
From:  bob koca 
Subject:  [Buggnubg] these rollout results don't make sense 
Date:  Sun, 26 Sep 2004 19:51:50 0400 
1) If I do a rollout of a cube action typically the value for cubeless when the cube is centered is not exactly half of the value for cubeless when gnubg owns a two cube. It is close to being exactly half but the differnce is often more than just a roundoff error in the last decimal place.
What is going on there? Is it not the same random number stream? If it isn't shouldn't it be?
2) I rolled out a position which gnubg says is a very clear double and close take. For the rollout I started with the cube in the middle. I got the values cubeless equity of .585 and cubeful equity .865. The gammon rates were about 25% for both sides so the cubeful number seemed low based on the cubeless equity.
So I did the rollout again, but this time starting with the cube already having been turned. This time I got a cubeless value of .595 and a cubeful value of .487 which goes to .974 cubeful when multiplied by 2. That cubeful seemed about right based on the cubeless equity and the gammon rates.
Shouldn't the cubeful numbers be the same regardless of which way I set it up since gnubg would just cube at the first roll for the centered cube situation anyways?
I am not a programmer so I can't check but I have two guesses; i) I am really confused by what is meant by cubeless ii) gnubg doesn't consider doubling until the second opportunity.
thanks in advance for any help, Bob Koca
[Prev in Thread]  Current Thread  [Next in Thread] 