Handicap engine matches

[JJJ]

For the preparation, Human could anticipate the first 10 or 20 move while training against the engine.

[Jesse Gersenson]

Knight odds against an FM hasn't been won by a machine.

On decent hardware Komodo may have a chance. In 20s+80ms Komodo won 5-4 against NanoSzachy 3.0 http://www.kalisz.mm.pl/~pic/nanochess/nanoSzachy30.zip.

[lkaufman]

What about 1 min vs 1h and no opening book for the engine ?

Yes, that sort of handicap is another way to go. Komodo while still single core did win a game from the late GM Walter Browne at 30 to one minute odds plus White pieces for human plus no book. But somehow crippling Komodo by giving it no time doesn't seem as interesting to me as some of these other ideas. The result can more or less be predicted just by doing some calculations.

[lkaufman]

For the preparation, Human could anticipate the first 10 or 20 move while training against the engine.

If that actually worked it wouldn't be so interesting, as the early moves would just be engine vs. engine. But MP makes programs unpredictible, and anyway assuming we could use a late unreleased version we could easily make sure it played rather differently without losing more than a couple elo.

[lkaufman]

Knight odds against an FM hasn't been won by a machine.

On decent hardware Komodo may have a chance. In 20s+80ms Komodo won 5-4 against NanoSzachy 3.0 http://www.kalisz.mm.pl/~pic/nanochess/nanoSzachy30.zip.

What elo has that engine? I never heard of it. Anyway it's not a good predictor of how humans would do. But I am inclined to try another knight odds match (with Komodo getting draw odds on each game) with John Meyer (FM) or another national master. Maybe at least this time we wouldn't lose every game, as Rybka did. That sort of match doesn't require significant expense.

Does anyone reading this have any experience playing handicap games against Komodo? If so state the handicap, the results, your elo, the time limit, the version of Komodo, and the hardware.

[Jesse Gersenson]

Knight odds against an FM hasn't been won by a machine.

On decent hardware Komodo may have a chance. In 20s+80ms Komodo won 5-4 against NanoSzachy 3.0 http://www.kalisz.mm.pl/~pic/nanochess/nanoSzachy30.zip.

What elo has that engine? I never heard of it. Anyway it's not a good predictor of how humans would do. But I am inclined to try another knight odds match (with Komodo getting draw odds on each game) with John Meyer (FM) or another national master. Maybe at least this time we wouldn't lose every game, as Rybka did. That sort of match doesn't require significant expense.

Does anyone reading this have any experience playing handicap games against Komodo? If so state the handicap, the results, your elo, the time limit, the version of Komodo, and the hardware.

http://www.computerchess.org.uk/ccrl/40 ... t_all.html
It's listed at 2300.

The largest tournament in europe is played in my town. Last year 60 FM's played. I could make an advertisement and find one willing to play at knight odds.

After a few more rounds. This is on one core, with no book: +12 =6 -14

If someone could tell me how to run a match in arena where Engine A always plays white I'd appreciate it.

[carldaman]

What about 1 min vs 1h and no opening book for the engine ?

Yes, that sort of handicap is another way to go. Komodo while still single core did win a game from the late GM Walter Browne at 30 to one minute odds plus White pieces for human plus no book. But somehow crippling Komodo by giving it no time doesn't seem as interesting to me as some of these other ideas. The result can more or less be predicted just by doing some calculations.

You might be crippling Komodo more by giving material odds rather than reasonable time odds. Besides, real chess is always more interesting than some artificial construct with missing pawns or pieces, although I'd be intrigued by the exchange odds games with the Black Rook on b8 that you've mentioned.

Regards,
CL

[lkaufman]

Knight odds against an FM hasn't been won by a machine.

On decent hardware Komodo may have a chance. In 20s+80ms Komodo won 5-4 against NanoSzachy 3.0 http://www.kalisz.mm.pl/~pic/nanochess/nanoSzachy30.zip.

What elo has that engine? I never heard of it. Anyway it's not a good predictor of how humans would do. But I am inclined to try another knight odds match (with Komodo getting draw odds on each game) with John Meyer (FM) or another national master. Maybe at least this time we wouldn't lose every game, as Rybka did. That sort of match doesn't require significant expense.

Does anyone reading this have any experience playing handicap games against Komodo? If so state the handicap, the results, your elo, the time limit, the version of Komodo, and the hardware.

http://www.computerchess.org.uk/ccrl/40 ... t_all.html
It's listed at 2300.

The largest tournament in europe is played in my town. Last year 60 FM's played. I could make an advertisement and find one willing to play at knight odds.

After a few more rounds. This is on one core, with no book: +12 =6 -14

If someone could tell me how to run a match in arena where Engine A always plays white I'd appreciate it.

Sure, feel free to do that. You can offer small prizes. The terms would be that Komodo always plays White, removes b1 knight, and Komodo wins any drawn games, since Komodo will always be playing just to avoid losing. I think the time limit for the comparable Rybka vs FM match was 30' +30", or maybe 45' + 15", you could offer the choice to the player. We paid $25 per win, so you can offer 25 Euros per win for an FM (or better) and of course we'll reimburse you. Limit four games per player. But don't play any matches until we modify Komodo to try harder to avoid trades.

[lkaufman]

What about 1 min vs 1h and no opening book for the engine ?

Yes, that sort of handicap is another way to go. Komodo while still single core did win a game from the late GM Walter Browne at 30 to one minute odds plus White pieces for human plus no book. But somehow crippling Komodo by giving it no time doesn't seem as interesting to me as some of these other ideas. The result can more or less be predicted just by doing some calculations.

You might be crippling Komodo more by giving material odds rather than reasonable time odds. Besides, real chess is always more interesting than some artificial construct with missing pawns or pieces, although I'd be intrigued by the exchange odds games with the Black Rook on b8 that you've mentioned.

Regards,
CL

"reasonable time odds" won't give the human a fair chance. Equalizing time odds will mean reducing the playing strength of Komodo by several hundred elo, might as well just use some 15 year old engine with equal time. We could try Komodo on a cellphone giving some moderate time handicap, but our cellphone sales don't generate much revenue so someone would have to sponsor this. Some people say they prefer a game starting with some short but bad opening sequence to explicit material odds. Examples are 1.e4 f6? 2.d4 Kf7?; or 1.e4 e5 2.Nf3 Nc6 3.Nxe5??, or 1.e4 f5?; or 1 e4.b5?, or 1.Nf3 e5?, or 1.e4 e5 2.Bc4 Nf6 3.Bxf7+ or 1.e4 Nc6 2.d4 Nb8?. Of course material handicap positions can also be reached by a long legal move sequence from the opening position, but those sequences involve silly moves by both sides, whereas the above lines only require inferior moves by the computer. I don't have a strong preference myself between explicit material handicaps and opening blunders like the above. I suppose the opening blunders could be considered more like normal chess.

[JJJ]

I think 2 pawn handicap is the best challenge for a 2600 - 2700 players