From: smartnose@gmail.com   
      
   On Oct 2, 10:35 pm, Ted Dunning  wrote:   
   > On Oct 1, 4:58 pm, smart  wrote:   
   >   
   > > Page 8. He talked about Beta(0,0) as prior.   
   >   
   > > But does Beta(0,0) exist? The integration in the definition of   
   > > Beta(0,0) simply don't converge? Or do I make any misunderstanding?   
   >   
   > > I'm confused.   
   >   
   > > Wei   
   >   
   > This is an improper, but it is still useful in some applications,   
   > especially as a prior.   
   >   
   > You are correct that the integral diverges.  Typically, though, you   
   > don't want to integrate the distribution itself, but some other   
   > distribution based on this distribution (as a posterior is based on a   
   > prior, for instance).  Other useful improper priors include the   
   > hyperbolic distribution (the limit of Gamma for constant shape * scale   
   > as shape => 0) and the infinite uniform (the limit of the normal   
   > distribution with 0 mean as \sigma => infinity).  These limits don't   
   > properly exist, but the corresponding limits on other expressions   
   > using these "distributions" may exist.   
   >   
   > There is considerable controversy regarding the explicit use of these   
   > distributions, but they are conceptually useful in certain cases.   
   >   
      
   Hum... Thank you.   
   I read other literatures and figure out so called 'improper'   
   distribution.   
      
   I also think it's ok to use it this way, since  it seems that  we need   
   only an ordered relationship between all the samples, so normalization   
   doesn't matter much.   
      
   Thanks   
      
   Wei   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)