|
PureBytes Links
Trading Reference Links
|
I believe that JavaScript's native format for numerical calculation is
double precision (64 bit floating point numbers). You may want to
code part or all of your .afs script in JScript.
Having single precision floating point numbers in AFS does cause some
problems. I wrote some .afs code that computed the least squares fit
to a quadratic equation. To do the computation, you must solve a 3x3
matrix - .afs doesn't have double dimensioned subscripts that I can
find. I chose the Gaussian Elimination method to solve for the
coefficiants of the quadratic. Since there was not double subscripts,
I created closed form algebraic equations that represented the
successive steps in the Gaussian solution - eliminating the need for
dual dimensioned subscripts.
Once this was done, I found the calculation would unpredictably "blow
up" - it produced wildly wrong answers. That was traced to the fact
that Gauss requires taking the difference between two numbers, both
having been raised to the 4th power. So if you had a number like 1000
- raising that to the 4th power produces something to the 12th power,
then the difference could be in the 10th decimal point - and BOOM it
blows up.
This was solved by creating new arrays of price + time that had fake
small numbers, but upon completion could be expanded back up to the
larger values.
WHEW! - it was many, many lines of .AFS - but all works perfectly now.
Sometimes you have to beat on a particular language to bend it to your
needs.
Cheers,
Reef-Break
--- In amibroker@xxxxxxxxxxxxxxx, bilbod <bilbod@xxxx> wrote:
> >> IIF(L<38.78,X,Y)
>
> the code will yield X when in fact it should yield Y because the low is
> not less than 38.78 (it is exactly 38.78). This occurs simply
because of
> the imprecise representation of the value in AB.
>
> Does anyone know how to get around this, and if its normal for the
value
> to be imprecise in AB? <<
>
> Comparing floating point numbers can be a problem if you are not
> careful. Fortunately financial price data is quantized so this problem
> is easy to work around. I your case, lets say the next legal price
below
> 38.78 is 38.77. You can use any number between 38.78 and 38.77 for your
> test. L<38.779 will work because it is below the computers
> representation of 38.78 but above 38.77. Also L<38.778 will work so
will
> L<38.777, etc.
>
> Bill
------------------------ Yahoo! Groups Sponsor --------------------~-->
<font face=arial size=-1><a href="http://us.ard.yahoo.com/SIG=12h2ak6od/M=362343.6886682.7839641.1493532/D=groups/S=1705632198:TM/Y=YAHOO/EXP=1124995493/A=2894354/R=0/SIG=11qvf79s7/*http://http://www.globalgiving.com/cb/cidi/c_darfur.html";>Help Sudanese refugees rebuild their lives through GlobalGiving</a>.</font>
--------------------------------------------------------------------~->
Please note that this group is for discussion between users only.
To get support from AmiBroker please send an e-mail directly to
SUPPORT {at} amibroker.com
For other support material please check also:
http://www.amibroker.com/support.html
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/amibroker/
<*> To unsubscribe from this group, send an email to:
amibroker-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|