|
PureBytes Links
Trading Reference Links
|
"The Omega Man" <editorial@xxxxxxxxxxxxx> writes:
>JimO wrote:
>
><<I can't imagine a simpler question than my original one regarding
>precision; is it single or double? I don't see what my question
>has to do with web browsers, or much of anything except EL.>>
>
>Jim, your original question was whether *TS 5.0* supported the option of
>double precision. To quote, your original question was:
>
>"Can you tell us if TS 5.0 is limited to single-precision calculations, or
>does it have the option of using double precision?"
>
>So you did not ask about EL. You asked about TS 5. I was attempting to
>answer your question regarding the data items that TS 5.0 can use.
Sorry, I thought I was implicitly asking about EL, as that's all that
came with my copy of TS. It's my one and only Microsoft-dependent
ap, and will likely remain so. If *I* were designing something like
EL, I'd either have it use double-precision, or provide the option
to declare certain vars as double, for all the reasons I've mentioned
earlier.
I once propositioned Doug Deming to write a DLL that would provide a
simple stack-based double-precision calculator, like HP's, so that we
could do recursive calculations, etc., without roundoff failures.
Doug didn't think there'd be a market for it, as most users find simple
EL intimidating, and the concept of roundoff error is utterly beyond them.
>Then you wrote:
>
><<I just want to know if the new version will fix one of the long-standing
>limitations of the current versions, and allow us to write some types of
>code that currently fail under the single-precision limit. Is that a
>"possibility the new architecture allows?">>
>
>
>In my mind, the answer to this question is YES. However, this would not be
>done through EL (at least, I would not do it that way) since there is no way
>to declare variable types in EL.
Then they should be double-precision to start with.
>We also need to be careful here that we don't get into discussing a
>distinction without a difference. How many significant figures are in the
>original data upon which you are basing your calculations? Remember, we
>cannot get more accuracy out than we put in. We cannot use price data with
>3-5 significant figures in a calculation and get a result with 12
>significant figures!
No, but it's the roundoff error within the calculation itself that
is the problem here. All too often, when you examine the details
of numerical algorithms, you need considerably more precision within
the calculation itself to retain the precision of its input.
There's a minor controversy within the Java community over the use
of extended-precision (more than double) within calculations. It's
a subject of real concern to folks used to working with numbers.
>You need not question my every motive - my effort to answer
>your original question was quite sincere.
Thanks for your thoughts,
Jim
|