I know posterior probability as,

$P(\theta|x)= [(P(x|\theta)*(P(\theta))/(P(x))]$,

as given in http://en.wikipedia.org/wiki/Posterior_probability

I am slightly confused with the term $P(y|x;\theta)$,

should I interpret it as, posterior of y given x on the parameter of $\theta$, and how may I interpret

$P(y|x;\theta)=[(P(x|y)*(P(y))/(P(x))]$ under the parameter theta,

or anything else?

If any one of the esteemed members may kindly suggest?

Thanks in Advance,

Regards, Subhabrata Banerjee.

  • $\begingroup$ The term $P(y\mid x;\theta)$ occurs nowhere in the linked to article. Where did you encounter it? $\endgroup$ – Graham Kemp Jun 10 '14 at 22:36

I think you will need to provide more context for where you found the second term $P(y \mid x ; \theta)$, as far as I know, it has not standard definition and its meaning is usually inferred from the context.

My guess is that $\theta$ governs the distribution of $y$. As an example, maybe $y$ is Bernoulli with parameter $\theta$, so $P(y)$ (which might be more properly written $P(y;\theta)$) is defined by $P(y=1):=\theta$ and $P(y=0):=1-\theta$. Again, this is just my guess; you might get better answers if you provide more context. Note that $\theta$ does not appear in your second equation explicitly.

Also, note that $\theta$, $x$, and $y$ play different roles in your two equations. You just need to permute them around, but I wanted to mention this in case you did not notice.

  • $\begingroup$ Dear Room, The equation of my question was not in the link I provided. It was actually Expectation calculation by Stanford Note CS229, given in [cs229.stanford.edu/notes/cs229-notes8.pdf]I was of late thinking if I could calculate it as, P(x,y|z)=P(x|z)*P(y|z) given in courses.engr.illinois.edu/cs440/sp2011/slides/lecture15.pdf or as P(C|F1....Fn) as given in en.wikipedia.org/wiki/Naive_Bayes_classifier. Please see if you can. Thanks for LaTex editing by team. Regards, Subhabrata Banerjee. $\endgroup$ – SUBHABRATA Jun 11 '14 at 6:40
  • $\begingroup$ Dear Room, I found another pointer an old post stats.stackexchange.com/questions/67318/… Regards, Subhabrata Banerjee. $\endgroup$ – SUBHABRATA Jun 11 '14 at 7:42
  • $\begingroup$ @SUBHABRATA From Andrew Ng's notes, he first has a notion of $p(x,z;\theta)$ and $p(x;\theta)$ where $\theta$ contains all parameters governing the $x$ and $z$ (i.e., mean/variance of Gaussian, probability of success of Bernoulli, etc.). You can see in the calculation in the center of page 4 what he means by $p(z \mid x;\theta)$. In essence, you can think of the semicolon as a "conditional" bar (i.e., $p(z\mid x;\theta) \simeq p(z\mid x,\theta)$ and $p(x;\theta) \simeq p(x\mid \theta)$); the use of the semicolon just clarifies that $\theta$ is a parameter and not a random variable. $\endgroup$ – angryavian Jun 11 '14 at 15:08
  • $\begingroup$ Dear Sir, Thank you for the clarification. Regards, Subhabrata Banerjee. $\endgroup$ – SUBHABRATA Jun 11 '14 at 17:13

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.