Variable Transform in Shifu
Such transform types are useful in Neural Network and Logistic Regression models.
- Woe transform is used to transform numerical values into discrete values;
- 'ln' function in woe implicit some normalization.
- Better model stability after woe transform.
- One-Hot Encoding is a good choice, while so far it is not supported in Shifu
- In Shifu, positive rate is used for each categorical feature.
- In tree models, no need transform categorical features to numerical.
