PureBytes Links
Trading Reference Links
|
selected messages from a former thread on the very same subject -- TJ
oh yeah, what would be the accumulated error on 1.6 million bar with
simple average functions in ts? now that boggles the mind even more
than okie man's ignorance of such things!!!
Resent-Date: Fri, 12 Sep 1997 06:40:00 -0700 (PDT)
X-Sender: bfulks@xxxxxxx
Date: Fri, 12 Sep 1997 09:29:08 -0400
To: Scientific Approaches <sci@xxxxxxxxxx>
From: Bob Fulks <bfulks@xxxxxxxxxxx>
Subject: Re: EL numerical accuracy
Cc: Omega Mailing List <omega-list@xxxxxxxxxxxxxxx>
Resent-From: omega-list@xxxxxxxxxx
X-Mailing-List: <omega-list@xxxxxxxxxx> archive/latest/9561
X-Loop: omega-list@xxxxxxxxxx
Resent-Sender: omega-list-request@xxxxxxxxxx
>Massimo Ciarafoni wrote:
>
>> does anyone know which numerical accuracy EL does
>> have? I mean how many decimal figures EL uses in
>> a result, (i.e. 10/7 equals 1.42, 1.428, 1.4285,
>> 1.42857, 1.428571 or ....). Does the price scale
>> have any effect on the numerical accuracy?
>
Several people seemed to question the need for high accuracy in
numerical
calculations. Obviously, three or four digits of accuracy would
normally be
satisfactory for the final results. but many operations require much
greater accuracy than this for intermediate calculations.
As a very simple example, the "Average" function supplied with
TradeStation
calculates the average by subtracting the old bar and adding the new
bar:
Sum = Sum[1] + Price - Price[Length]
This is done because it is faster than recalculating everything on each
new
bar.
If there were even a small error on each calculation, the total error
accumulated after hundreds or thousands of bars could be very
significant.
As it is, the present accuracy is barely adaquate for many operations.
--
Bob Fulks
bfulks@xxxxxxxxxxx
Resent-Date: Fri, 12 Sep 1997 10:40:40 -0700
Date: Fri, 12 Sep 97 10:44:24 PDT
From: chris@xxxxxxxx (Chris Norrie)
To: omega-list@xxxxxxxxxx
Subject: Re: EL numerical accuracy
Resent-From: omega-list@xxxxxxxxxx
X-Mailing-List: <omega-list@xxxxxxxxxx> archive/latest/9571
X-Loop: omega-list@xxxxxxxxxx
Resent-Sender: omega-list-request@xxxxxxxxxx
Bob make a very good point in the following text. If I understand
corectly,
TS5 will eliminate the 13,000 bar limit, allowing systems to run on a
much
greater range of data. Accumulating numerical error will become an
even
greater problem problem unless floating point operands go to either 64
bits
or 80 bits.
Chris Norrie
--- Mark Brown <markbrown@xxxxxxxxxxxxx> wrote:
I think the point is that Easy Language builds up a cumulative bunch of
errors that anyone of by themselves is not that significant. However if
you do big jobs and complicated systems that do require precision TS
just can not do it. The errors will add up to the point to where you
will be buying where you should be selling. This is the point and you
know it weather you and PO will admit it or not. If what you say is
true then what would your interest in Fortran be? Why what would be
the purpose?
|