BGonline.org Forums

Apples and oranges

Posted By: Chuck Bower
Date: Tuesday, 26 May 2009, at 4:37 p.m.

In Response To: An example of donkage + false ER (Stick)

My experience is that error rates for rolled out matches are considerably different than evaluated matches. Obviously selective rollout analyses are a third variation.

We should all realize that the small error rates reported are going to be worse if all plays are rolled out, since (as somewhere here -- probably Stick or UBK or... -- has pointed out) sometimes the bot eval agrees with the human but both are wrong (when rolled out). Few humans ever roll out their 'correct' plays.

Messages In This Thread

• An example of donkage + false ER
  Stick -- Friday, 15 May 2009, at 8:29 p.m.
  • An example of donkage + false ER
    neilkaz -- Saturday, 16 May 2009, at 2:46 p.m.
  • Apples and oranges
    Chuck Bower -- Tuesday, 26 May 2009, at 4:37 p.m.