Name

math::fuzzy - Fuzzy comparison of floating-point numbers

Table Of Contents

• Table Of Contents
• Synopsis
• Description
• PROCEDURES
• TEST CASES
• REFERENCES
• Bugs, Ideas, Feedback
• Keywords
• Category

Synopsis

Description

The package Fuzzy is meant to solve common problems with floating-point numbers in a systematic way:

• Comparing two numbers that are "supposed" to be identical, like 1.0 and 2.1/(1.2+0.9) is not guaranteed to give the intuitive result.

• Rounding a number that is halfway two integer numbers can cause strange errors, like int(100.0*2.8) != 28 but 27

The Fuzzy package is meant to help sorting out this type of problems by defining "fuzzy" comparison procedures for floating-point numbers. It does so by allowing for a small margin that is determined automatically - the margin is three times the "epsilon" value, that is three times the smallest number eps such that 1.0 and 1.0+$eps canbe distinguished. In Tcl, which uses double precision floating-point numbers, this is typically 1.1e-16.

PROCEDURES

Effectively the package provides the following procedures:

::math::fuzzy::teq value1 value2

Compares two floating-point numbers and returns 1 if their values fall within a small range. Otherwise it returns 0.

::math::fuzzy::tne value1 value2

Returns the negation, that is, if the difference is larger than the margin, it returns 1.

::math::fuzzy::tge value1 value2

Compares two floating-point numbers and returns 1 if their values either fall within a small range or if the first number is larger than the second. Otherwise it returns 0.

::math::fuzzy::tle value1 value2

Returns 1 if the two numbers are equal according to [teq] or if the first is smaller than the second.

::math::fuzzy::tlt value1 value2

Returns the opposite of [tge].

::math::fuzzy::tgt value1 value2

Returns the opposite of [tle].

::math::fuzzy::tfloor value

Returns the integer number that is lower or equal to the given floating-point number, within a well-defined tolerance.

::math::fuzzy::tceil value

Returns the integer number that is greater or equal to the given floating-point number, within a well-defined tolerance.

::math::fuzzy::tround value

Rounds the floating-point number off.

::math::fuzzy::troundn value ndigits

Rounds the floating-point number off to the specified number of decimals (Pro memorie).

Usage:

if { [teq $x $y] } { puts "x == y" }
if { [tne $x $y] } { puts "x != y" }
if { [tge $x $y] } { puts "x >= y" }
if { [tgt $x $y] } { puts "x > y" }
if { [tlt $x $y] } { puts "x < y" }
if { [tle $x $y] } { puts "x <= y" }
set fx      [tfloor $x]
set fc      [tceil  $x]
set rounded [tround $x]
set roundn  [troundn $x $nodigits]

TEST CASES

The problems that can occur with floating-point numbers are illustrated by the test cases in the file "fuzzy.test":

• Several test case use the ordinary comparisons, and they fail invariably to produce understandable results

• One test case uses [expr] without braces ({ and }). It too fails.

The conclusion from this is that any expression should be surrounded by braces, because otherwise very awkward things can happen if you need accuracy. Furthermore, accuracy and understandable results are enhanced by using these "tolerant" or fuzzy comparisons.

Note that besides the Tcl-only package, there is also a C-based version.

REFERENCES

Original implementation in Fortran by dr. H.D. Knoble (Penn State University).

P. E. Hagerty, "More on Fuzzy Floor and Ceiling," APL QUOTE QUAD 8(4):20-24, June 1978. Note that TFLOOR=FL5 took five years of refereed evolution (publication).

L. M. Breed, "Definitions for Fuzzy Floor and Ceiling", APL QUOTE QUAD 8(3):16-23, March 1978.

D. Knuth, Art of Computer Programming, Vol. 1, Problem 1.2.4-5.

Bugs, Ideas, Feedback

This document, and the package it describes, will undoubtedly contain bugs and other problems. Please report such in the category math :: fuzzy of the Tcllib Trackers. Please also report any ideas for enhancements you may have for either package and/or documentation.

When proposing code changes, please provide unified diffs, i.e the output of diff -u.

Note further that attachments are strongly preferred over inlined patches. Attachments can be made by going to the Edit form of the ticket immediately after its creation, and then using the left-most button in the secondary navigation bar.

Keywords

floating-point, math, rounding

Category

Mathematics