I was also trying to implement a toy logistic regression before building something more complex. I found it somewhat hard to follow the existing regression tutorial because of the way the code is structured.
I prefer the structure of the VAE tutorial, in which the model and guide are contained within a reusable module and aren’t relying on any global variables located elsewhere in the notebook.
It took a lot of debugging and messing around with dimensions (I still don’t really grok how Pyro does this but I’m slowly learning), but I have a working example here: