1
$\begingroup$

Given a finite set of non-negative numbers $S={s_1,...,s_n}$, we can divide them by a normalization constant $Z$ (i.e. their sum) to get a probability distribution.

Then we typically say (and write) something like: the probability of event $x_i$ is $p(x_i)=\frac{s_i}{Z}$.

In this context, is there any standard name and notation conventions for the "unnormalized probability" $s_i$?

I usually just call these unnormalized probabilities and use some arbitrary notation (e.g. $\tilde{p}(x_i)$), but I am not very fond of this name.

$\endgroup$
0
$\begingroup$

Probability is a measure that assigns the value 1 to the entire space. Without that, you just have a non-negative measure (i.e., its just measure theory w/o any probabilistic interpretation)

$\endgroup$
  • $\begingroup$ Thanks. I know these are just non-negative numbers, but when used in in this probabilistic context they have an additional "meaning" and it is sometimes useful to talk about the quantities before and after the normalization. I am just wondering if there is some standard name/notation for this. $\endgroup$ – Bitwise Jan 24 '14 at 1:09
  • $\begingroup$ @Bitwise the best term I can think of is likelihood as it is a measure of relative support as opposed to absolute support $\endgroup$ – user76844 Jan 24 '14 at 16:22
  • $\begingroup$ I see how this would work in a Bayesian context, however in my case I actually use likelihood later for a different term so it is not suitable. I will wait to see if I get other answers before proceeding. $\endgroup$ – Bitwise Jan 24 '14 at 16:29
  • $\begingroup$ @Bitwise based on your description in the post, you are dividng a number by its sum to get a probability. In that case, mustn't your $s_i$'s be raw counts of some sort? Why wouldn't you just call them counts? $\endgroup$ – user76844 Jan 24 '14 at 17:21
  • $\begingroup$ Actually, these are not the counts. It is more of a parametrized probability distribution, e.g. $p(x_i)=\frac{e^{i}}{Z}$. $\endgroup$ – Bitwise Jan 24 '14 at 19:10

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.