some questions about the code

[hanruisong00]

Hello, your code has provided me with great inspiration for my work. I also have some questions to ask you. Firstly, why do we need to calculate the self.cv_squared loss for both importance and load? Secondly, in _prob_in_top_k, can we replace prob_if_out with 1 - prob_if_in?