at the
lowest levels, i hope_
:)

you wrote:
Clearly we need steps in between for use in matches between Dragon and human GMs. The obvious ones are various two Black pawn handicaps and various three White pawn handicaps, and/or knight for pawn handicaps.
My preference is for knight vs pawn handicap (eg. b7 for 'Nakamura'
(we can test this maybe with RodentIII with the naka settings :)
Reason: you wrote yourself the game will change a lot when there
are too much pawn handicaps>

As for the ratio's i proposed, i'm now thinking about levels
as 1.1, 1.2, 1.3, 1.45 etc. (and in between ofcourse).
It can be done, obviously not perfectly, but then
at least we get a more standard procedure than only
-ever changing- exhibition games or so

Meanwhile i think about it a bit further, and in a few
days revert with a more refined proposal. Meantime i suggest
we not only think about Nakamura but also other Gm's
and then the comps may come back in the game again
(well at least in some special, possibly informal tourns)
Knight for pawn was tested quite thoroughly between Komodo (just before Dragon) and GM Alex Lenderman in Rapid (15' + 10"). We played eight games, two with b7 off, two with c7 off, two with d7 off, two with e7 off, in each case once with b1 knight off and once with g1 knight off. The first two games were with the standard Komodo version, the last six with the MCTS version. Komodo drew six and lost two of these games; the ones with b7 off were both drawn. There were also three games between Komodo and superstars Nakamura and MVL at knight for pawn odds; Nakamura lost with knight b1 for c7 pawn, MVL lost with Ng1 for f7 pawn (the hardest knight for pawn handicap for the human), and Nakamura drew with a third Black knight replacing the f7 pawn (!). I think that c7 and b7 are the most neutral and appropriate pawns for this, they minimize the distortion.
Komodo rules!

Chessqueen
Posts: 1440
Joined: Wed Sep 05, 2018 12:16 am
Full name: George Pichard

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by Chessqueen » Wed Feb 17, 2021 5:55 am

mhouppin wrote:
Thu Feb 11, 2021 5:49 pm
FWIW, I ran a minimatch between SF dev and Stash 10.0 (roughly 1650 Elo at CCRL 40/4), time control 3'+2", 1 thread, 256 MB hash, 5-man Syzygy tablebases for SF, I used starting position with one of the 4 rooks removed. Here are the games:


[Event "?"]
[Site "?"]
[Date "2021.02.11"]
[Round "?"]
[White "Stockfish Dev"]
[Black "Stash 10.0"]
[Result "1-0"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBN1 w Qkq - 0 1"]
[GameDuration "00:02:47"]
[GameEndTime "2021-02-11T18:45:39.902 Paris, Madrid"]
[GameStartTime "2021-02-11T18:42:52.379 Paris, Madrid"]
[PlyCount "41"]
[SetUp "1"]
[TimeControl "180+2"]

1. e4 {-6.46/24 3.8s} d5 {+5.38/9 5.6s} 2. exd5 {-6.36/23 2.3s}
Qxd5 {+5.58/8 5.5s} 3. Nc3 {-6.37/24 3.7s} Qe5+ {+6.29/8 5.5s}
4. Nge2 {-6.40/25 2.7s} Qxh2 {+6.69/8 5.4s} 5. d4 {-6.21/26 2.5s}
e5 {+6.91/8 5.3s} 6. dxe5 {-6.13/24 4.9s} Qxe5 {+7.01/8 5.3s}
7. Bf4 {-6.01/26 5.7s} Qc5 {+7.21/8 5.2s} 8. Nd5 {-5.62/24 4.4s}
Bd6 {+7.34/8 5.1s} 9. b4 {-5.55/26 3.8s} Qc6 {+7.35/8 5.1s}
10. Nec3 {-5.12/25 4.1s} a6 {+7.55/8 5.0s} 11. b5 {-2.55/26 3.1s}
Qd7 {+7.88/8 5.0s} 12. Qd4 {-2.34/27 3.8s} Nf6 {+7.74/8 4.9s}
13. Qe3+ {-3.00/33 9.7s} Kf8 {+8.47/9 4.8s} 14. Nxf6 {-2.37/26 3.1s}
gxf6 {+4.07/9 4.8s} 15. Bh6+ {+M13/66 3.1s} Kg8 {+3.99/8 4.7s}
16. O-O-O {+M11/74 3.2s} Ba3+ {+8.05/8 4.7s} 17. Kb1 {+M9/102 0.52s}
Bd6 {-M8/8 4.6s} 18. Ne4 {+M7/245 2.8s} Qxb5+ {-M6/8 4.6s}
19. Bxb5 {+M5/245 0.014s} Nd7 {-M4/8 4.5s} 20. Nxf6+ {+M3/245 0.007s}
Nxf6 {-M2/8 4.5s} 21. Qg5# {+M1/245 0.006s, White mates} 1-0

[/pgn]

It would seem Rook odds are still too tactical for low rated engines to handle. I may ran another minimatch with a more recent version of Stash to get an idea of where the turning point is.
Can Stockfish still beat your latest version of stash 27.0 with a Rook Odds

Chessqueen
Posts: 1440
Joined: Wed Sep 05, 2018 12:16 am
Full name: George Pichard

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by Chessqueen » Wed Feb 17, 2021 1:39 pm

WOW Stash27.0 is stronger than MVL against Komodo 13.3


jefk
Posts: 502
Joined: Sun Jul 25, 2010 8:07 pm
Location: the Netherlands
Contact:

Re: How many of you believe that Stockfish 12 or Komodo can give a rook Odds to a 1450 engine ?

Post by jefk » Wed Feb 17, 2021 1:56 pm

Cq, one game isn't enough to say one engine is stronger,
further more 2.e5 wasn't a good (book?) move against the Scandinavian,
maybe it was in Kom's book with 3 pct chance or so dunno.

Once we have Stash Nnue we'll see again :)

talking about computer chess nostalgia like Socrates, i've once
been an operator for this engine in a single game, in The Hague during
one of the Aegon comp-vs (grand)master events (slow time controls).
Didn't know much about chess during that time, but remember it,
my opponent writer /chess player Tim Krabbe managed to lose
with a Petroff defense, his site still is here:
https://timkr.home.xs4all.nl/chess/chess.html
So yes, Socrates then started with 1.e4 (during that time had no idea
who (the late) Don Daily or Larry Kaufman were, btw)
:)
Last edited by jefk on Wed Feb 17, 2021 1:58 pm, edited 1 time in total.
jefk

Post Reply