Parallel for HDDM's `find_starting_values()` in HSSM

[jainraj]

The HDDM library has a method called find_starting_values(), which apparently helps if parameters don't converge (I never used HDDM personally, so I can only assume).

I am wondering:

• Is there a similar method in the HSSM library?
• If not, are you planning to build one?
• If not, what are the constraints? Is HSSM so general that such a method may not ever exist?

[AlexanderFengler]

find_starting_values() basically ran the equivalent of find_MAP() in PyMC, to initialize the chains. Initializing chains with MAP values is discouraged from a theoretical viewpoint, however we are thinking of adding this feature again because in practice it can de facto help.

However by "adding this feature" I just mean creating a convenience wrapper around manually calling myhssmmodel.pymc_model.find_MAP() and assigning the results as starting values to the chains, which you can in principle do already.

[jainraj]

Awesome! Thanks for this!