3 ms·
Is this equivalent to encoding the ranks as decimals in a floating-point probabilistic PL? Like if a program uses ranks 0-9 in RankPL, could those could be enco
by zaptheimpaler 10y ago
Is this equivalent to encoding the ranks as decimals in a floating-point probabilistic PL? Like if a program uses ranks 0-9 in RankPL, could those could be encoded as probabilities 1/55, 2/55, 3/55, ... ,10/55 (assuming rank 9 is 10 times less likely than rank 0) in a regular probabilistic PL?
- nerdponx 10y agoIt looks like the values are un-normalized density values, i.e the numerator in Bayes' theorem. So, what you said. But determining the correct denominator for such a collection of values, in general, isn't easy when the collection is infinite (like a function). It amounts to a very very gnarly integration problem. This is why MCMC solvers are used in Bayesian statistics, and why abstracting away the "nasty integral" part from the "probability density" part is useful to statisticians and other researchers.
- tr352 10y agoThere is indeed a probabilistic interpretation of ranks. It's discussed here: http://ac.els-cdn.com/0004370295000909/1-s2.0-0004370295000909-main.pdf?_tid=b5361720-1647-11e7-bb2d-00000aacb361&acdnat=1490988469_b0f8c1932e015b072c0569c252376961 http://ac.els-cdn.com/0004370295000909/1-s2.0-00043702950009...