Using the compile_logp function to evaluate the log-posterior of a point estimate

[ricardoV94]

@Antonis_Mav

I have tried using model.compile_logp(point_estimate_dict)

Should be model.compile_logp()(poont_estimate_dict)

The first call simply compiles and returns a function (you should store this if you intend to do multiple evaluations). The second actually evaluates it.

@Antonis_Mav

Which of the model variables should be used in the point estimate dictionary?

You should include all the unobserved variables.

@Antonis_Mav

Why does the compile_logp function require to add _log_ after my parameter names?

Because those variables are transformed on the log scale, so your input should actually be the log of the variable. The name is to avoid surprises. You can disable a variable transformation by passing transform=None when defining it in the Model.

Finally, you can call model.initial_point() to have an example dict of all the variable names and values/shapes that are required for the model logp. model.compile_logp()(model initial_point()), should always work.