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

BGonline.org Forums

Clarification about Janowski's formulas

Posted By: Daniel Murphy
Date: Monday, 25 January 2010, at 5:13 p.m.

In Response To: Clarification about Janowski's formulas (Timothy Chow)

I did not mean to imply that we can now assume that bots play the checkers perfectly, leaving only cube action in rollouts needing expert scrutiny.

Sometimes, in response to play X, we can identify an immediate error in response. So then, if instead of rolling out play X, we'd get a better answer by rolling out all 21 responses. Better, but not necessarily correct, since we still have no guarantee that the bot will not make the same mistake (or another) on a subsequent turn.

An example is Bob's post 59008. After some plays, some bots on some plies may immediately and incorrectly split. We can roll out the splitting and nonsplitting responses, but having done that, our results may still not be correct, because they may reflect the possibility that the bot is making the same mistake further on.

A: My 46,656 trial rollout says play X is 0.010 better than play Y.

B: Impressive! Does your bot play perfectly?

A: No, but it plays both sides, so mistakes even out, so the result is correct.

B: And your proof of that assertion is .... ?

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.