| |
BGonline.org Forums
Long / Close Rollouts
Posted By: David Rockwell
Date: Wednesday, 3 June 2009, at 7:35 p.m.
I was glad to see the rollout for the 64 opening extended to 139k. The more rollouts I do, the more I appreciate how much variability can be left long rollouts
I think that once we reach ridiculous lengths such as we have with the 64 opening, there may be a better approach. That is, to compile the rollout from a weighted average of its 21 component parts (i.e the rollout for the 21 different subsequent rolls). The chief advantage of this is that the "correct" response can be used in the rollout rather than the GNU 2 ply evaluation used in the basic rollout which is not always correct.
For example, we know that the best response 41$ 21 is to slot. However, GNU 2 ply prefers split. The difference between these two is approximately .04. When a rollout is done of 41$, GNU incorrectly splits instead of slotting on a 21 response. I have done a little investigating and have estimated the potential error in the rollout due to these errors in the 1st ply to be between .0000 and .0055 for the small sample that I have. In general, this is small enough to ignore. However, with moves such as the opening 64, this is large enough to potentially change the conclusion.
There are unresolved technical issues with using this approach which I won't take the time to list here. And, there is the question of whether this exercise is worth the trouble. It probably is not. But, if we are going to go to take time to do a 139k rollout, it might be worth investigating as a supplement. I am sure that going deeper will get us closer to the truth than going longer. And, most of the supporting rollouts are available already anyway.
I'm working on this very slowly since I'm currently on a strict diet of study which will lower my ER in the short term. Improving my play of the opening 64 isn't likely to make the list. But it IS very interesting.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.