[ExI] elo system for two-player game rankings
spike66 at att.net
Sun Dec 16 18:47:27 UTC 2007
A couple years ago Anders asked about rating systems that might be
applicable to a web based simulation game he was designing. Chess players
use the Elo system, which is described in detail, along with the equations
for calculation here:
This system can be used for any two player game.
I am working on a macro that would calculate Elos for boxers. [Am I the
only one here who sees the interesting strategy in boxing? It isn't merely
two enormous galoots pounding each other senseless to the point of drain
brammage. That is merely a curiously entertaining side-effect.]
Using the equations given in the article, one can take a subset of the
players' records and still come up with a reasonably accurate estimate of
their Elo over time. This can be proven thus: take the performances of the
top rated chess players and look at just one annual tournament in which they
met. Using this subset of data, one can use the Elo equations and calculate
a ratings, then compare with actual ratings based on the full dataset.
The rating system in dependent on time. If I had defeated Kasparov when he
was an eight year old expert, for instance, its meaning to my rating is far
different from if I defeated him now. (I see that Putin has defeated
One can take the professional records of one's favorite couple dozen boxers
(which are readily available online, along with who managed to inflict their
few defeats and the dates thereon) then use this data to create a three
dimensional hash table. Any time-slice of the three dimensional data
structure resembles a chess tournament hash table. Most of the hash table
is empty at any time slice, but my claim is that I can derive a reasonable
estimate of any boxer's Elo at any arbitrary date.
This isn't ready for prime time, but I invite others to attempt to create
such a system with your favorite game.
More information about the extropy-chat