Human odds games

[koedem]

A friend of mine was asking whether there have been serious attempts to find good handicaps for various rating differences. In engine games there seems to be some rough understanding of it, however what about human vs human games?
Maybe @lkaufman has some insights?

[Uri Blass]

A friend of mine was asking whether there have been serious attempts to find good handicaps for various rating differences. In engine games there seems to be some rough understanding of it, however what about human vs human games?
Maybe @lkaufman has some insights?

It is not about rating difference and you need smaller handicap at higher level at the same rating difference.

From my experience I can beat most human players with queen and 2 rooks handicap.
There are some children that I even can beat with only knight bishop and pawns against all the pieces and in the weakest level even only pawns are enough to win because the opponent does not capture the pawns and let a pawn to promote.

These players never played for rating so they have no known rating that I know but most chess players have no known chess rating.

[koedem]

A friend of mine was asking whether there have been serious attempts to find good handicaps for various rating differences. In engine games there seems to be some rough understanding of it, however what about human vs human games?
Maybe @lkaufman has some insights?

It is not about rating difference and you need smaller handicap at higher level at the same rating difference.

From my experience I can beat most human players with queen and 2 rooks handicap.
There are some children that I even can beat with only knight bishop and pawns against all the pieces and in the weakest level even only pawns are enough to win because the opponent does not capture the pawns and let a pawn to promote.

These players never played for rating so they have no known rating that I know but most chess players have no known chess rating.

Yes, of course at GM level a N is way bigger handicap than at beginner level. That is the point of this question, to get a more precise handicap system across different rating ranges.
Though it is in the scenario of both players having a known rating. (and the question which handicap to pick for a close game)

[lkaufman]

A friend of mine was asking whether there have been serious attempts to find good handicaps for various rating differences. In engine games there seems to be some rough understanding of it, however what about human vs human games?
Maybe @lkaufman has some insights?

Yes, I know a lot about that topic. Much depends on the time limit. Normally people don't play material handicap games at 2 hours per side or anything like that, usually it's either rapid or blitz or untimed games that are somewhere in between rapid and blitz. My own experience is that at this sort of speed (let's say ten min plus five sec inc) if someone is even with me at knight odds I'll call him 1800, rook odds 1600, two knights or queen for knight 1400, rook and knight 1200, queen odds 1000. Of course with elementary school kids the odds are generally much higher, as in most schools the best player is below 1000 strength. At the top end, in blitz Bobby Fischer used to give f7 and three or four moves or even knight odds to masters. More recently, Carlsen and MVL played handicap blitz matches with IM Lawrence Trent. I think Carlsen won at f2 + g2 but lost at rook, so maybe knight odds would have been reasonable, but I don't have the details. In slow (near standard time limit) chess, we have the four game match Kasparov vs. a British tycoon who became an FM somewhat later. Kasparov gave two pawns (a +b,d,e, and h, alternating colors, if memory serves) and won by 2.5 to 1.5, but I would only call the "a" pawn a half-pawn handicap so I would call this 1 1/2 pawn handicap. I would love to see Carlsen (or MVL or Naka) play rapid matches giving handicaps, perhaps f7 to some midlevel GM for example. I'd be glad to take a suitable handicap in a match from one of those guys, but at my age it might not prove anything other than that I'm no longer the player I used to be.

[Henk]

I remember Smyslow played a good tournament when he was 65 years old.
Steinitz also played well when 65. Although he died a few years later.

[lkaufman]

I remember Smyslow played a good tournament when he was 65 years old.
Steinitz also played well when 65. Although he died a few years later.

Since Steinitz died at 64, he probably didn't play so well at 65! Korchnoi is by far the best example of playing well at an advanced age, up to age 80 he was a strong GM. I tied for second in the Washington International Blitz tournament a few weeks ago, so I'm not doing poorly for my age (71), but I've clearly declined in strength over the past decade.

[Henk]

Ok my memory not that great.

Steinitz won tournament at 1897 (tied). Born in 1836 he was about 61 years old.

[koedem]

A friend of mine was asking whether there have been serious attempts to find good handicaps for various rating differences. In engine games there seems to be some rough understanding of it, however what about human vs human games?
Maybe @lkaufman has some insights?

Yes, I know a lot about that topic. Much depends on the time limit. Normally people don't play material handicap games at 2 hours per side or anything like that, usually it's either rapid or blitz or untimed games that are somewhere in between rapid and blitz. My own experience is that at this sort of speed (let's say ten min plus five sec inc) if someone is even with me at knight odds I'll call him 1800, rook odds 1600, two knights or queen for knight 1400, rook and knight 1200, queen odds 1000. Of course with elementary school kids the odds are generally much higher, as in most schools the best player is below 1000 strength. At the top end, in blitz Bobby Fischer used to give f7 and three or four moves or even knight odds to masters. More recently, Carlsen and MVL played handicap blitz matches with IM Lawrence Trent. I think Carlsen won at f2 + g2 but lost at rook, so maybe knight odds would have been reasonable, but I don't have the details. In slow (near standard time limit) chess, we have the four game match Kasparov vs. a British tycoon who became an FM somewhat later. Kasparov gave two pawns (a +b,d,e, and h, alternating colors, if memory serves) and won by 2.5 to 1.5, but I would only call the "a" pawn a half-pawn handicap so I would call this 1 1/2 pawn handicap. I would love to see Carlsen (or MVL or Naka) play rapid matches giving handicaps, perhaps f7 to some midlevel GM for example. I'd be glad to take a suitable handicap in a match from one of those guys, but at my age it might not prove anything other than that I'm no longer the player I used to be.

Thanks for the response. I think my friends aim was to create a way to play against other rated players and have the board set up so that it would be a close match based on the ratings.
So the question might be if there's a way to express piece odds in pawn handicap units that scale accordingly. (i.e. 2 pawn handicap units would be for twice as big rating difference as one pawn handicap) Not sure if that's feasible and/or comparable across different ratings.
A concrete example, I played some odds games against Komodo some time ago, with Jesse transmitting the moves. In the rook odds handicap games I usually got outplayed because a rook doesn't help very much until I eventually had the idea to sac the exchange to then be up a piece which does help.
Small sample size, but I felt playing with the piece was a lot easier than with the rook so maybe at least at my level a piece handicap might be just as good?!

The other question is of course based on the rating of the weaker player, what handicap translates to which rating difference. I.e. is there a function f: R -> R s.t. f(rating) = ratingDiffWithPawnOdds and/or how it would look like. Quite clearly this wouldn't be a constant function since at GM level a pawn will be much more than at beginner level. Maybe a linear function might be an alright approximation?! But probably a decent approximation would at least have to be quadratic, so that e.g. twice the rating means the handicap is 4 times as big or so.

Yet another problem are time controls which probably makes the problem too complicated to be solved well. Maybe at least for the rapid/blitz case with the available data there could be a rough interpolation based on existing data.

[lkaufman]

Okay, here is my best guess. Let's assume a time limit of the World Rapid Championship, 15' + 10", I think that most casual chess games would adhere pretty well to this speed. For now let's assume the average rating of the opponents is 2000 (it's better to go by the average rating, it makes the handicaps more linear). My guide for fair handicaps would be: 100 = f2 (pawn); 200 = f7 (pawn and move); 300 = f7 with e4 already played, wtm (pawn and two moves); 400 = f7 with e4 and d4 already played, wtm (pawn and three moves); 500 = b1 or g1 (knight odds); 600 = b8 or g8 (knight and move odds); 700 = a1 (rook); 800 = a8 (rook and move). You can extend it with 900 = a1 + f2 (rook and pawn) etc. If the average rating is not 2000, a quadratic adjustment feels about right, so for example if the average rating is 3000 you multiply the handicap value by 3/2 squared, which is 2.25. So f7 is 450 and f7 + second White move is 675, which agree pretty well with data from engine vs. engine and engine vs. top GM games.
Regarding your question about rook odds vs. knight odds, rook odds is a larger handicap at any level, though not 5/3 as large, more like the 7 to 5 ratio implied by my above table. It's true that the rook is less useful than the knight in the opening, most in most games pieces and pawns get traded and the game opens up, and in the endgame the rook clobbers the knight. Beating a top engine at knight odds is very difficult unless you are a master and are using a reasonably slow time control, whereas at rook odds a near-master can win even at "slow blitz", like 5' + 2" or so. I think that odds of rook for knight is an excellent handicap at any level, roughly equal to f7 odds. Our experience with Komodo (and earlier Rybka) vs GMs shows that the human player normally reaches a position that he "should" win, but in fact no human ever actually beat Komodo with rook for knight odds (even with the White pieces), but there were about half draws and one GM (Milov) did beat Rybka once (with 3 draws) a decade ago. With equal engines playing the Exchange almost always wins.

[jefk]

this topic seems to be revived in a new thread now,
about SF with queen odds vs 1450 player

(this is just for reference)