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

BGonline.org Forums

Introduction & ND vs. D/T win probabilities question

Posted By: Chuck Bower
Date: Tuesday, 11 September 2012, at 4:59 p.m.

In Response To: Introduction & ND vs. D/T win probabilities question (Joshua Small)

Welcome, Joshua! You've come to the right place if you are interested in learning the game at a high level.

From my understanding, the modern bots (XG and GNU-bg) stop a rollout trial when they reach a double/pass decision. So if you enter a position that is (in the bot's opinion) D+P, then on the NoDouble branch of the rollout it likely will only run one play sequence, then evaluate the position as a pass, and use it's evaluation W/G/BG values for that trial.

For the D+T branch, if it's really ND then quite a few plays should be needed before arriving at a D+P (ending the rollouts for that trial) and if it's actually a D+P, you get a LOT of plays since the taker has to turn the game around (or it gets played to conclusion). If the position is D+T, it also should take several plays to reach a decision for that trial.

What all this means (IMO) is that the D+T result posted by the bot typically has more rollout plays per trial and therefore should be more reliable in terms of W/G/BG breakdowns. Sometimes the ND branch isn't much more reliable than simply an evaluation -- i.e. you don't get the power of the rollout.

I assume this is why Caspar recommended rolling out without a cube in play. I don't think that is an option in XG. (It is an option in GNU-bg.)

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.