Talk:Pseudocount

Why is the rule of succession "a bit of a fudge"?

Because its justification is heuristic, and has no theoretical basis whatsoever. Will that do for you? 81.102.133.198 19:41, 25 September 2007 (UTC)[reply]

The rule of succession says that if you have a uniform prior on [0, 1] for a frequency parameter p, with the independent probability of a success on each trial being p, then the probability of a success after s successes and n total trials is (s+1)/(n+2). The proof is given on the Rule of Succession page. Sure, you don't always have a uniform prior, but I hardly see how this is "no theoretical basis whatsoever." —Preceding unsigned comment added by 72.94.217.197 (talk) 07:44, 26 September 2007 (UTC)[reply]

The statement "Neither approach is completely satisfactory and both are a bit of a fudge" should be removed as it's expressing a point of view. As far as I'm concerned, Laplace's rule is very satisfactory in practice and I'll go on using it, just as I'll go on using uniform priors. I know this is only a subjective belief but that's what probability's all about. ;-)

I'll remove the statement but if anyone feels strongly about it, let them reinstate it.

--84.9.85.135 10:52, 25 October 2007 (UTC)[reply]