Sklearn - predict_proba equivalents

python numpy scikit-learn

530 观看

1回复

361 作者的声誉

So some of the models in Sci-kit learn such as Logistic Regression support the predict_proba method which I use heavily. Is there an other way for models such as Lasso to output a similar probability array, since they don't support predict_proba?

Also: I'm working with a three outcome dataset where the probabilities between the outcomes can be relatively even - any general suggestions for models/tunings to try to improve probability precision, that can handle 50+ feature columns? =)

Thanks!

作者: Daniel Slätt 的来源 发布者: 2017 年 12 月 27 日

回应 1


1

20505 作者的声誉

Sklearn introduced Probability calibration exactly for this purpose. Improving or adding support for classifiers without a natural probability-output.

There is also a blog-post about this.

Usage will be based on CalibratedClassifierCV.

Of those two methods, sigmoid and isotonic, the former is quite popular as the underlying method behind libsvm's probability-output, which you can see in sklearn's wrapper SVC

作者: sascha 发布者: 2017 年 12 月 27 日
32x32