BGonline.org Forums

An alternative to error rate

Posted By: NJ
Date: Sunday, 27 September 2009, at 10:56 p.m.

In Response To: An alternative to error rate (Timothy Chow)

In my iphone app, I created a statistic called "Checker Play Skill". I wanted it to account for forced moves and easy plays. The way I computed it is this:

For each play:

Error = equity difference from best move

Average Error = average equity difference from best move (i.e. take all possible moves and average all the errors).

Then after a game:

Checker play skill = 1 - Sum(errors)/Sum(average errors)

So playing all the best moves will result in checker play skill of 100% and playing moves at random will result in checker play skill of 0%.

Notice that for forced moves and moves which do not matter (such as when one side is certain to win), error = 0 and average error = 0. So these do not contribute to the final checker play skill statistic. Note also that moves that have a larger average error are weighted more than moves with lower average error. This is what differentiates "easy" plays from "difficult" plays.

This isn't exactly perfect. Just because the average error is large doesn't necessarily mean that the play is difficult. For example, it could be that a hit is the only good play to make and every other play is terrible. With my system, the average error for that play would be large but the correct play would be easy to find. Because of this, I capped the average error for each play at 0.5 for contact positions and 0.1 for race positions. I also weighted race moves 3x compared to contact moves because otherwise race moves would not count enough (because the errors and average errors are so small).

I'm not sure about your proposed W3 stat. It seems a little too limited because it only takes your three worst moves into account. Perhaps it would be useful for top players, but for mortals like me I am sure that I make so many blunders that my W3 rating would never budge. Perhaps a modification where it took your top N mistakes and weighted them. So for example if N were 10 and the weightings were something like:

 

1: 3.3% 

2: 6.6% 

3: 10.0% 

4: 13.3% 

5: 16.6% 

6: 16.6% 

7: 13.3% 

8: 10.0% 

9: 6.6% 

10: 3.3% 

Messages In This Thread

• An alternative to error rate
  Timothy Chow -- Sunday, 27 September 2009, at 7:17 p.m.
  • An alternative to error rate
    Rich Munitz -- Sunday, 27 September 2009, at 7:29 p.m.
  • An alternative to error rate
    Phil Simborg -- Sunday, 27 September 2009, at 9:55 p.m.
  • An alternative to error rate
    NJ -- Sunday, 27 September 2009, at 10:56 p.m.