BGonline.org Forums

GNUbg JSD and rollout settings

Posted By: leobueno
Date: Sunday, 17 April 2011, at 8:39 p.m.

Tim, Neil and others here have convinced me that GNUbg rollouts should be at least 2-ply.

I realize that the JSD setting gives us an indication of the confidence with which we can say that play A is better than play B.

The default GNUbg setting, for stopping rollouts when JSD reaches a given level, is 3. However, I recall reading a thread which showed that a JSD of 3 may not be enough (as shown by a deeper and longer rollout of a position, I think).

I recall also seeing a posting (by Tom Keith?) that indicated that the way to gauge the confidence that play A is superior to play B, is to divide the equitiy differnce by the JSD, and then compare it to a table which provides the confidence level.

A couple of questions come to mind.

1. Why doesn't GNUbg use that formula (confidence derived from the equity divided by JSD and the table) as the rollount stopping paramenter instead of the raw JSD?

2. How much better generally is setting the JSD cutoff at 4, instead of 3?

3. What about doing 2-ply rollouts including extra candidate plays and stopping when JSD is 4, but with a minimium number of trials of 36?

Thanks for your thoughts.

Messages In This Thread

• GNUbg JSD and rollout settings
  leobueno -- Sunday, 17 April 2011, at 8:39 p.m.
  • GNUbg JSD and rollout settings
    Tom Keith -- Sunday, 17 April 2011, at 9:45 p.m.
  • GNUbg JSD and rollout settings
    mtuhtan -- Monday, 18 April 2011, at 5:19 a.m.
  • GNUbg JSD and rollout settings (a request to Timothy or Maik)
    Philippe Michel -- Wednesday, 20 April 2011, at 4:02 p.m.
    • GNUbg JSD and rollout settings (a request to Timothy or Maik)
      Maik Stiebler -- Wednesday, 20 April 2011, at 7:20 p.m.
      • GNUbg JSD and rollout settings (a request to Timothy or Maik)
        Philippe Michel -- Wednesday, 20 April 2011, at 8:26 p.m.