Thanks Paul and Ed.
It is a shame because i manage to code AR Burg method all in AFL native.
Two points are not so good in my coding :
1- Loop for need to flip an array. There is no such function in native AFL,
but it is very usefull sometimes for many algorythm.
Exemple : [1 5 4 8] -> [8 4 5 1]
Is there a way do do it without loop ? (if not possible thomas you cannot
add a function flip_array ? very usefull for all lattice filter or backwards
recursion)
Code :
for (j = 1; j < i; j++) a_old_flip[j] = a_old[i-j];
2- Precision digit make many sensitive algorythm not done by AFL. Bu i think
it is done to have full speed from AB, so great.
But I have not the same result for SUM and FOR(i=0,...)
Code :
AFL SUM function :
Num = 0; Den = 0;
Num = Sum(W*f*b_shifted, LongBar - i);
Den = Sum(W*(f^2 + b_shifted^2), LongBar - i);
k[i] = 2*Num[LongBar - 1]/Den[LongBar - 1];
FOR :
Num = 0; Den = 0;
for (j = i; j < LongBar; j++) {
Num = Num + 100000*W[j]*( f[j]*b[j-1] );
Den = Den + 100000*W[j]*( f[j]^2 + b[j-1]^2 );
}
k[i] = 2*Num/Den;
I have not the same result for the number k[i] !?
For small number on small period formula give back same result... but for
long periods... i end up with slightly different number (and for AR coef, it
means big diff... arhhh).
which is better SUM or FOR (I had to test by coding it in JSscript or
VBscript and compare maybe) ?
Thanks again.
Cheers,
Mich.
----- Original Message -----
From: Ed Hoopes
To: [email protected]
Sent: Thursday, November 16, 2006 5:17 PM
Subject: [amibroker] Re: Digit precision on AB ?
Also, the default precision in javascript is double.
--- In [email protected], "Paul Ho" .> wrote:
>
>you can only doing single precision floating points within AFL all
>functions that I called returned single precision, 1. I thought it was 8
>significant digits
>http://en.wikipedia.org/wiki/IEEE_754;
>2. yes or dll
>3. for and sum should give the same answer.
>cheers
>Paul.
>
>_____
>
>From: [email protected] [mailto:[EMAIL PROTECTED]
On Behalf
>Of Tom Tom
>Sent: Thursday, 16 November 2006 7:01 PM
>To: [email protected]
>Subject: [amibroker] Digit precision on AB ?
>
>
>
>Hello,
>
>Does someone can help me about digit precision on AB please.
>
>1- I think it is 7 signifiant digit +exposant +signe. Am i right ?
>
>2- For more precision we need to code in VBscript and enable double
float ?
>
>3- But if i prefer stay on Amibroker AFL. For a sum loop for exemple, i
>didn't have same result with looping with FOR(i=...) and with
SUM(array,
>periods). SUM is better ? Native AFL included function work inside
with only
>
>7 digit precision too, or they work in more precision and send back
function
>
>result in 7 digit format. If it is the case, SUM should give better
calcul
>accuracy than FOR(i=...) ?
>
>Thanks,
>Mich.
>
>__________________________________________________________
>Windows Live Messenger sur i-modeT : dialoguez avec vos amis depuis
votre
>mobile comme sur PC ! http://mobile.
><http://mobile.live.fr/messenger/bouygues/> live.fr/messenger/bouygues/
>
_________________________________________________________________
Retrouvez tout en un clin d'oeil avec Live Search !
http://www.live.com/?mkt=fr-fr
