Backgammon Rollouts

Rollouts

Reporting results of rollouts

From:   David Montgomery
Address:   monty@cs.umd.edu
Date:   14 June 1995
Subject:   Re: comments on instructional materials
Forum:   rec.games.backgammon
Google:   3rn5c5$ps2@twix.cs.umd.edu

> 1) It is sometimes not clear to me what cube protocol was used > when rollout results are mentioned. Are there some conventions > I am missing? I don't think there are. Nor can there be, really, as different positions call for different approaches, and as the computer tools for analyzing backgammon positions become more sophisticated, there will be even more options. I would like to encourage everyone (both here and in backgammon publications) to specify the details when giving rollout results. Assuming a JF rollout, that would mean: 1) identifying the program (JF, not EXBG) and version (if there is more than one); 2) the number of rollouts; 3) truncated or complete rollouts; 4) if truncated, the horizon; 5) if not truncated, the settlement limit and the results with the cube in the relevant positions, including cubeless results; 6) if a play vs. play comparison, whether the same random seed was used for all plays (note that JF doesn't do this unless you specifically enter a non-zero seed). All of this would really only take one sentence, like: "For each play I did 1296 20-ply truncated rollouts with JF 1.0, using the same random seed for each." > 2) One thing I've been trying to do recently is expand my library > of reference positions for over the board use. Are there any good > annotated matches available in which comments include how expert players > use these. For example, when to be in a reference position mindset and > how to adjust for differences between reference and actual position. I haven't seen anything like this in the literature. There are only a few references to reference positions in the annotated matches I've read and even then the comments usually go like: "I knew this position [reference position] was a take, and the game position looks a little better for me, so I took." Seems like an interesting idea, although I'm not sure if very much in the way of general comments could be made. Beyond establishing reference positions, an important goal is to determine for each position what the relevant features are, and how important changing each of them is to the correct strategy (for checker play) or equity (for cube play). And a goal beyond that is to find categories of positions such that within a category, the features behave in the same way. David Montgomery monty on FIBS

Did you find the information in this article useful?

Do you have any comments you'd like to add?