Breaking M&M prototyping to using DSC¶

This is an attempt to use DSC in a novel way.

Specifically I've got various components of the model written in R and Python. Some part of it is already implemented elsewhere (mashr package for some core updates). To debug all intermediate steps it might be interesting to attempt breaking M&M iterations to DSC modules, thus keeping track of all intermediate outputs, comparing versions of implementations etc.

There are roughly 4 steps in the DSC:

Generate some data
Set some initializing parameter
Run M&M updates -- exploit existing mashr functions in R
Diagnose the run, including but not limited to computing ELBO -- implemented in Python for my own comfort

Steps 3 and 4 will iterate. In DSC's terms it is something like:

run: set_data * set_params * (fit * diagnose) ^ 5

if we want to do 5 iterations. I implement the benchmark under src/model_dsc.