enter image description here

I summarize this based on http://www.cs.ubc.ca/~murphyk/Papers/bayesGauss.pdf

However, there are some questions about this.

  1. When the mean and variance are both unknown, can I set the prior of of variance a Chi-square distribution? If so, is the posterior normal chi-square distribution?
  2. I know that wishart distribution is generalization of chi-square distribution, then are there any generalizations of gamma distribution?

No. Only lines 1 and 6 are correct.

For a univariate normal,

  • the conjugate prior for a variance (with known mean) is an inverse gamma (which is a reparameterization of a scaled-inverse $\chi^2$ distribution and thus the inverse $\chi^2$ is a special case) and
  • the conjugate prior for mean and variance is a normal-inverse-gamma distribution.

For a multivariate normal,

  • the conjugate prior for a covariance matrix (with known mean) is an inverse Wishart (which is a generalization of an inverse-gamma distribution) and
  • the conjugate prior for the mean and variance is a normal-inverse-Wishart distribution.

Note that you do not need to specify the posterior family once you have specified the prior family because by definition of conjugacy, these will be the same.

  • $\begingroup$ thanks very much. Based on your answers, I think lines 2 and 3 are incorrect, lines 4 and 5 are correct, right? For multivariate normal, lines 7 and 8 are incorrect, I will remove them. $\endgroup$ – DuFei Nov 3 '17 at 1:26
  • $\begingroup$ The way you have laid it out, lines 4, 5, and 9 are not correct because you need a joint distribution for the mean and (co-)variance in both the prior and the posterior. You have indicated this in the posterior, but have not indicated this in the prior. $\endgroup$ – jaradniemi Nov 3 '17 at 14:22
  • $\begingroup$ Feel free to mark it as the answer. $\endgroup$ – jaradniemi Nov 8 '17 at 2:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.