We know that the predictive distribution in a bayesian scheme is:

p(x'|x,param)=integral(p(x'|theta)p(theta|x,param))d(thetha)

Where x' is a new observation

This is defined as averaging over the posterior p(theta|x,param).

I would like to know if anyone has a link to this proof, since I have yet to find it.

Thanks

asked Mar 29 '11 at 13:08

Leon%20Palafox's gravatar image

Leon Palafox
31265471107

closed Mar 29 '11 at 13:56

1

I think this is a definition, a consequence of the concept of marginal probabilities, no?

(Mar 29 '11 at 13:39) Alexandre Passos ♦

Really? Im kinda missing it. Is it something like:

p(a|b)=integral(p(a|c)p(c|b))d(c)

??

(Mar 29 '11 at 13:46) Leon Palafox
1

Yes, with c being theta. You are just marginalizing theta http://en.wikipedia.org/wiki/Marginal_probability

(Mar 29 '11 at 13:48) Alexandre Passos ♦

Cool, now I see it, was a bit overwhelmed with so many equations.

(Mar 29 '11 at 13:51) Leon Palafox
1

Probability theory is indeed more confusing than I'd like.

(Mar 29 '11 at 13:53) Alexandre Passos ♦

The question has been closed for the following reason "Figured it out" by Leon Palafox Mar 29 '11 at 13:56

powered by OSQA

User submitted content is under Creative Commons: Attribution - Share Alike; Other things copyright (C) 2010, MetaOptimize LLC.