How do I set the desired treshold at the model training stage?

Hi, everybody, I have an algorithm: 

clf = LGBMClassifier()
clf.fit(X_train, y_train)

When I use this algorithm to predict the probabilities of a binary classification clf.predict(X_test), the algorithm uses treshold=0.5 to determine the classification class. I can change the treshold when using predict_proba, but this option is not very suitable for me.

I am interested in whether it is possible to set the treshold value I need at the fit stage (training the model)? I.e. for example, I want that when treshold>=0.6, class 1 was predicted, and class 2 was predicted for treshold fit

Author: ivan100096, 2020-03-05
Source

1 answers

Solved the question: 

class_distrib = {0:treshold, 1:1-treshold}
clf = LGBMClassifier(class_weight = class_distrib)
clf.fit(X_train, y_train)

Thank you for your answers

 0
Author: ivan100096, 2020-03-05 15:01:26