Subject: | |
From: | |
Reply To: | |
Date: | Mon, 1 Oct 2001 19:18:54 -0600 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Dennis Handly wrote:
>Jeff Woods ([log in to unmask]) wrote:
>>But if you are working with floating point values, one should perform a
>>summation of many values (in this case the total of the value of the
>>digits) beginning with the smallest values and adding in the largest
>>value last in order to minimize the accumulation of rounding error in the
>>final result.
>
>You are correct sir! I asked the IPF math lib guru and he agreed.
Of course. (FWIW, I avoid false modesty. ;) One of my favorite computer
courses (many years ago) in college was "Numerical Analysis" wherein I
learned a few things regarding how to turn floating point to my advantage
rather than to my downfall.
It's amazing how many otherwise intelligent and well-educated people insist
on treating floating point as though it has infinite precision when they've
been told in very clear terms it doesn't. They seem to expect 1/3 (one
third) to be an approximation (presumably because it's a repeating
decimal), but never consider that 1/10 (one tenth) is just as much an
approximation in floating point. If you ignore rounding error, it *will*
bite you when you least expect it.
-- Jeff Woods
* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *
|
|
|