t*****2 发帖数: 94 | 1 如果logistic回归自变量x不是线性的,怎么办?
谢谢各位大侠了 | h***x 发帖数: 586 | 2 transformation
【在 t*****2 的大作中提到】 : 如果logistic回归自变量x不是线性的,怎么办? : 谢谢各位大侠了
| k*******a 发帖数: 772 | | t*****2 发帖数: 94 | 4 你能举个例子吗?
【在 h***x 的大作中提到】 : transformation
| t*********e 发帖数: 71 | 5 binning your variable, no more than 100 | h***x 发帖数: 586 | 6 Some nonlinear transformations such as log, sqrt ...
and the binning mentioned by LS. If your dependent variable is category, and
predictor is continues, group the variable into 10 or 20 bins, calculate
response rate in each bin. Combine adjacent bins with same response rate.
This method works better than nonlinear transformation, but has its own
problems.
Both methods can be automated easily...
【在 t*****2 的大作中提到】 : 你能举个例子吗?
| n*****3 发帖数: 1584 | 7
then how to interpret it to the clients? what if
the log and sqrt gives you diff stories?
and
You sure lose info/power this way.
【在 h***x 的大作中提到】 : Some nonlinear transformations such as log, sqrt ... : and the binning mentioned by LS. If your dependent variable is category, and : predictor is continues, group the variable into 10 or 20 bins, calculate : response rate in each bin. Combine adjacent bins with same response rate. : This method works better than nonlinear transformation, but has its own : problems. : Both methods can be automated easily...
| h***x 发帖数: 586 | 8 那你给一种nonlinear关系下不丢失信息的线性fitting方法?
实际上就是根据dependent variable和predictor直接的非线性关系进行线性分段拟合
,你没有意识到而已。
【在 n*****3 的大作中提到】 : : then how to interpret it to the clients? what if : the log and sqrt gives you diff stories? : and : You sure lose info/power this way.
|
| A*******s 发帖数: 3942 | 9 my 2 cents--GAM might be the best choice for nonlinear modeling in business
for a few reasons, say
can use cross validation to govern the trace of the penalization matrix and
avoid overfitting, which is a big concern of nonlinear modeling;
easy to illustrate the marginal effect of each variable, which is not
available in MARS and CART.
Additive assumption make some ad-hoc tasks much easier. say convenient
missing imputation (may not theoretically sound tho), "neutralizing" one
variable, etc... | n*****3 发帖数: 1584 | 10 I am not a expert on this topic,
but I think SVM could be a better choice here,
since you are talking about a non-linear logistic regression here.
agree, Generalized additive model can be a good choice too.
business
and
【在 A*******s 的大作中提到】 : my 2 cents--GAM might be the best choice for nonlinear modeling in business : for a few reasons, say : can use cross validation to govern the trace of the penalization matrix and : avoid overfitting, which is a big concern of nonlinear modeling; : easy to illustrate the marginal effect of each variable, which is not : available in MARS and CART. : Additive assumption make some ad-hoc tasks much easier. say convenient : missing imputation (may not theoretically sound tho), "neutralizing" one : variable, etc...
| h***i 发帖数: 3844 | 11 上非参
【在 t*****2 的大作中提到】 : 如果logistic回归自变量x不是线性的,怎么办? : 谢谢各位大侠了
| A*******s 发帖数: 3942 | 12 If we only care predictive performance, yes I do agree SVM, or more common,
kernel method, is a better way to capture nonlinear functional form. But as
I said in "business modeling", kernel method has no interpretability and is
hence less appealing.
I am not a expert on this topic,but I think SVM could be a better choice
here,since you ........
【在 n*****3 的大作中提到】 : I am not a expert on this topic, : but I think SVM could be a better choice here, : since you are talking about a non-linear logistic regression here. : agree, Generalized additive model can be a good choice too. : : business : and
| s*********e 发帖数: 1051 | 13 gam没办法确保单调,另外scoring code不portable.
business
and
【在 A*******s 的大作中提到】 : my 2 cents--GAM might be the best choice for nonlinear modeling in business : for a few reasons, say : can use cross validation to govern the trace of the penalization matrix and : avoid overfitting, which is a big concern of nonlinear modeling; : easy to illustrate the marginal effect of each variable, which is not : available in MARS and CART. : Additive assumption make some ad-hoc tasks much easier. say convenient : missing imputation (may not theoretically sound tho), "neutralizing" one : variable, etc...
|
|