Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: general conversions of floats to intervals



Vladik et al,

At present, I'd prefer to stick with the string conversions the way 754-2008
defines them, although you have a valid point of view upon which
we can vote.  Yes, in some engineering contexts the number of
digits actually written down conveys accuracy information, but using
this convention in our standard would complicate it a bit.  However,
I'm beginning to think of arguments in favor of including it, too.

Baker

Kreinovich, Vladik wrote:
Dear Baker, dear Friends,

Yes, I meant conversion of strings to intervals, because a floating
point 1.5 and 1.50 is the same, so this has to be done with strings.
If the user wants the exact value 1.5, then it can be represented as a
floating point number 1.5; if the user wants 1.50 meaning [1.45,1.55],
then this can be represented as a string '1.50'.
I did not intend to propose a completely new idea, just tried to propose
an additional argument in favor of things that have already been
discussed: that we can keep exact transformation for numbers and expand
an intervals when transforming strings into intervals, so that
intval(1.5) is [1.5,1.5] while intval('1.50') is [1.45,1.55] (to be more
precise, its binary-rational enclosure, with endpoints appropriately
rounded). It may be that I am misreading the opinion of others.
I agree with you that if this is a completely new idea, let us put it on
hold, because we already have enough ideas for this standard.
However, if I did capture the intent of some of the proposers correctly,
folks, please let me know, we can then work on a proposal as Baker
suggested.

-----Original Message-----
From: stds-1788@xxxxxxxx [mailto:stds-1788@xxxxxxxx] On Behalf Of R.
Baker Kearfott
Sent: Tuesday, November 11, 2008 9:23 AM
To: owner-stds-1788@xxxxxxxxxxxxxxxxx
Cc: stds-1788
Subject: Re: general conversions of floats to intervals

Vladik,

Yes.  In fact, Siegfried Rump captures this meaning in INTLAB's
"significance notation."  ( Try "intvalinit('Display_')" and see how
subsequent dialog proceeds.)

However, how is that relevant to converting a float to an interval?
We are given a float, and are converting it to an interval, without
context.  If we're thinking of it with uncertainty, already there
is additional information (e.g. given as two end points or as a midpoint
and radius).

Are you speaking of conversion of strings to intervals?  I suppose
that's
something we can consider, but I would vote against it in a formal
vote, because, what if the user really wanted the exact number input?
If you want it, please continue to advocate it, and bring it up as
a motion or position paper when we have our formal procedures in place.

Baker




--

---------------------------------------------------------------
R. Baker Kearfott,    rbk@xxxxxxxxxxxxx   (337) 482-5346 (fax)
(337) 482-5270 (work)                     (337) 993-1827 (home)
URL: http://interval.louisiana.edu/kearfott.html
Department of Mathematics, University of Louisiana at Lafayette
(Room 217 Maxim D. Doucet Hall, 1403 Johnston Street)
Box 4-1010, Lafayette, LA 70504-1010, USA
---------------------------------------------------------------