topic Re: Programing question in Libraries & objects

Programing question

Anonymous — Thu, 09 Dec 2004 21:09:32 GMT

What is the best bit of code to use to avoid "divide by zero" errors?
I run into this issue often when using circular functions
such as in the following piece of code.

ratio=b/a
ang=atn(ratio)

If "a" happens to be zero there is an error.
Usually I resort to placing the following at the
beginning of the script.

if a=0 then a=0.0000001

This "work-around" is logically very dissatisfying to me.
Is there a better way?

Thank you,
Peter Devlin

Re: Programing question

TomWaltz — Thu, 09 Dec 2004 22:16:04 GMT


IF A <> 0 THEN
     ratio=b/a
     ang=atn(ratio)
ELSE
   (something to fix the problem)
ENDIF

Re: Programing question

Anonymous — Thu, 09 Dec 2004 22:45:22 GMT

Tom,
Thank you for your reply.
Its the "(something to fix the problem)"
that I was looking for.
Something more elegant than if a=0 then a=0.00000001.
Peter

Re: Programing question

TomWaltz — Thu, 09 Dec 2004 23:59:00 GMT

Where are A and B coming from?

Re: Programing question

Anonymous — Fri, 10 Dec 2004 00:47:40 GMT

Tom,
In the example I gave "a" and "b" are the obligatory
x and y dimensions of a library part.
But they could be any numeric parameter or variable
that has a possible value of zero either as initialized or
as the result of calculation.
Peter

Re: Programing question

TomWaltz — Fri, 10 Dec 2004 01:44:13 GMT

Peter

I was wondering if there was a certain situation this was coming from, that maybe there would be some kind of logic that would prevent it from happening. Since objects should probably never have an A or B value of 0, that should kickj up an error if its own.

Re: Programing question

Anonymous — Fri, 10 Dec 2004 02:47:53 GMT

Tom,
I appreciate your interest in this issue.
I made a library part that is a detail marker.
(see attached image)
I wanted (0,0) to be at the center of the "bubble"
and the point (a,b) to be at the edge of the call-out symbol
so that the connecting line between bubble and symbol
can be stretched to any angle and length and
still point to the center of the bubble similar to
what Red was trying to do in the thread titled "tangent".
The code must allow the library part's "A" to equal zero
to display the shape labeled #2
I know that there are ways to code this part so that
"A" equaling zero does not cause a problem, but
this is not the point I only use this lib part as an
example of where one value divides another value
and the dividing value may equal zero
and therefor cause an error.
Thanks,
Peter

Re: Programing question

rocorona — Fri, 10 Dec 2004 10:09:55 GMT

I don't have a more efficient solution. Usually, when I know that a divisor can became zero, I use the following form:

epsilon = 0.000001 ! <--at the beginning of the current (or Master) Script

ratio=b/MAX(a, epsilon)
ang=atn(ratio)

This way the error message can be avoided.
But, if the number can also be negative, the formula has to be more complex. ... I think something like this:

ratio=b / (MAX(ABS(a), epsilon) * (sgn(a)+(a=0)))

GASP!
_______________________________

--Roberto Corona--

Re: Programing question

Laszlo Nagy — Fri, 10 Dec 2004 10:48:37 GMT

rocorona wrote:
I don't have a more efficient solution. Usually, when I know that a divisor can became zero, I use the following form:

epsilon = 0.000001 ! <--at the beginning of the current (or Master) Script

ratio=b/MAX(a, epsilon)
ang=atn(ratio)

This way the error message can be avoided.
But, if the number can also be negative, the formula has to be more complex. ... I think something like this:

ratio=b / (MAX(ABS(a), epsilon) * (sgn(a)+(a=0)))

GASP!
_______________________________

--Roberto Corona--

Yes, this is similar to what GS programmers do in their GDL Scripts.
One can check out any AC9 Door or Window script and will find many examples similar to this:

IF ABS(Whatever) > EPS THEN
...............
ENDIF

EPS is for Epsilon.

Re: Programing question

Anonymous — Fri, 10 Dec 2004 11:07:58 GMT

I have always preferred a 'logical check';

oEpsilon = 0.0000001
ratio = oA / ( (oB=0) * Epsilon + oB )

For angles this is extended to;

oAngle = atn ( oA / ( (oB=0) * Epsilon + oB)

Oh how I wish we had 'proper' macro statements ...

oAngle = myATN (oA, oB)
!------
define macro "myATN" (value1, value2)
return atn ( value1 / ( (value2=0) * 0.000001 + value2)

- Stuart

Re: Programing question

Durval — Fri, 10 Dec 2004 16:19:03 GMT

Instead of conditional change of 'a', why not simply change 'ang' directly?

IF a<epsilon THEN ang=90

Stuart, wouldn't this code
ratio = oA / ( (oB=0) * Epsilon + oB )
trigger the annoying 'precision problems' warning?

Re: Programing question

Anonymous — Fri, 10 Dec 2004 18:52:47 GMT

Durval;

I might be wrong, but IMHE, "x=0" doesn't seem to create a precision problem. x=1 (or any other number) can.

Will test a little over the w/e.

? - Stuart

Re: Programing question

Oleg — Fri, 10 Dec 2004 20:33:30 GMT

I would prefer:

IF ABS(a)<EPS THEN ang=SGN(b)*90 ELSE ang=TAN(b/a)

as imho it is more readable and clear

Re: Programing question

Anonymous — Sat, 11 Dec 2004 03:25:49 GMT

Hello,
Looking at all of the replies to my original question I can only conclude
that some conditional, somewhat arbitrary, change to the devisor must be made when the devisor acquires a zero value.
At the time I posted my question I was wondering if there was some mathematical algorithm or programing command that would say in effect
if a=0 then
a=(smallest non-zero quantity allowed by the program/math processor).
I was thinking that in floating point arithmetic "zero" must have some
internal definition limited by precision.
Since I don't have a computer science background the above statements
may be complete gibberish.
This thread has certainly brought up some fascinating issues.
Thank you all for your comments,
Peter Devlin