Problem with Tick

[Chihastonick]

Hi guys, I have the follow code that extends Actor :

var float temp;

simulated State MyState
{
	function BeginState(Name PreviousNameState)
	{
		`log(" -------------------------------------- MyState ");
	}
          
	functiom Tick(float DeltaTime)
	{
		Super.Tick(DeltaTime);

		temp += 0.4;

		`log(" ------------------------- TEMP : "@temp);
	}
}

The result log :

[0060.34] ScriptLog:  -------------------------  TEMP :  47.6000
[0060.36] ScriptLog:  -------------------------  TEMP :  48.0000
[0060.37] ScriptLog:  -------------------------  TEMP :  48.4000
[0060.39] ScriptLog:  -------------------------  TEMP :  48.8000
[0060.41] ScriptLog:  -------------------------  TEMP :  49.2000
[0060.42] ScriptLog:  -------------------------  TEMP :  49.6000
[0060.44] ScriptLog:  -------------------------  TEMP :  50.0000
[0060.45] ScriptLog:  -------------------------  TEMP :  50.4000
[0060.47] ScriptLog:  -------------------------  TEMP :  50.8000
[0060.49] ScriptLog:  -------------------------  TEMP :  51.2001
[0060.50] ScriptLog:  -------------------------  TEMP :  51.6001
[0060.52] ScriptLog:  -------------------------  TEMP :  52.0001
[0060.53] ScriptLog:  -------------------------  TEMP :  52.4001
[0060.55] ScriptLog:  -------------------------  TEMP :  52.8001
[0060.57] ScriptLog:  -------------------------  TEMP :  53.2001

Why ???

[Mooser]

Floating point rounding errors!

http://docs.oracle.com/cd/E19957-01/..._goldberg.html

http://stackoverflow.com/questions/2...rounding-error

[Chihastonick]

Floating point rounding errors!

http://docs.oracle.com/cd/E19957-01/..._goldberg.html

http://stackoverflow.com/questions/2...rounding-error

solutions to fix it quickly?

[Angel_Mapper]

There's almost certainly a better way to do this, but this worked for me:

temp = float(int(10 * (temp + 0.4))) / 10;

[Mooser]

Maybe just change temp to an Int, and divide before the log? This way the gradual rounding errors won't accumulate and you'll always start from a cleanly represented int.

This may have other floating point issues though.

(Untested code warning)

var Int temp; 

simulated State MyState
{
	function BeginState(Name PreviousNameState)
	{
		`log(" -------------------------------------- MyState ");
	}
          
	functiom Tick(float DeltaTime)
	{
		local float floatTemp;

		Super.Tick(DeltaTime);

		temp += 4;
		floatTemp = temp / 10;

		`log(" ------------------------- TEMP : "@floatTemp );
	}
}

[Chihastonick]

Thankyou very much guys !!!

[Chihastonick]

WHY ???

var float temp;

simulated State MyState
{
	function BeginState(Name PreviousNameState)
	{
		`log(" -------------------------------------- MyState ");
	}
          
	functiom Tick(float DeltaTime)
	{
		Super.Tick(DeltaTime);

		temp = float(int(10*(temp + 0.4))) / 10;;

		`log(" ------------------------- TEMP : "@temp);
	}
}

Result :

[0038.31]   ------------- TEMP :  7.6000  
[0038.32]   ------------- TEMP :  8.0000 
[0038.33]   ------------- TEMP :  8.4000 
[0038.34]   ------------- TEMP :  8.7000 
[0038.35]   ------------- TEMP :  9.0000 
[0038.35]   ------------- TEMP :  9.4000 
[0038.36]   ------------- TEMP :  9.7000 
[0038.37]   ------------- TEMP :  10.0000 
[0038.38]   ------------- TEMP :  10.4000

[Mooser]

Wow, very interesting.

Maybe the compiler is doing something weird behind the scenes? Or the multiple conversions is abusive on code? Not sure. Did you try the Integer approach?

[VendorX]

Try this one: TestFloat = (int(TestFloat*10)+4)*0.1;

[Chihastonick]

Try this one: TestFloat = (int(TestFloat*10)+4)*0.1;

VendorX Thank you as always for your great help! Thanks also to others for answers.

[Angel_Mapper]

Ah yeah going that direction would be better, what was happening with mine was it was getting say, 8.79999 and truncating instead of rounding.

[Wormbo]

Wow, very interesting.

Maybe the compiler is doing something weird behind the scenes? Or the multiple conversions is abusive on code? Not sure. Did you try the Integer approach?

Read up on how floating-point numbers work internally. You'll see the compiler is not responsible for the rounding errors at all.

The real reason is, that 0.4 does not have a precise float value representation. That's because the format defines a number to be of the sum of 1 and a possibly empty combination of 1/2, 1/4, 1/8, ..., 1/4194304, 1/8388608. That sum is multiplied by 2 to the power of -126 to +127 and the result's sign can be plus or minus.
Like most fractional numbers, 0.4 lies between two actual floating point values. These are:
+(1 + 1/2 + 1/16 + 1/32 + 1/256 + 1/512 + 1/4096 + 1/8192 + 1/65536 + 1/131072 + 1/1048576 + 1/2097152) * 2^-2, ~= 0.399999976158
and
+(1 + 1/2 + 1/16 + 1/32 + 1/256 + 1/512 + 1/4096 + 1/8192 + 1/65536 + 1/131072 + 1/1048576 + 1/2097152 + 1/8388608) * 2^-2 ~= 0.40000000596
The latter is closer to 0.4, so it's the value used when you specify the literal "0.4" in your code.

Except for performing integer arithmetics as long as possible before applying a float operation, there's nothing you could do about this.

Personally I'd recommend:

var int counter;

function Tick(float DeltaTime)
{
  counter++;
  `log(counter * 0.4);
}

But I actually see another question arising here: Why add 0.4 in Tick and not DeltaTime? Or, if adding a constant value, why in Tick and not in a timer event?