
BGonline.org Forums
Jacobs/Trice Fish book  a proposed computer experiment
Posted By: Timothy Chow
Date: Thursday, 27 November 2014, at 3:16 p.m.
While playing with Murat the Troll over on r.g.b., I was led to consider the following computer experiment. I'm not adept enough with GNU to actually carry it out, but I hope someone else will attempt it.
The idea is to play GNU against itself with different settings. Call the two sides GNU A and GNU B. GNU A is simply GNU 2ply with a standard MET. GNU B, however, has noise added to its checker plays (but not its cube decisions), and uses a fish MET. The exact choice of noise level and the exact choice of fish MET will probably need to be tuned experimentally, but the idea is to pick a fish MET that will compensate for the checkerplay errors. The goal is to pick a combination that will cause the standard formulas for converting error rate into Elo difference (or more precisely, odds of GNU A beating GNU B) to be as inaccurate as possible.
The goal of the experiment is twofold. First, it will give a quantitative estimate of how much a weaker player can hope to benefit from aggressive cube play. (The estimate won't be perfect of course, since human players make cube errors, and don't make "random" checkerplay errors, but it should give us some idea.) Second, it will give us some idea of how robust the standard formulas for converting between error rate and Elo are.
Anyone game?

BGonline.org Forums is maintained by Stick with WebBBS 5.12.