Incorporating known variable distributions

I have a situation where I have a variable that I am predicting that I know the distribution of a set of parameters I’m trying to learn and I want to incorporate that knowledge into pymc3. The specific case is trying to determine the length of the principal axes of an ellipse from the projection of that ellipse onto a line at uniformly randomly distributed angles. (I posted about this previously here Improvements to geometry based model form).

I can estimate what the length of the principal axes are and then compute the different angles using pymc3 but I’m struggling to also have pymc3 make an estimate of the length of the principal axes because the number of equations is less than my number of unknowns so I get lots of divergences. I would like to encode my knowledge that the angles should be distributed uniformly which will help define my solution. Is there a good way to do this?