Talk:Pointwise mutual information

If the measure is symmetric, then why is the higher bound said to be -log(p(x))? That does not make any sense. ("The measure is symmetric (SI(x,y) = SI(y,x).) It is zero if X and Y are independent, and equal to -log(p(x)) if X and Y are perfectly associated. Finally, SI(x,y) will increase if p(x|y) is fixed, but p(x) decreases.") It implies that SI(x,y) would be bounded from 0 to -log(p(x)) but SI(y,x) from 0 to -log(p(y)) but both values would be equal. Sense, this does not make. pfl (talk) 11:50, 18 July 2008 (UTC)[reply]

If x and y are "perfectly associated", then they always have the same value. So -log(p(x)) = -log(p(y)) and the bounds are the same. —Preceding unsigned comment added by 140.32.16.100 (talk) 00:22, 5 November 2008 (UTC)[reply]

There is no range for PMI other than -infinity to infinity. An example has been added which shows this. Bazugb07 (talk) 20:54, 6 February 2009 (UTC)[reply]

Why is the equation named "SI()" when the article is called "Pointwise Mutual Information"? It should at least be explained if it refers to "Specific MI"...Skywise 78 (talk) 14:58, 17 March 2009 (UTC)[reply]

There seems to be a typo. It should be

   \operatorname{pmi}(x;yz) = \operatorname{pmi}(x;y) + \operatorname{pmi}(x; z|y) instead of 
   \operatorname{pmi}(x;yz) = \operatorname{pmi}(x;y) + \operatorname{pmi}(x; y|z)  ???  — Preceding unsigned comment added by 128.174.241.42 (talk) 20:35, 25 May 2013 (UTC)[reply]

Event vs outcome[edit]

Pretty sure the intro should use the word "outcomes" rather than "events"? The section under the table of contents seems to use the correct language. 203.63.183.112 (talk) 01:14, 24 January 2018 (UTC)[reply]