Oihana PHP

shedFloatNoise.php

Table of Contents

Functions

shedFloatNoise()  : float
Sheds the binary noise a computed figure carries, and rounds no business figure.

Functions

shedFloatNoise()

Sheds the binary noise a computed figure carries, and rounds no business figure.

shedFloatNoise(int|float $value[, int $digits = 12 ][, int|float|null $scale = null ]) : float

Adding floats leaves an error that grows with the figure : a sum of two hundred thousand comes back as 202799.4000000001, a sum of forty five thousand as 45292.249999999985. A fixed count of decimals — ten, say — cleans the second and leaves the first, because the noise of a double sits at a relative distance of the value, not at a fixed decimal.

🔑 The scale follows the size of the value. The figure is rounded to a count of significant digits : twelve by default, where a double holds fifteen to seventeen — the last ones are left to the noise, and every decimal a source records survives, a quantity of 13039.42858, a weight of 55999.81828, a volume of 0.0073.

⚠️ Not a rounding. Two decimals on an amount or five on a quantity are business rules, written where the figure is shown or stored (roundValue()). This function only removes what an addition of floats left behind : a figure that was never computed needs none of it.

🚨 A sum whose terms cancel needs the scale of its terms. The noise an addition leaves is relative to the size of the figures added, not to the size of the result. When they nearly cancel, the result is tiny and twelve of its own digits are the noise itself : 1000.01 - 1000.00 holds 0.009999999999990905. Given the size of the largest term as $scale, the digits are reckoned on it — twelve at the scale of a thousand are eight decimals — and the figure comes back as 0.01. The larger of the value and the scale is the one used, so a scale never keeps more digits than the value alone would.

The result is never a negative zero : a dust below zero shed to nothing is 0.0, not -0.0. An infinite figure and a NaN are answered as they are.

Parameters
$value : int|float

The computed figure.

$digits : int = 12

How many significant digits to keep. Default 12.

$scale : int|float|null = null

The size the digits are reckoned on — the largest term of a sum. Default null : the size of the value itself.

Tags
example 
use function oihana\core\maths\shedFloatNoise;

echo shedFloatNoise( 0.1 + 0.2 )          ; // 0.3
echo shedFloatNoise( 202799.4000000001 )  ; // 202799.4
echo shedFloatNoise( 45292.249999999985 ) ; // 45292.25
echo shedFloatNoise( 13039.42858 )        ; // 13039.42858 — a real decimal survives
echo shedFloatNoise( 0.0073 )             ; // 0.0073

// Terms that cancel : the result alone is too small to say where the noise starts.
echo shedFloatNoise( 1000.01 - 1000.00 )                   ; // 0.00999999999999
echo shedFloatNoise( 1000.01 - 1000.00 , scale: 1000.01 )  ; // 0.01
echo shedFloatNoise( 1000.00 - 999.99 - 0.01 , scale: 1000 ) ; // 0
author

Marc Alcaraz (ekameleon)

since
1.3.0
Return values
float —

The figure, rid of its noise.

On this page

Search results