Hi authors, thanks for your work.

I have a question in Prior_Sup, when y has uniform prior.

Why isn't log_p_y = -math.log(self.num_data)? I understand this value won't affect overall training. Can you help with it?