Steve Gordon posted to the WGP mailing list:
Date: Thu, 29 Mar 2001 06:47:06 -0700
Subject: Re: [wgp] probability of drawing tiles
I would like to point out a different consequence of the fact that the number of different racks is "manageable" (one that was first suggested years ago, I believe by John Chew):
It is quite feasible to build a file of opening moves. This file would simply list the best opening move to make with each of these different racks. (Note that computing a unique index from the rack itself would mean that this file could just be an indexed array of opening moves.)
While this approach begs the question of how to decide what the "best" opening move is given any particular rack, making this decision before the game has to give a better answer than trying to decide during the game (because there is more time available to compute the answer). And best of all, it would then take virtually no time during the game to just lookup the move in the file.
We could build this file as a community, by running the same simulation program on each of our computers on a different subset of these racks. For many racks the choice will be clear-cut, but for others the "best" move might simulate only a fraction of a point better than the "second best" move. The "real" best move for those racks could make for interesting discussions for years to come. Still, I reiterate, no matter how one decides the best opening move, it is better to do the computations involved before the game than during the game.
This file could also be used in reverse on the second move of each game. The program could find all racks for which the opponent's first move was the "best" move, and use this to infer the most likely rack leaves for the opponent.
So, if anyone wants to help build this, here's what to do.
AAAAAAA AAAAAAB AAAAAAC ...
AAAAAAA -AAAAAA/MN AAAAAAB -AAAAA/MN;ABA 8F/SMN/15 ...
to indicate that Maven (M) with its North American (N) lexicon recommends trading those As, or playing ABA at 8F with a loss of 15 points of equity when calculated by simulated (S) Maven. Blanks to be written in lower case, other simulation utilities to have short names registered here.
Finally, we need someone to look after receiving submissions to the database and coordinating matters to reduce duplication of effort. I'll do both until (1) someone else steps forward or (2) it gets to be too much work for me to handle.