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

BGonline.org Forums

Multi pass and PR

Posted By: Karol Szczerek
Date: Saturday, 31 December 2016, at 5:17 p.m.

In Response To: Multi pass and PR (Bob Koca)

2nd pass can enlarge the error discovered in 1st pass for sure, but it's not what's relevant here.

If we are comparing what I understand we do, which is the case where the 2-step algorithm uses the same strength on its 2nd pass as the 1-step does for its single pass, then the error in 2-step will never be bigger than in 1-step.

Set of moves analyzed by 2nd pass in 2-step algorithm are included in set of moves analyzed in 1-step algorithm. Therefore the 2-step has to have lower error sum.

Didn't hard-checked it right now, but from what I remember and understand from XG's algorithm description, if the human & bot (at filter strength) choose the same play and it's best after 1st pass, it isn't passed again.

The same happens when human&bot agree but the play is slightly worse (eq diff is lower than value set for "error" - typically .02) than the top one after 1st pass.

There are probably several cases more.

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.