I just have a question about why Gibbs sampling can correctly estimate parameters with random initial value? That is to say,we can sample z by: \begin{align} p(z_i=k \,|\, \cdot) &\propto \prod_i^n \prod_k^K \left( \pi_k \frac{1}{\sigma\sqrt{2\pi}} \text{exp}\left\{ \frac{-(x_i-\theta_k)^2}{2\sigma^2} \right\} \right)^{\delta(z_i, k)} \\ &\propto \pi_k \cdot \frac{1}{\sigma\sqrt{2\pi}} \text{exp}\left\{ \frac{-(x_i-\theta_k)^2}{2\sigma^2} \right\} \end{align} Then update parameters After several iteration of these two-step, it gets true distribution with random initialization. Why it converges to the true distribution? I don't understand. Which information contribute to the change of parameters to the true value?


Your Answer

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

Browse other questions tagged or ask your own question.