| |
BGonline.org Forums
The Great Rollout Conspiracy
Posted By: Peter Edwards
Date: Saturday, 27 September 2008, at 3:33 p.m.
Before I show my ignorance, allow me to confess. I am a newbie.
Why, oh why do we set such store by rollouts? Stick’s excellent web pages give us rollouts for opening rolls and replies. He rolls out positions at world class level (Snowie 3-ply and Gnu 2-ply) using 46656 trials. However, Gnu can assess a single position at 3-ply, 4-ply or higher, all far better than world class level. Average the equities at 3-ply and 4-ply and you get an assessment of the possible plays. Check by averaging the equities at 4-ply and 5-ply and you get a very similar result.
Here is an example: look at the rollout results for an opening 12 for money. Stick’s results show that the equities of the two front runners are both close to zero, with the slot marginally favoured over the split. Indeed the plays are so close that either could be better. Now average the same plays with 4- and 5-ply analysis and the two candidate plays are still close to zero, but now the split is marginally preferred. Can anybody put their hand on their heart and claim to know for certain which is the long term winner? I thought not.
The real question is how should a play be assessed? Should one look ahead four or five plies, examine all the options and then make an assessment? Alternatively, is it preferable to play out the position to a conclusion, thousands of times, but only at world class level? In the case of an opening 12 roll, it hardly matters which of the two candidate plays is preferred. Complicated mid-game positions are another matter entirely. Rollouts frequently contradict assessments, and the rollout results are invariably believed. Why? I suspect a conspiracy.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.