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

BGonline.org Forums

Long / Close Rollouts

Posted By: David Rockwell
Date: Wednesday, 3 June 2009, at 10:25 p.m.

In Response To: Long / Close Rollouts (Rich Munitz)

It isn't all that tough to compile the component parts. And, the information collected is valuable in itself. I'll share my spreadsheets with you offline at some point if you are interested. The only reason not to do this exercise is that that the numbers involved are usually very small.

Let me be clear that I'm Stick's biggest fan with what he is doing with the opening. I think there is great value in it and I think it is almost always going to lead to the right answer. It certainly isn't a waste of time. I just think we may be able to find a few instances where the rollouts do NOT lead to the right answer if we take existing information and apply a little elbow grease.

Actually, one doesn't really have to have all component parts to do this exercise. If you have rollouts for the plays which GNU gets wrong, you can just substitute the correct move's equity for the GNU 2-ply move's equity (properly weighted of course) and the equity difference between move choices should be very close to what you would get with a complete compilation.

What interests me most is finding systematic errors in GNUs evaluation. For example, my first pass at 41S and 41$ show that GNU 2-ply is making some pretty big mistakes after the slot but is almost perfect after the split. If this patterns holds up to further scrutiny, it tells us something about how to evaluate similar split versus slot choices when the rollouts are close. For example, the rollouts show that 21$ is best at Gammon Save by a small margin (2.7). I've seen enough to believe that fixing GNUs slotting errors might reverse this result favoring the split.

I quickly took a look at 64 a while ago and it appeared that the 2-ply errors were not uniform across the three choices. I believe that Run gained the most from GNU's mistakes. So, it seems to me that drilling down a level will increase accuracy more than letting GNU pound away making the same errors with every repetition.

As for your wish list, yes that would be great. And, it would only be a matter of capturing data that was already compiled.

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.