Re: relative error for 'zero' input

Sorry, i didn't specify very clearly xi and xo.

Of course xo=measured (=aproximate) value
xi = absolute value (resp. what i'm supposing to be there)

the thing is that I have signals which have xo nonzero while xi is zero. What is the 'standard' procedure to deal with such a case? Express an absolute error like sqrt((xi-xo)^2)?
d.



David W. Cantrell napsal(a):
"hagman" <google@xxxxxxxxxxxxx> wrote:

Dejfson schrieb:


Dear All,

is there any way how to calculate the relative error defined as:

E=100(1-xo/xi)

for xi=0 ?

Obviously, no.


???
If xo is nonzero, as indicated below, then the relative error is infinite.

BTW, I would normally calculate relative error using

(approximate value)/(exact value) - 1.

Of course, you can then multiply that result by 100% if you want a
percentage, or take its absolute value, if that's what you want.

If the exact value is 0 and the approximation isn't,
then |rel. error| = +infinity.

David


Resp. what is the standard way of expressing the error for the
zero-input non-zero-output measurables?

If you expect a result near zero, you might want to investigate the
absolute error instead.
Of course this causes trouble when investigating sequences of
operations.
You may have a "badly conditioned" problem if this case arises.

BTW, I hope you don't understand xo and xi as "output" and "input"?
You should rather use "calculated result" and "exact result".
.

Relevant Pages

  • Re: Truncation of infinite series
    ... the relative error is the absolute divided by the ... Since the sum of the series is strictly increasing, an upper bound is ...
    (sci.math)
  • Re: relative error for zero input
    ... There's more than one standard way to deal with it. ... impression that you wanted the _relative_ error instead, which is infinite. ... percentage, or take its absolute value, if that's what you want. ...
    (sci.math)
  • Re: Floating point precision problem
    ... Or, more generally, using a comparison function that uses a relative error ... instead of an absolute one. ... Prev by Date: ...
    (microsoft.public.vc.language)
  • Re: relative error for zero input
    ... But having infinite relative error doesn't say about the error anything at all. ... I think that any attempt to calculate relative error finishes with infinity for zero 'absolute' value. ... impression that you wanted the _relative_ error instead, ...
    (sci.math)
  • Re: Infinity......
    ... if the sizes of A and B are different ... see william I know what you mean very ... Of course the term "size" in the above system means ABSOLUTE size. ... Then this mean that infinite sets do not have an absolute set size., ...
    (sci.math)