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

BGonline.org Forums

In defense of single trial rollouts

Posted By: Tom Keith
Date: Sunday, 17 October 2010, at 7:46 p.m.

In Response To: Minimum # of Trials and Confidence (Timothy Chow)

This is a little off topic, but I thought I'd mention that I do single trial rollouts "all the time". Yes, they're not as accurate as multiple trial rollouts, but single trial rollouts have a couple of unique advantages:

(1) Single trial rollouts are very fast. And not just because of the low number of trials. If you have a recorded game that you want to roll out, you can roll out all the positions of the game with one rollout. (If your game has 60 rolls in it, you are achieving 60 times faster results than doing a separate rollout of each position.)

(2) Single trial rollouts allow for a second type of variance reduction that isn't available with multiple trials. (This applies only if you are rolling out out the whole game.) We know the equity of the starting position of the game is zero. If your rollout gives a equity for the starting position that is higher than zero, then the first level of variance reduction must have added too much to that player's side overall. You can adjust the value of that side downward and get a more accurate equity for that position and all subsequent positions. (This gives perfect accuracy for the starting position; improved accuracy, on average, for all later positions.)

I admit that these features may not be applicable for everyday rollouts using standard bots. But I can assure you that single trial rollouts are indeed very useful in certain applications.

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.