How to: calibration -> inverse prediction

5

You can write an equivalent model that doesn’t use automatic imputation. Maybe that’s more understandable. Here I am using something simpler but you should be able to translate to your own model

with pm.Model() as m:
  x_fixed = np.array([1, 2, 3])
  x_unknown = pm.Normal("x_unknown", shape=2)
  x = pm.math.concatenate([x_fixed[None], x_unknown[None]], axis=0)
  y = pm.Normal("y", x, 0.1, observed=np.array([1, 2, 3, 4, 5]))