| Floating underflow in a plot [message #90771] |
Tue, 14 April 2015 00:50  |
Miguel
Messages: 18
Registered: April 2015
|
Junior Member |
|
|
Hi,
I'm trying to plot an implicit function f(x,y)=a by computing the function for many values of x and y and plotting the result.
But when the number of point (x,y) become too large, there is a problem of floating underflow.
Here is my code :
-------------------------------------------------------
set_plot, 'ps'
device, filename="Essai",/color, bits_per_pixel=8
loadct,13, /silent
!EXCEPT=2
close,/all
plot,[1,1],xrange=[-30,30],yrange=[-30,100],psym=3
openw,1,"bla.cat"
l=(dindgen(45)+0)/1*!PI/180
d=dindgen(1000)/10.+0
l_n=0
inc_color=0
Ro=8.5
To=220
resultat=dblarr(3,n_elements(l)*n_elements(d))
foreach i,l do begin
foreach j,d do begin
R=sqrt(j^2+Ro^2-2*Ro*j*cos(i))
if (R NE 0.) then begin
T=To*(1.00767*(R/Ro)^(0.0394)+0.00712)
V=Ro*sin(i)*((T/R)-(To/Ro))
resultat(0,l_n)=i
resultat(1,l_n)=j
resultat(2,l_n)=V
printf,1,-sin(!PI-i)*j,-cos(!PI-i)*j;,i*180/!PI,j,V
l_n=l_n+1
endif
endforeach
endforeach
k=min(resultat(2,*))
while ( k LE max(resultat(2,*)) ) do begin
ind=where(resultat(2,*) GE k AND resultat(2,*) LE k+0.5)
if (ind(0) NE -1.) then begin
trajectoire=resultat(*,ind)
oplot,-sin(!PI-trajectoire(0,*))*trajectoire(1,*),-cos(!PI-t rajectoire(0,*))*trajectoire(1,*),psym=3,color=fix(inc_color ),NSUM=1
endif
k=k+0.5
inc_color=inc_color+1
if (inc_color EQ 256) then inc_color=0
endwhile
close,1
--------------------------------------------------------
Th bla.cat file contains all the points and when I plot them, it seems fine (except for the last value of l)
What is the problem here =
Thanks
Miguek
|
|
|
|
|
|
|
|
|
|
|
|
| Re: Floating underflow in a plot [message #90777 is a reply to message #90776] |
Tue, 14 April 2015 11:28  |
Heinz Stege
Messages: 189
Registered: January 2003
|
Senior Member |
|
|
On Tue, 14 Apr 2015 08:55:05 -0700 (PDT),
miguelfigueirasebastiao@gmail.com wrote:
> El martes, 14 de abril de 2015, 13:36:44 (UTC+2), Heinz Stege escribió:
>> Hi Miguel,
>>
>> there is a system variable named !EXCEPT. You can change its value to
>> !EXCEPT=2. This makes IDL running slower. But the floating underflow
>> message should be accompanied by another message, which tells you the
>> line of the code, where the floating underflow happens.
>>
>> HTH, Heinz
>
> The !EXCEPT=2 is actually in the code (line 4) and the problem arises at the line where oplot is used.
>
Oh, yes, of cause. I was too lazy to look into the code in detail.
Sorry for this.
Seems to be a very strange error. I can't explain it. However, are you
really sure, that the floating underflow error leads to missing
points, as you say in the answer to Craig's post?
You can eliminate the floating-underflow-error by skipping the points
near the position x=0.0 and y=0.0. This can be done be defining
eps=(machar(/double)).eps
somewhere in the head of your code, and replacing the oplot command by
following lines:
x=-sin(!PI-trajectoire(0,*))*trajectoire(1,*)
y=-cos(!PI-trajectoire(0,*))*trajectoire(1,*)
ii=where(abs(x) ge eps and abs(y) ge eps,count)
if count ge 1 then $
oplot,x[ii],y[ii],psym=3,color=fix(inc_color),NSUM=1
This should not make a visible change to your plot. However for me the
plot still looks some kind of "incomplete".
Cheers, Heinz
|
|
|
|
| Re: Floating underflow in a plot [message #90778 is a reply to message #90776] |
Tue, 14 April 2015 13:04 |
Lajos Foldy
Messages: 176
Registered: December 2011
|
Senior Member |
|
|
On Tuesday, April 14, 2015 at 5:55:06 PM UTC+2, miguelfigue...@gmail.com wrote:
> El martes, 14 de abril de 2015, 13:36:44 (UTC+2), Heinz Stege escribió:
>> Hi Miguel,
>>
>> there is a system variable named !EXCEPT. You can change its value to
>> !EXCEPT=2. This makes IDL running slower. But the floating underflow
>> message should be accompanied by another message, which tells you the
>> line of the code, where the floating underflow happens.
>>
>> HTH, Heinz
>
> The !EXCEPT=2 is actually in the code (line 4) and the problem arises at the line where oplot is used.
>
> Miguel
Set !EXCEPT to 0 and add "if check_math() ne 0 then stop" after the oplot line. Now IDL will stop on the underflow and you can examine the input to oplot.
regards,
Lajos
|
|
|
|
|
|
| Re: Floating underflow in a plot [message #90785 is a reply to message #90779] |
Wed, 15 April 2015 12:04 |
Jeremy Bailin
Messages: 618
Registered: April 2008
|
Senior Member |
|
|
On Wednesday, April 15, 2015 at 12:58:16 PM UTC-5, miguelfigue...@gmail.com wrote:
>> Seems to be a very strange error. I can't explain it. However, are you
>> really sure, that the floating underflow error leads to missing
>> points, as you say in the answer to Craig's post?
>
> The underflow problem was not the cause of the error
>
> I found the problem : the subscript l_n for the array "resultat" can be very high so I needed this variable to have a double precision.
>
> l_n=double(l_n) fixed the problem.
>
> Thank you for your help ;)
>
> Miguel
Ah, interesting. You don't really want to be using a double for something that's fundamentally an integer. But I see that you're actually using a short integer -- any integer that you don't know a priori will never go about 32000 should always be defined as a long:
l_n=0L
And anything that is really a floating-point number should not be declared as an integer:
To=220.0
-Jeremy.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
