Float bug, how can I quickly fix it please ? :(

[Phlos]

Code: Select all

Test.f = 1.50
For i = 1 To 5
  Test + 0.30
Next i

Div.f = Round(Test / 1.00, 0) * 0.05

Debug Div ; Should be 0.15 instead of 0.10 ((3.0 / 1.0) * 0.05 = 0.15) ...)

How can I get accurate value in div please ?

Running PB 4.10 b4

[Derek]

Change to doubles instead of floats.

Code: Select all

Test.d = 1.50 
For i = 1 To 5 
  Test + 0.30 
Next i 

Div.d = Round(Test / 1.00, 0) *0.05

Debug Div

You will still get some small errors in accuracy but it will be a lot closer to what you are expecting.

[Psychophanta]

See comments here in the snipet:

Code: Select all

Test.f = 1.50
For i = 1 To 5
  Test + 0.30
Next i
;test.f is now 2.99999....
Div.f = Round(Test, 0) * 0.05 ; Round(2.999999999999999999999999999999,0) is 2.0 , And then 2.0 * 0.05 is 0.1
Debug Div; Correct result is 0.1

[Derek]

But, of course, the 2.999999999 part is wrong as it should total 3.0 which it will if you use doubles.

[Psychophanta]

But, of course, the 2.999999999 part is wrong as it should total 3.0 which it will if you use doubles.

Agree, agree

[Phlos]

I am not a newbie, I knew why it was bugged Psychophanta

Derek, I cannot use double for 2 reasons :
- my project is big (5000+ lines) and I use float since the beginning (so many many stuffs to change, and maybe new bug )
- double doesn't really fix the issue, it hides it better

So here is my tricky solution :

Code: Select all

Test.f = 1.50
For i = 1 To 5
  Test + 0.30
Next i

Div.f = Round(ValF(StrF(Test / 1.00)), 0) * 0.05

Debug Div ; Should be 0.15 and it is 0.15 now :D

[Thade]

Round is wrong named ... it replaces Floor and Ceil of other languages but does not round in the actual sense.

To get a rounded result always use + 0.5

Code: Select all

Result.f = Round(Test.f+0.5,0)

[Psychophanta]

Round is wrong named ... it replaces Floor and Ceil of other languages but does not round in the actual sense.

To get a rounded result always use + 0.5

Code: Select all

Result.f = Round(Test.f+0.5,0)

Exactly!
So this is perhaps the more elegant solution to take:

Code: Select all

Test.f = 1.50
For i = 1 To 5
  Test + 0.30
Next i
Div.f = Round(Test.f+0.5,0) * 0.05
Debug Div ; Should be 0.15 and it is 0.15 now :D

[Derek]

Yep, Thade is right, the round function doesn't work like you might of learned about rounding but instead rounds up or down depending on your choice so added 0.5 and rounding down works.

[Demivec]

Wouldn't it be simpler to use Int() instead of Round()?

Code: Select all

Result.f = Int(Test.f+0.5)

[Psychophanta]

Yes, but it depends on what Phlos want to do for his program. Round() returns a float while Int() returns a long.

[Phlos]

I am sorry guys, but your fix seems to be wrong
If I have the float value 2.7 and I want the integer part "2", doing Round(2.7 + 0.5, 0) will return me "3" instead of "2"

You can't fix the float bad accuracy by adding a constant value

[Dare]

I am sorry guys, but your fix seems to be wrong
If I have the float value 2.7 and I want the integer part "2", doing Round(2.7 + 0.5, 0) will return me "3" instead of "2"

You can't fix the float bad accuracy by adding a constant value

Hi Phlos.

As 2.7 plus 0.5 is 3.2, Round(2.7 + 0.5,0) = Round(3.2,0) = 3 and so is giving you an accurate result, no?

Edit:

By adding 0.5 and rounding down (flooring) you ensure you get anything with a fractional part of .5 or higher to the next whole number and then truncate the digits. Whereas anything with a decimal fraction part under 0.5 the next whole number is not reached and thus is truncated down to the current whole number. This, of course, within the vague-eries of how floating point numbers are stored (an industry standard).

[Derek]

If I have the float value 2.7 and I want the integer part "2", doing Round(2.7 + 0.5, 0) will return me "3" instead of "2"

If you just want the integer part then you should use int() not round() as round() implied that you wanted the nearest whole number which is what the +0.5 was obtaining.

[Psychophanta]

I am sorry guys, but your fix seems to be wrong
If I have the float value 2.7 and I want the integer part "2", doing Round(2.7 + 0.5, 0) will return me "3" instead of "2"

You can't fix the float bad accuracy by adding a constant value

At the moment may be this workaround can help you, even it is not very elegant, it helps if you need floats and no doubles:

Code: Select all

Test.f = 1.50 
For i = 1 To 5 
  Test + 0.30 
Next i 
Div.f = Round(Test.f+0.01,0) * 0.05 
Debug Div ; Should be 0.15 and it is 0.15 now :D