1.. _tut-fp-issues: 2 3************************************************** 4Floating Point Arithmetic: Issues and Limitations 5************************************************** 6 7.. sectionauthor:: Tim Peters <tim_one@users.sourceforge.net> 8 9 10Floating-point numbers are represented in computer hardware as base 2 (binary) 11fractions. For example, the decimal fraction :: 12 13 0.125 14 15has value 1/10 + 2/100 + 5/1000, and in the same way the binary fraction :: 16 17 0.001 18 19has value 0/2 + 0/4 + 1/8. These two fractions have identical values, the only 20real difference being that the first is written in base 10 fractional notation, 21and the second in base 2. 22 23Unfortunately, most decimal fractions cannot be represented exactly as binary 24fractions. A consequence is that, in general, the decimal floating-point 25numbers you enter are only approximated by the binary floating-point numbers 26actually stored in the machine. 27 28The problem is easier to understand at first in base 10. Consider the fraction 291/3. You can approximate that as a base 10 fraction:: 30 31 0.3 32 33or, better, :: 34 35 0.33 36 37or, better, :: 38 39 0.333 40 41and so on. No matter how many digits you're willing to write down, the result 42will never be exactly 1/3, but will be an increasingly better approximation of 431/3. 44 45In the same way, no matter how many base 2 digits you're willing to use, the 46decimal value 0.1 cannot be represented exactly as a base 2 fraction. In base 472, 1/10 is the infinitely repeating fraction :: 48 49 0.0001100110011001100110011001100110011001100110011... 50 51Stop at any finite number of bits, and you get an approximation. 52 53On a typical machine running Python, there are 53 bits of precision available 54for a Python float, so the value stored internally when you enter the decimal 55number ``0.1`` is the binary fraction :: 56 57 0.00011001100110011001100110011001100110011001100110011010 58 59which is close to, but not exactly equal to, 1/10. 60 61It's easy to forget that the stored value is an approximation to the original 62decimal fraction, because of the way that floats are displayed at the 63interpreter prompt. Python only prints a decimal approximation to the true 64decimal value of the binary approximation stored by the machine. If Python 65were to print the true decimal value of the binary approximation stored for 660.1, it would have to display :: 67 68 >>> 0.1 69 0.1000000000000000055511151231257827021181583404541015625 70 71That is more digits than most people find useful, so Python keeps the number 72of digits manageable by displaying a rounded value instead :: 73 74 >>> 0.1 75 0.1 76 77It's important to realize that this is, in a real sense, an illusion: the value 78in the machine is not exactly 1/10, you're simply rounding the *display* of the 79true machine value. This fact becomes apparent as soon as you try to do 80arithmetic with these values :: 81 82 >>> 0.1 + 0.2 83 0.30000000000000004 84 85Note that this is in the very nature of binary floating-point: this is not a 86bug in Python, and it is not a bug in your code either. You'll see the same 87kind of thing in all languages that support your hardware's floating-point 88arithmetic (although some languages may not *display* the difference by 89default, or in all output modes). 90 91Other surprises follow from this one. For example, if you try to round the 92value 2.675 to two decimal places, you get this :: 93 94 >>> round(2.675, 2) 95 2.67 96 97The documentation for the built-in :func:`round` function says that it rounds 98to the nearest value, rounding ties away from zero. Since the decimal fraction 992.675 is exactly halfway between 2.67 and 2.68, you might expect the result 100here to be (a binary approximation to) 2.68. It's not, because when the 101decimal string ``2.675`` is converted to a binary floating-point number, it's 102again replaced with a binary approximation, whose exact value is :: 103 104 2.67499999999999982236431605997495353221893310546875 105 106Since this approximation is slightly closer to 2.67 than to 2.68, it's rounded 107down. 108 109If you're in a situation where you care which way your decimal halfway-cases 110are rounded, you should consider using the :mod:`decimal` module. 111Incidentally, the :mod:`decimal` module also provides a nice way to "see" the 112exact value that's stored in any particular Python float :: 113 114 >>> from decimal import Decimal 115 >>> Decimal(2.675) 116 Decimal('2.67499999999999982236431605997495353221893310546875') 117 118Another consequence is that since 0.1 is not exactly 1/10, summing ten values 119of 0.1 may not yield exactly 1.0, either:: 120 121 >>> sum = 0.0 122 >>> for i in range(10): 123 ... sum += 0.1 124 ... 125 >>> sum 126 0.9999999999999999 127 128Binary floating-point arithmetic holds many surprises like this. The problem 129with "0.1" is explained in precise detail below, in the "Representation Error" 130section. See `The Perils of Floating Point <http://www.lahey.com/float.htm>`_ 131for a more complete account of other common surprises. 132 133As that says near the end, "there are no easy answers." Still, don't be unduly 134wary of floating-point! The errors in Python float operations are inherited 135from the floating-point hardware, and on most machines are on the order of no 136more than 1 part in 2\*\*53 per operation. That's more than adequate for most 137tasks, but you do need to keep in mind that it's not decimal arithmetic, and 138that every float operation can suffer a new rounding error. 139 140While pathological cases do exist, for most casual use of floating-point 141arithmetic you'll see the result you expect in the end if you simply round the 142display of your final results to the number of decimal digits you expect. For 143fine control over how a float is displayed see the :meth:`str.format` method's 144format specifiers in :ref:`formatstrings`. 145 146 147.. _tut-fp-error: 148 149Representation Error 150==================== 151 152This section explains the "0.1" example in detail, and shows how you can 153perform an exact analysis of cases like this yourself. Basic familiarity with 154binary floating-point representation is assumed. 155 156:dfn:`Representation error` refers to the fact that some (most, actually) 157decimal fractions cannot be represented exactly as binary (base 2) fractions. 158This is the chief reason why Python (or Perl, C, C++, Java, Fortran, and many 159others) often won't display the exact decimal number you expect:: 160 161 >>> 0.1 + 0.2 162 0.30000000000000004 163 164Why is that? 1/10 and 2/10 are not exactly representable as a binary 165fraction. Almost all machines today (July 2010) use IEEE-754 floating point 166arithmetic, and almost all platforms map Python floats to IEEE-754 "double 167precision". 754 doubles contain 53 bits of precision, so on input the computer 168strives to convert 0.1 to the closest fraction it can of the form *J*/2**\ *N* 169where *J* is an integer containing exactly 53 bits. Rewriting :: 170 171 1 / 10 ~= J / (2**N) 172 173as :: 174 175 J ~= 2**N / 10 176 177and recalling that *J* has exactly 53 bits (is ``>= 2**52`` but ``< 2**53``), 178the best value for *N* is 56:: 179 180 >>> 2**52 181 4503599627370496 182 >>> 2**53 183 9007199254740992 184 >>> 2**56/10 185 7205759403792793 186 187That is, 56 is the only value for *N* that leaves *J* with exactly 53 bits. 188The best possible value for *J* is then that quotient rounded:: 189 190 >>> q, r = divmod(2**56, 10) 191 >>> r 192 6 193 194Since the remainder is more than half of 10, the best approximation is obtained 195by rounding up:: 196 197 >>> q+1 198 7205759403792794 199 200Therefore the best possible approximation to 1/10 in 754 double precision is 201that over 2\*\*56, or :: 202 203 7205759403792794 / 72057594037927936 204 205Note that since we rounded up, this is actually a little bit larger than 1/10; 206if we had not rounded up, the quotient would have been a little bit smaller 207than 1/10. But in no case can it be *exactly* 1/10! 208 209So the computer never "sees" 1/10: what it sees is the exact fraction given 210above, the best 754 double approximation it can get:: 211 212 >>> .1 * 2**56 213 7205759403792794.0 214 215If we multiply that fraction by 10\*\*30, we can see the (truncated) value of 216its 30 most significant decimal digits:: 217 218 >>> 7205759403792794 * 10**30 // 2**56 219 100000000000000005551115123125L 220 221meaning that the exact number stored in the computer is approximately equal to 222the decimal value 0.100000000000000005551115123125. In versions prior to 223Python 2.7 and Python 3.1, Python rounded this value to 17 significant digits, 224giving '0.10000000000000001'. In current versions, Python displays a value 225based on the shortest decimal fraction that rounds correctly back to the true 226binary value, resulting simply in '0.1'. 227