b******e
发帖数: 445 | 1 I have data for some contest result. I have to come up some score for each
id.
data looks like -
id winning% tie% loss%
1 .33 .12 .55
2 .33 .34 .33
3 .20 .40 .40
....
I don't want to use something just like: winning%+tie%- loss%
ideally the score should range [0, 100] , the higher the value the higher
percentage of winning.
thanks |
k*******a
发帖数: 772 | |
b******e
发帖数: 445 | 3 i am thinking (win+tie)/loss.
but how can i make 100% loss get a score 0; and 100% win get a score 100. |
w*******9
发帖数: 1433 | 4 (win + 0.5*tie)? If you have game level data, try Bradley-Terry model.
【在 b******e 的大作中提到】
: i am thinking (win+tie)/loss.
: but how can i make 100% loss get a score 0; and 100% win get a score 100.
|
b******e
发帖数: 445 | 5 总共只有2个team 也可以用Bradley-terry model吗？
★ 发自iPhone App: ChineseWeb 8.2.2
【在 w*******9 的大作中提到】
: (win + 0.5*tie)? If you have game level data, try Bradley-Terry model.
|
m****D
发帖数: 686 | 6 How about (2win+tie)/(2win+tie+2loss)?
If you know the difficulty level of the contest or the following effect of
the contest, more reasonable weights can be generated. |
