From: Paul A. Bristow (pbristow_at_[hidden])
Date: 2001-04-24 12:31:56
I am sure there is a BIG cost in using intervals rather
than single values, and for SOME applications it isn't worth it.
But for calculations involving real measured values with known
uncertainties, estimating the uncertainty of the calculated result
is exceedingly valuable because it allows you to avoid drawing
the wrong conclusions from the measurement.
Confidence intervals ARE important. For example the predictions
of vCJD deaths from the BSE epidemic were something like 10 000,
but the 95% confidence intervals were 100 and 10 000 000.
When more data was available, the estimates were refined to
1 000 to 100 000, the sillier newspapers screamed that
10 times as many might die or 100 000 might die -
despite this being really good news.
So if the calculation is not too intensive (or you can afford the
interval calculation IS useful. (But I'm not yet convinced that it
isn't more complicated than the methods in use so far.
Statisticians always keep degrees of freedom in mind,
and distributions are not just normal...
C++ (and fast processor chips) make these things conceivable.
> -----Original Message-----
> From: David Abrahams [mailto:abrahams_at_[hidden]]
> Sent: Tuesday, April 24, 2001 11:53 AM
> To: Boost_at_Yahoogroups. Com
> Subject: Re: [boost] Re: Boost.MathConstant (& Boost.Interval)
> ----- Original Message -----
> From: <k.hagan_at_[hidden]>
> To: <boost_at_[hidden]>
> Sent: Tuesday, April 24, 2001 6:13 AM
> Subject: [boost] Re: Boost.MathConstant (& Boost.Interval)
> > Peter Schmitteckert asks if there is a boost interval library.
> > Yes, Jens Maurer submitted one over a year ago, but it is sitting
> > in the vault because people like me haven't got around to doing
> > anything with it. I would expect a good interval library to replace
> > the use of plain float & double, in the same way that std::string
> > has (?) replaced char. (We only need to convince the performance
> > freaks that there is little or no cost involved.)
> Is that really possible? Surely one has to do 2-3 times as many
> floating-point ops to handle intervals as raw numbers. I know that in my
> application domain (simulation), nobody would be willing to pay
> that price.
> Typically these simulations use iterative methods that only
> strive for a few
> digits of accuracy before producing a result anyway.
> To unsubscribe, send email to: <mailto:boost-unsubscribe_at_[hidden]>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk