c*******o
发帖数: 8869 | 1 用logistic regression 得到了predicted probability(for case=1), 结果在
predicted likelihood 很高 (0.8-1) 的区间 和很低的区间 （0-0.1), case
enrichment 很高， 而在中间区 （0.5左右）， case 的比例很底， 为什么predicted
likelihood 会有这种curvature的情况， 如何处理？
跪谢了。 | h***x
发帖数: 586 | 2 Check you data if there is any variable in your model has missing value ...
No need 跪谢, need baozi if helps.
predicted
【在 c*******o 的大作中提到】
: 用logistic regression 得到了predicted probability(for case=1), 结果在
: predicted likelihood 很高 (0.8-1) 的区间 和很低的区间 （0-0.1), case
: enrichment 很高， 而在中间区 （0.5左右）， case 的比例很底， 为什么predicted
: likelihood 会有这种curvature的情况， 如何处理？
: 跪谢了。
| c*******o
发帖数: 8869 | 3 I have a lots of zero, in another word, matrix is sparse. Is that the reason?
【在 h***x 的大作中提到】
: Check you data if there is any variable in your model has missing value ...
: No need 跪谢, need baozi if helps.
:
: predicted
| h***x
发帖数: 586 | 4 According to my understanding to your question, you are saying that in the
top 2 deciles and bottom decile, there are lots of targets (you name it case
) while in the middle, fewer targets.
The distribution of the targets is based on the scores. The reason why lots
of targets are in the bottom is because they have low scores. So you need to
check the variable values for those records.
As you said, you have a lots of zero, that should explain the reason.
reason?
【在 c*******o 的大作中提到】
: I have a lots of zero, in another word, matrix is sparse. Is that the reason?
| s*r
发帖数: 2757 | 5 would a quadratic term be helpful | p*****y
发帖数: 34 | 6 I guess regular logistic regression can't deal with sparse case
reason?
【在 c*******o 的大作中提到】
: I have a lots of zero, in another word, matrix is sparse. Is that the reason?
|
|
