Closed ziyyadali closed 1 year ago
Hey Ziyyad! Thanks for opening this issue. Good catch! I'll make this change. It does not affect any existing code because we always assume sigma
is constant. I'm just going to remove that first term since it's constant to make the code simpler.
The math for the compute_lnprior function in the Gaussian prior should have a sigma squared.
Current:
lnprob = -0.5*np.log(2.*np.pi*self.sigma) - 0.5*((param - self.mu) / self.sigma)**2
Fixed:lnprob = -0.5*np.log(2.*np.pi*self.sigma**2) - 0.5*((param - self.mu) / self.sigma)**2