| |
BGonline.org Forums
An alternative to error rate
Posted By: Timothy Chow
Date: Sunday, 27 September 2009, at 7:17 p.m.
While idly pondering the concept of "error rate," I came up with another possible measure of the quality of one's play: The 3rd largest error you've made in your last 50 games.
Obviously the numbers 3 and 50 can be adjusted according to taste, but the idea is clear. The usual definition of ER has some well-known difficulties:
1. It seems that forced plays and easy plays shouldn't count, but how do you decide which plays are "easy"?
2. Occasionally an "erroneous" play may actually be "correct" when you take into account the players' levels of skill.
3. Small "errors" may not really be errors because the bots aren't perfect.
4. Rolling out every move is too much work.
These difficulties are largely addressed by my new measure, which perhaps we can call "W3" (for "Whopper number 3"). Small "errors" will not affect your W3. Deliberate "traps" that are set to catch an unwary opponent will also rarely affect your W3, because it will be rare that a trap that loses a really large amount of equity will actually be a good idea. And since only a few moves will be candidates for your W3, you can afford to roll them out.
The reason for choosing W3 rather than W1 is to eliminate outliers that may occur under very atypical circumstances. For example, say you do something really stupid like miss an obvious hit. If something that bad happens to you only very rarely, then the W3 will forgive you. But if you do it three times then it counts as a pattern of poor play.
Obviously I don't expect W3 to replace ER, but if anyone decides to try using W3 as a tool for monitoring their own performance, I would be curious to hear whether it seems to be useful.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.