Init values with PyStan and variational inference

I would like to provide starting values for the variational distribution in PyStan but I don’t quite understand how to do it. For sampling-based algorithms it is straightforward but it is not obvious for mean-field and full-rank.
In my case, I can get good starting values for the mean-field distribution using Laplace-style approximations.
The ‘vb’ function returns a dictionary with a ‘init’ key but, as far as I can understand, it does not look like the variational parameters.
Any help would be appreciated.

Hi, sorry for not answering this one earlier.

I think init takes the same input as does .sampling.

cc. @Bob_Carpenter Do we have documentation for the variational inference init?

Hi, did you manage to get it working or do you still need some help?

I have the same question for PyStan.