Log-posterior calculation involve PDE numerical analysis

[yangxu698]

@Bob_Carpenter

As Ben said, you need to use optimization, not sampling. Sampling finds typical parameter values, not ones that maximize the log density.

Sorry, I just wanna ask, I remember the non-formative prior(uniform) will result in the same result as MLE, is that correct? I’m not sure the difference between optimization and sampling.

@Bob_Carpenter

The only concern is that differentiating the iterative algorithm (which is an approximation to the value) may not give a good approximation to the derivative everywhere.

Thanks, I will try as long as there’s some possibility that Stan may work.

@Bob_Carpenter

P.S. It’s “Stan”, not “STAN”.

Thank you for the kind reminder!