• ref2401 (6/6) Dec 15 2012 What does means 'maxRelDiff' parameter?
• bearophile (4/10) Dec 15 2012 Maybe you want to use std.math.feqrel instead.
• ref2401 (1/2) Dec 15 2012 that's not what i was asking about
• =?UTF-8?B?QWxpIMOHZWhyZWxp?= (21/26) Dec 15 2012 If you consider two values to be equal if the difference between them is...
• ref2401 (1/1) Dec 15 2012 now i got it, thanks.
• js.mdnq (10/16) Dec 16 2012 basically floating point types are not accuracy. Is 0.000001 =

"ref2401" <refactor24 gmail.com> writes:

What does means 'maxRelDiff' parameter?
I looked at the source code of this method and I still didn't get 
it.

return fabs((lhs - rhs) / rhs) <= maxRelDiff
|| maxAbsDiff != 0 && fabs(lhs - rhs) <= maxAbsDiff;

In what cases can I use this parameter?

"bearophile" <bearophileHUGS lycos.com> writes:

ref2401:

 What does means 'maxRelDiff' parameter?
 I looked at the source code of this method and I still didn't 
 get it.

 return fabs((lhs - rhs) / rhs) <= maxRelDiff
 || maxAbsDiff != 0 && fabs(lhs - rhs) <= maxAbsDiff;

 In what cases can I use this parameter?

Maybe you want to use std.math.feqrel instead.

Bye,
bearophile

"ref2401" <refactor24 gmail.com> writes:

 Maybe you want to use std.math.feqrel instead.

that's not what i was asking about

=?UTF-8?B?QWxpIMOHZWhyZWxp?= <acehreli yahoo.com> writes:

On 12/15/2012 11:01 AM, ref2401 wrote:
 What does means 'maxRelDiff' parameter?
 I looked at the source code of this method and I still didn't get it.

 return fabs((lhs - rhs) / rhs) <= maxRelDiff
 || maxAbsDiff != 0 && fabs(lhs - rhs) <= maxAbsDiff;

 In what cases can I use this parameter?

If you consider two values to be equal if the difference between them is 
1%, then you provide 0.01 as maxRelDiff. The concept of a percentage 
difference makes sense for some programs.

Sometimes you know the magnitude of the values up front (and you are 
sure of it). In such cases you can provide an absolute difference:


import std.math;

void main()
{
     // By relative difference
     assert(approxEqual(1.01, 1.015, 0.01));
     assert(approxEqual(1010, 1019, 0.01));

     // I am sure that the values are around 10 thousand and a 
difference of 10
     // is sufficient to accept them as equal:
     assert(approxEqual(9995, 10_003, 1.0, 10.0));
}

Since the concept of equality is related to precision, sometimes it 
makes more sense to use feqrel() as it returns the number of bits of the 
mantissa that the two values have equal.

Ali

"ref2401" <refactor24 gmail.com> writes:

now i got it, thanks.

"js.mdnq" <js_adddot+mdng gmail.com> writes:

On Saturday, 15 December 2012 at 19:01:23 UTC, ref2401 wrote:
 What does means 'maxRelDiff' parameter?
 I looked at the source code of this method and I still didn't 
 get it.

 return fabs((lhs - rhs) / rhs) <= maxRelDiff
 || maxAbsDiff != 0 && fabs(lhs - rhs) <= maxAbsDiff;

 In what cases can I use this parameter?


basically floating point types are not accuracy. Is 0.000001 = 
0.00000000000000000000000000000001?

It all depends! floating point calculations can accumulate 
rounding errors which result in comparisons that should be valid 
are not.

Hence, it is not technically correct to compare two floating 
point types. Hence, the calculation above simply compares how 
close the two are and accepts them as == if they are within some 
distance of each other.