为决策树创建我自己的标准函数

Question

I'm using the sklearn DecisionTreeClassifier and I would like to create my own criteria function (by default you can use gini or entropy, but it's not what i'm looking for). Something like that :

clf = DecisionTreeClassifier( criterion = 'my_function')

Is it possible to do that ?

Is there a similar algorithm that allows to do it (in Python or R ) ?

Thanks.

Answer 1

For R, you can use the rpart package . In particular, see the User Written Split Functions vignette. Despite having limited decision tree experience, I was able to follow the examples to handle a multivariate output using a custom algorithm.

However, note that the built-in classifiers use a fast external library, so if you write your algorithm in plain R, the processing may be considerably slower. As the vignette notes, cross-validation is disabled by default due to this expected slowdown. If your data is "small" enough (or you can wait a bit longer), this may not be an issue in your case.