1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
|
[manpage_begin math::fuzzy n 0.2]
[keywords floating-point]
[keywords math]
[keywords rounding]
[moddesc {Tcl Math Library}]
[titledesc {Fuzzy comparison of floating-point numbers}]
[category Mathematics]
[require Tcl [opt 8.3]]
[require math::fuzzy [opt 0.2]]
[description]
[para]
The package Fuzzy is meant to solve common problems with floating-point
numbers in a systematic way:
[list_begin itemized]
[item]
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.
[item]
Rounding a number that is halfway two integer numbers can cause
strange errors, like int(100.0*2.8) != 28 but 27
[list_end]
[para]
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 [emph 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.
[section "PROCEDURES"]
Effectively the package provides the following procedures:
[list_begin definitions]
[call [cmd ::math::fuzzy::teq] [arg value1] [arg value2]]
Compares two floating-point numbers and returns 1 if their values
fall within a small range. Otherwise it returns 0.
[call [cmd ::math::fuzzy::tne] [arg value1] [arg value2]]
Returns the negation, that is, if the difference is larger than
the margin, it returns 1.
[call [cmd ::math::fuzzy::tge] [arg value1] [arg 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.
[call [cmd ::math::fuzzy::tle] [arg value1] [arg value2]]
Returns 1 if the two numbers are equal according to
[lb]teq[rb] or if the first is smaller than the second.
[call [cmd ::math::fuzzy::tlt] [arg value1] [arg value2]]
Returns the opposite of [lb]tge[rb].
[call [cmd ::math::fuzzy::tgt] [arg value1] [arg value2]]
Returns the opposite of [lb]tle[rb].
[call [cmd ::math::fuzzy::tfloor] [arg value]]
Returns the integer number that is lower or equal
to the given floating-point number, within a well-defined
tolerance.
[call [cmd ::math::fuzzy::tceil] [arg value]]
Returns the integer number that is greater or equal to the given
floating-point number, within a well-defined tolerance.
[call [cmd ::math::fuzzy::tround] [arg value]]
Rounds the floating-point number off.
[call [cmd ::math::fuzzy::troundn] [arg value] [arg ndigits]]
Rounds the floating-point number off to the
specified number of decimals (Pro memorie).
[list_end]
Usage:
[example_begin]
if { [lb]teq $x $y[rb] } { puts "x == y" }
if { [lb]tne $x $y[rb] } { puts "x != y" }
if { [lb]tge $x $y[rb] } { puts "x >= y" }
if { [lb]tgt $x $y[rb] } { puts "x > y" }
if { [lb]tlt $x $y[rb] } { puts "x < y" }
if { [lb]tle $x $y[rb] } { puts "x <= y" }
set fx [lb]tfloor $x[rb]
set fc [lb]tceil $x[rb]
set rounded [lb]tround $x[rb]
set roundn [lb]troundn $x $nodigits[rb]
[example_end]
[section {TEST CASES}]
The problems that can occur with floating-point numbers are illustrated
by the test cases in the file "fuzzy.test":
[list_begin itemized]
[item]
Several test case use the ordinary comparisons, and they
fail invariably to produce understandable results
[item]
One test case uses [lb]expr[rb] without braces ({ and }). It too
fails.
[list_end]
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.
[para]
Note that besides the Tcl-only package, there is also a C-based version.
[section REFERENCES]
Original implementation in Fortran by dr. H.D. Knoble (Penn State
University).
[para]
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).
[para]
L. M. Breed, "Definitions for Fuzzy Floor and Ceiling",
APL QUOTE QUAD 8(3):16-23, March 1978.
[para]
D. Knuth, Art of Computer Programming,
Vol. 1, Problem 1.2.4-5.
[vset CATEGORY {math :: fuzzy}]
[include ../common-text/feedback.inc]
[manpage_end]
|
