Diagonal log-likelihood function diag_logll

[dhanajitb]

In this line,

pytorch-ensembles/utils/swag_model.py

Line 231 in 76b71c9

logprob += Normal(mean, scale).log_prob(param).sum()

Isn't scale supposed to be the square root of variance?

So, it seems that lines 230-231 should be the following instead:

for param, mean, var in zip(param_list, mean_list, var_list):
    logprob += Normal(mean, var**(0.5)).log_prob(param).sum()

[alexlyzhov]

Yes, sounds right to me.