How to troubleshoot sampling error with custom, not differentiable likelihood?

Okay, that’s helpful, thanks. Right now I’m experimenting with just forcing multiple step functions not to try NUTS. It seems to run but has yet to find a sample.

I looked into SunODE early on, but it seems to only support ODEs, and I need to solve a PDE. fenics_pymc3 is the only suitable package I’ve found, and I couldn’t get it to play nice with PyMC 4, which is the only reason I’m using PyMC3.