A Question Regarding the Bivariate Normal Distribution Loss

[DKandrew]

At line https://github.com/apple/ml-multiple-futures-prediction/blob/402f82d3d4969c51b71ac9a2d63f0ef4818c4462/multiple_futures_prediction/my_utils.py#L51-L52

For here we want to compute P(Y|Z, X) which is a Bivariate Gaussian Distribution, but the loss function does not look like a negative log likelihood of Gaussian to me. First, the variable ohr is very odd, shouldn't it be sqrt(1-square(rho))? Second, it seems like all the denominators in the exponent component are multiplying with the nominator. So they are very strange.

I am using the definition of Bivariate Gaussian here: https://en.wikipedia.org/wiki/Multivariate_normal_distribution#Bivariate_case