-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathfloatingpoint.html
571 lines (565 loc) · 21.1 KB
/
floatingpoint.html
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
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml" lang="" xml:lang="">
<head>
<meta charset="utf-8" />
<meta name="generator" content="pandoc" />
<meta
name="viewport"
content="width=device-width, initial-scale=1.0, user-scalable=yes"
/>
<title>floatingpoint</title>
<style type="text/css">
code {
white-space: pre-wrap;
}
span.smallcaps {
font-variant: small-caps;
}
span.underline {
text-decoration: underline;
}
div.column {
display: inline-block;
vertical-align: top;
width: 50%;
}
</style>
</head>
<body>
<h3 id="navigation">Navigation</h3>
<ul>
<li>
<a href="https://docs.python.org/3/genindex.html" title="General Index"
>index</a
>
</li>
<li>
<a
href="https://docs.python.org/3/py-modindex.html"
title="Python Module Index"
>modules</a
>
|
</li>
<li><a href="appendix.html" title="16. Appendix">next</a> |</li>
<li>
<a
href="interactive.html"
title="14. Interactive Input Editing and History Substitution"
>previous</a
>
|
</li>
<li><img src="../_static/py.png" /></li>
<li><a href="https://www.python.org/">Python</a> »</li>
<li>
<a href="https://docs.python.org/3/index.html">3.9.5 Documentation</a> »
</li>
<li><a href="index.html">The Python Tutorial</a> »</li>
<li></li>
</ul>
<p><span id="tut-fp-issues"></span></p>
<h1 id="floating-point-arithmetic-issues-and-limitations">
<span class="section-number">15. </span>Floating Point Arithmetic: Issues
and Limitations<a
href="#floating-point-arithmetic-issues-and-limitations"
class="headerlink"
title="Permalink to this headline"
>¶</a
>
</h1>
<p>
Floating-point numbers are represented in computer hardware as base 2
(binary) fractions. For example, the decimal fraction
</p>
<pre><code>0.125</code></pre>
<p>
has value 1/10 + 2/100 + 5/1000, and in the same way the binary fraction
</p>
<pre><code>0.001</code></pre>
<p>
has value 0/2 + 0/4 + 1/8. These two fractions have identical values, the
only real difference being that the first is written in base 10 fractional
notation, and the second in base 2.
</p>
<p>
Unfortunately, most decimal fractions cannot be represented exactly as
binary fractions. A consequence is that, in general, the decimal
floating-point numbers you enter are only approximated by the binary
floating-point numbers actually stored in the machine.
</p>
<p>
The problem is easier to understand at first in base 10. Consider the
fraction 1/3. You can approximate that as a base 10 fraction:
</p>
<pre><code>0.3</code></pre>
<p>or, better,</p>
<pre><code>0.33</code></pre>
<p>or, better,</p>
<pre><code>0.333</code></pre>
<p>
and so on. No matter how many digits you’re willing to write down, the
result will never be exactly 1/3, but will be an increasingly better
approximation of 1/3.
</p>
<p>
In the same way, no matter how many base 2 digits you’re willing to use,
the decimal value 0.1 cannot be represented exactly as a base 2 fraction.
In base 2, 1/10 is the infinitely repeating fraction
</p>
<pre><code>0.0001100110011001100110011001100110011001100110011...</code></pre>
<p>
Stop at any finite number of bits, and you get an approximation. On most
machines today, floats are approximated using a binary fraction with the
numerator using the first 53 bits starting with the most significant bit
and with the denominator as a power of two. In the case of 1/10, the
binary fraction is <code>3602879701896397 / 2 ** 55</code> which is close
to but not exactly equal to the true value of 1/10.
</p>
<p>
Many users are not aware of the approximation because of the way values
are displayed. Python only prints a decimal approximation to the true
decimal value of the binary approximation stored by the machine. On most
machines, if Python were to print the true decimal value of the binary
approximation stored for 0.1, it would have to display
</p>
<pre><code>>>> 0.1
0.1000000000000000055511151231257827021181583404541015625</code></pre>
<p>
That is more digits than most people find useful, so Python keeps the
number of digits manageable by displaying a rounded value instead
</p>
<pre><code>>>> 1 / 10
0.1</code></pre>
<p>
Just remember, even though the printed result looks like the exact value
of 1/10, the actual stored value is the nearest representable binary
fraction.
</p>
<p>
Interestingly, there are many different decimal numbers that share the
same nearest approximate binary fraction. For example, the numbers
<code>0.1</code> and <code>0.10000000000000001</code> and
<code>0.1000000000000000055511151231257827021181583404541015625</code> are
all approximated by <code>3602879701896397 / 2 ** 55</code>. Since all of
these decimal values share the same approximation, any one of them could
be displayed while still preserving the invariant
<code>eval(repr(x)) == x</code>.
</p>
<p>
Historically, the Python prompt and built-in
<a
href="https://docs.python.org/3/library/functions.html#repr"
class="reference internal"
title="repr"
><code class="sourceCode python"
><span class="bu">repr</span>()</code
></a
>
function would choose the one with 17 significant digits,
<code>0.10000000000000001</code>. Starting with Python 3.1, Python (on
most systems) is now able to choose the shortest of these and simply
display <code>0.1</code>.
</p>
<p>
Note that this is in the very nature of binary floating-point: this is not
a bug in Python, and it is not a bug in your code either. You’ll see the
same kind of thing in all languages that support your hardware’s
floating-point arithmetic (although some languages may not
<em>display</em> the difference by default, or in all output modes).
</p>
<p>
For more pleasant output, you may wish to use string formatting to produce
a limited number of significant digits:
</p>
<pre><code>>>> format(math.pi, '.12g') # give 12 significant digits
'3.14159265359'
>>> format(math.pi, '.2f') # give 2 digits after the point
'3.14'
>>> repr(math.pi)
'3.141592653589793'</code></pre>
<p>
It’s important to realize that this is, in a real sense, an illusion:
you’re simply rounding the <em>display</em> of the true machine value.
</p>
<p>
One illusion may beget another. For example, since 0.1 is not exactly
1/10, summing three values of 0.1 may not yield exactly 0.3, either:
</p>
<pre><code>>>> .1 + .1 + .1 == .3
False</code></pre>
<p>
Also, since the 0.1 cannot get any closer to the exact value of 1/10 and
0.3 cannot get any closer to the exact value of 3/10, then pre-rounding
with
<a
href="https://docs.python.org/3/library/functions.html#round"
class="reference internal"
title="round"
><code class="sourceCode python"
><span class="bu">round</span>()</code
></a
>
function cannot help:
</p>
<pre><code>>>> round(.1, 1) + round(.1, 1) + round(.1, 1) == round(.3, 1)
False</code></pre>
<p>
Though the numbers cannot be made closer to their intended exact values,
the
<a
href="https://docs.python.org/3/library/functions.html#round"
class="reference internal"
title="round"
><code class="sourceCode python"
><span class="bu">round</span>()</code
></a
>
function can be useful for post-rounding so that results with inexact
values become comparable to one another:
</p>
<pre><code>>>> round(.1 + .1 + .1, 10) == round(.3, 10)
True</code></pre>
<p>
Binary floating-point arithmetic holds many surprises like this. The
problem with “0.1” is explained in precise detail below, in the
“Representation Error” section. See
<a href="http://www.lahey.com/float.htm" class="reference external"
>The Perils of Floating Point</a
>
for a more complete account of other common surprises.
</p>
<p>
As that says near the end, “there are no easy answers.” Still, don’t be
unduly wary of floating-point! The errors in Python float operations are
inherited from the floating-point hardware, and on most machines are on
the order of no more than 1 part in 2**53 per operation. That’s more than
adequate for most tasks, but you do need to keep in mind that it’s not
decimal arithmetic and that every float operation can suffer a new
rounding error.
</p>
<p>
While pathological cases do exist, for most casual use of floating-point
arithmetic you’ll see the result you expect in the end if you simply round
the display of your final results to the number of decimal digits you
expect.
<a
href="https://docs.python.org/3/library/stdtypes.html#str"
class="reference internal"
title="str"
><code class="sourceCode python"><span class="bu">str</span>()</code></a
>
usually suffices, and for finer control see the
<a
href="https://docs.python.org/3/library/stdtypes.html#str.format"
class="reference internal"
title="str.format"
><code class="sourceCode python"
><span class="bu">str</span>.<span class="bu">format</span>()</code
></a
>
method’s format specifiers in
<a
href="https://docs.python.org/3/library/string.html#formatstrings"
class="reference internal"
><span class="std std-ref">Format String Syntax</span></a
>.
</p>
<p>
For use cases which require exact decimal representation, try using the
<a
href="https://docs.python.org/3/library/decimal.html#module-decimal"
class="reference internal"
title="decimal: Implementation of the General Decimal Arithmetic Specification."
><code class="sourceCode python">decimal</code></a
>
module which implements decimal arithmetic suitable for accounting
applications and high-precision applications.
</p>
<p>
Another form of exact arithmetic is supported by the
<a
href="https://docs.python.org/3/library/fractions.html#module-fractions"
class="reference internal"
title="fractions: Rational numbers."
><code class="sourceCode python">fractions</code></a
>
module which implements arithmetic based on rational numbers (so the
numbers like 1/3 can be represented exactly).
</p>
<p>
If you are a heavy user of floating point operations you should take a
look at the Numerical Python package and many other packages for
mathematical and statistical operations supplied by the SciPy project. See
<<a href="https://scipy.org/" class="reference external"
>https://scipy.org</a
>>.
</p>
<p>
Python provides tools that may help on those rare occasions when you
really <em>do</em> want to know the exact value of a float. The
<a
href="https://docs.python.org/3/library/stdtypes.html#float.as_integer_ratio"
class="reference internal"
title="float.as_integer_ratio"
><code class="sourceCode python"
><span class="bu">float</span>.as_integer_ratio()</code
></a
>
method expresses the value of a float as a fraction:
</p>
<pre><code>>>> x = 3.14159
>>> x.as_integer_ratio()
(3537115888337719, 1125899906842624)</code></pre>
<p>
Since the ratio is exact, it can be used to losslessly recreate the
original value:
</p>
<pre><code>>>> x == 3537115888337719 / 1125899906842624
True</code></pre>
<p>
The
<a
href="https://docs.python.org/3/library/stdtypes.html#float.hex"
class="reference internal"
title="float.hex"
><code class="sourceCode python"
><span class="bu">float</span>.<span class="bu">hex</span>()</code
></a
>
method expresses a float in hexadecimal (base 16), again giving the exact
value stored by your computer:
</p>
<pre><code>>>> x.hex()
'0x1.921f9f01b866ep+1'</code></pre>
<p>
This precise hexadecimal representation can be used to reconstruct the
float value exactly:
</p>
<pre><code>>>> x == float.fromhex('0x1.921f9f01b866ep+1')
True</code></pre>
<p>
Since the representation is exact, it is useful for reliably porting
values across different versions of Python (platform independence) and
exchanging data with other languages that support the same format (such as
Java and C99).
</p>
<p>
Another helpful tool is the
<a
href="https://docs.python.org/3/library/math.html#math.fsum"
class="reference internal"
title="math.fsum"
><code class="sourceCode python">math.fsum()</code></a
>
function which helps mitigate loss-of-precision during summation. It
tracks “lost digits” as values are added onto a running total. That can
make a difference in overall accuracy so that the errors do not accumulate
to the point where they affect the final total:
</p>
<pre><code>>>> sum([0.1] * 10) == 1.0
False
>>> math.fsum([0.1] * 10) == 1.0
True</code></pre>
<p><span id="tut-fp-error"></span></p>
<h2 id="representation-error">
<span class="section-number">15.1. </span>Representation Error<a
href="#representation-error"
class="headerlink"
title="Permalink to this headline"
>¶</a
>
</h2>
<p>
This section explains the “0.1” example in detail, and shows how you can
perform an exact analysis of cases like this yourself. Basic familiarity
with binary floating-point representation is assumed.
</p>
<p>
<em>Representation error</em> refers to the fact that some (most,
actually) decimal fractions cannot be represented exactly as binary (base
2) fractions. This is the chief reason why Python (or Perl, C, C++, Java,
Fortran, and many others) often won’t display the exact decimal number you
expect.
</p>
<p>
Why is that? 1/10 is not exactly representable as a binary fraction.
Almost all machines today (November 2000) use IEEE-754 floating point
arithmetic, and almost all platforms map Python floats to IEEE-754 “double
precision”. 754 doubles contain 53 bits of precision, so on input the
computer strives to convert 0.1 to the closest fraction it can of the form
<em>J</em>/2**<em>N</em> where <em>J</em> is an integer containing exactly
53 bits. Rewriting
</p>
<pre><code>1 / 10 ~= J / (2**N)</code></pre>
<p>as</p>
<pre><code>J ~= 2**N / 10</code></pre>
<p>
and recalling that <em>J</em> has exactly 53 bits (is
<code>>= 2**52</code> but <code>< 2**53</code>), the best value for
<em>N</em> is 56:
</p>
<pre><code>>>> 2**52 <= 2**56 // 10 < 2**53
True</code></pre>
<p>
That is, 56 is the only value for <em>N</em> that leaves <em>J</em> with
exactly 53 bits. The best possible value for <em>J</em> is then that
quotient rounded:
</p>
<pre><code>>>> q, r = divmod(2**56, 10)
>>> r
6</code></pre>
<p>
Since the remainder is more than half of 10, the best approximation is
obtained by rounding up:
</p>
<pre><code>>>> q+1
7205759403792794</code></pre>
<p>
Therefore the best possible approximation to 1/10 in 754 double precision
is:
</p>
<pre><code>7205759403792794 / 2 ** 56</code></pre>
<p>
Dividing both the numerator and denominator by two reduces the fraction
to:
</p>
<pre><code>3602879701896397 / 2 ** 55</code></pre>
<p>
Note that since we rounded up, this is actually a little bit larger than
1/10; if we had not rounded up, the quotient would have been a little bit
smaller than 1/10. But in no case can it be <em>exactly</em> 1/10!
</p>
<p>
So the computer never “sees” 1/10: what it sees is the exact fraction
given above, the best 754 double approximation it can get:
</p>
<pre><code>>>> 0.1 * 2 ** 55
3602879701896397.0</code></pre>
<p>
If we multiply that fraction by 10**55, we can see the value out to 55
decimal digits:
</p>
<pre><code>>>> 3602879701896397 * 10 ** 55 // 2 ** 55
1000000000000000055511151231257827021181583404541015625</code></pre>
<p>
meaning that the exact number stored in the computer is equal to the
decimal value 0.1000000000000000055511151231257827021181583404541015625.
Instead of displaying the full decimal value, many languages (including
older versions of Python), round the result to 17 significant digits:
</p>
<pre><code>>>> format(0.1, '.17f')
'0.10000000000000001'</code></pre>
<p>
The
<a
href="https://docs.python.org/3/library/fractions.html#module-fractions"
class="reference internal"
title="fractions: Rational numbers."
><code class="sourceCode python">fractions</code></a
>
and
<a
href="https://docs.python.org/3/library/decimal.html#module-decimal"
class="reference internal"
title="decimal: Implementation of the General Decimal Arithmetic Specification."
><code class="sourceCode python">decimal</code></a
>
modules make these calculations easy:
</p>
<pre><code>>>> from decimal import Decimal
>>> from fractions import Fraction
>>> Fraction.from_float(0.1)
Fraction(3602879701896397, 36028797018963968)
>>> (0.1).as_integer_ratio()
(3602879701896397, 36028797018963968)
>>> Decimal.from_float(0.1)
Decimal('0.1000000000000000055511151231257827021181583404541015625')
>>> format(Decimal.from_float(0.1), '.17')
'0.10000000000000001'</code></pre>
<h3 id="table-of-contents">
<a href="https://docs.python.org/3/contents.html">Table of Contents</a>
</h3>
<ul>
<li>
<a href="#" class="reference internal"
>15. Floating Point Arithmetic: Issues and Limitations</a
>
<ul>
<li>
<a href="#representation-error" class="reference internal"
>15.1. Representation Error</a
>
</li>
</ul>
</li>
</ul>
<h4 id="previous-topic">Previous topic</h4>
<p>
<a href="interactive.html" title="previous chapter"
><span class="section-number">14. </span>Interactive Input Editing and
History Substitution</a
>
</p>
<h4 id="next-topic">Next topic</h4>
<p>
<a href="appendix.html" title="next chapter"
><span class="section-number">16. </span>Appendix</a
>
</p>
<h3 id="this-page">This Page</h3>
<ul>
<li><a href="https://docs.python.org/3/bugs.html">Report a Bug</a></li>
<li>
<a
href="https://github.com/python/cpython/blob/3.9/Doc/tutorial/floatingpoint.rst"
>Show Source</a
>
</li>
</ul>
<h3 id="navigation-1">Navigation</h3>
<ul>
<li>
<a href="https://docs.python.org/3/genindex.html" title="General Index"
>index</a
>
</li>
<li>
<a
href="https://docs.python.org/3/py-modindex.html"
title="Python Module Index"
>modules</a
>
|
</li>
<li><a href="appendix.html" title="16. Appendix">next</a> |</li>
<li>
<a
href="interactive.html"
title="14. Interactive Input Editing and History Substitution"
>previous</a
>
|
</li>
<li><img src="../_static/py.png" /></li>
<li><a href="https://www.python.org/">Python</a> »</li>
<li>
<a href="https://docs.python.org/3/index.html">3.9.5 Documentation</a> »
</li>
<li><a href="index.html">The Python Tutorial</a> »</li>
<li></li>
</ul>
<p>
©
<a href="https://docs.python.org/3/copyright.html">Copyright</a>
2001-2021, Python Software Foundation.<br />
The Python Software Foundation is a non-profit corporation.
<a href="https://www.python.org/psf/donations/">Please donate.</a>
</p>
<p>
Last updated on May 30, 2021.
<a href="https://docs.python.org/3/bugs.html">Found a bug</a>?<br />
Created using <a href="https://www.sphinx-doc.org/">Sphinx</a> 2.4.4.
</p>
</body>
</html>