Check out Balbo's Chess, our featured variant for October, 2024.


[ Help | Earliest Comments | Latest Comments ]
[ List All Subjects of Discussion | Create New Subject of Discussion ]
[ List Earliest Comments Only For Pages | Games | Rated Pages | Rated Games | Subjects of Discussion ]

Single Comment

Zillions of Games. It can play an endless variety of abstract board games, and we have a large collection of Chess variants you can play on it.[All Comments] [Add Comment or Rating]
H. G. Muller wrote on Mon, Sep 23 05:42 AM UTC in reply to Fergus Duniho from 02:35 AM:

The rule of thumb is that engines gain about 70 Elo for each doubling in speed or thinking time (100 ln(speed)). Ofcourse it is difficult to know the speed advantage the new computer gives to Zilions, as, as far as I know, Zillions does not print the number of positions it has searched. But you could benchmark this with some other chess engine.

Tests with as few as 10 games are usually meaningless. The statistical error in the result of a match of N independent games is 40%/sqrt(N). For N=10 that amounts to 13%, and each 1% score difference equates to about 7 Elo. So the standard deviation of the measured Elo difference would be nearly 100 Elo, more than you would get from a speed doubling. To have a good chance (like 86%) that a speed increase of (say) 21%, (~21 Elo or 3% result improvement, i.e. expected match result 53%) would result in winning a match you would need about 170 games.