differential-privacy-library
differential-privacy-library copied to clipboard
predict_proba method
The predict_proba method of logistic regression (and I think maybe for all models) simply outputs 1 for the predicted label and 0 otherwise.
Hi there,
It seems that the logistic regression model does indeed give a binary 0 or 1 output when using predict_proba in some instances, however there are occasions when a more granular output is produced (in the cases I have considered, this happens for large epsilon). This will require further investigation. If you have any code examples you can share, that would be helpful.