From: Thorsten Ottosen (nesotto_at_[hidden])
Date: 2004-11-08 14:57:14
"Paul A Bristow" <pbristow_at_[hidden]> wrote in message
| But I believe strongly that "at most a few ulps" is entirely the wrong
| When the existing Standards make ABSOLUTELY NO accuracy requirements,
| I find this a surprising target.
| If Quality is Fitness-for-Purpose,
| then a much lower accuracy is entirely acceptable in the Real World.
I agree with this. Probabilities are often accompanied by a significant second
order uncertainty. For example,
it many real applications it is hard to justify a probability with a precision
| The loss of even 3 decimal digits precision in the incomplete beta
| still make a negligible difference to the probability calculated
| (which has a much greater uncertainty because of the sensitivity
| to physical measurement and degrees of freedom).
not to say when probabilities are expert jugdments based on a sample size of
I'm with you here Paul.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk