[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Comparing rollouts A, B, and C

Posted By: Taper_Mike
Date: Wednesday, 17 July 2013, at 3:58 p.m.

In Response To: How to set up Rollout A for 1296 trials, no truncation (Seth)

Seth: What's the difference between a and b-obviously, b less trials-I mean how accurate is rollout b compared to a rollout using a?

I am not sure I am qualified to answer your question.

There is a tradeoff between the number of trials and the number of plies. Let us assume variance reduction is used, which should always be the case. Given sufficiently many trials, more plies will always be better than fewer plies. When the number of trials is small, however, as it is in these truncs, fewer plies may yield better results.

The extreme example is a single game played to completion. That’s a rollout with 1 trial. Obviously, the equity of the result should not be taken as the equity of a move on the 3rd roll. You would do better just to use GnuBg’s 2-ply eval, which is the equivalent of a 1-ply trunc.

Performance is also an issue. 7-ply truncs take longer than 5-ply.

My guess is that answering your question would require extensive testing, both with full rollouts and truncs. What has been done, and what conclusions can be drawn, I can’t say.

With the bliss that ignorance brings, I have happily accepted Xavier’s judgement that a reasonable balance is to use rollout B for checker plays, and rollout C for cube decisions.

Mike

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

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