The bigfloat package is a Python package providing arbitrary-precision
correctly-rounded binary floating-point arithmetic. It is implemented as a
Cython wrapper around the GNU MPFR library. A couple of lines of Python code should give the
idea:
>>> from bigfloat import *
>>> with precision(200) + RoundTowardZero:
... print(sqrt(2))
...
1.4142135623730950488016887242096980785696718753769480731766796
>>> with quadruple_precision:
... const_pi()
...
BigFloat.exact('3.14159265358979323846264338327950280', precision=113)
- Supports Python 2 (version 2.7) and Python 3 (version 3.5 or later).
- Exactly reproducible correctly-rounded results across platforms; precisely-defined semantics compatible with the IEEE 754-2008 standard.
- Support for mixed-type operations with Python integers and floats.
- Support for emulating IEEE 754 arithmetic in any of the IEEE binary interchange formats described in IEEE 754-2008. Infinities, NaNs, signed zeros, and subnormals are all supported.
- Easy control of rounding modes and precisions via
Contextobjects and Python'swithstatement.
Full package documentation is hosted at Read the Docs. Read on for a quick tour.
The bigfloat package is small and simple to use. Here's a quick
tour of some of its features.
For demonstration purposes, start with:
>>> from bigfloat import *
Note that this import shadows some builtin Python functions, namely abs,
max, min, pow, round, sum and (on Python 2 only) cmp.
In normal usage you'll probably only want to import the classes and functions
that you actually need.
The main class is the BigFloat class:
>>> BigFloat(1) # can be constructed from an integer, float or string
BigFloat.exact('1.00000000000000000000000000000000000', precision=113)
>>> BigFloat('3.14159') ** 2 / 6.0 # can combine with ints and floats
BigFloat.exact('1.64493128801666666666666666666666670', precision=113)
>>> BigFloat('0.1', precision(200)) # high-precision value from string
BigFloat.exact('0.1000000000000000000000000000000000000000000000000000
0000000002', precision=200)
Newly-created BigFloat instances refer to the current context to
determine what precision and rounding modes to use. This current
context is represented by a Context instance, and can be retrieved
by calling getcontext:
>>> getcontext()
Context(precision=113, emax=16384, emin=-16493,
subnormalize=True, rounding=ROUND_TIES_TO_EVEN)
The precision(200) argument passed to the BigFloat constructor
above is also an example of a Context:
>>> precision(200) Context(precision=200)
The context used for a calculation can be set using the setcontext
function, but a better way to make a temporary change to the context
is to use Python's with statement:
>>> with precision(1000): ... print sqrt(2) ... 1.41421356237309504880168872420969807856967187537694807317667973 7990732478462107038850387534327641572735013846230912297024924836 0558507372126441214970999358314132226659275055927557999505011527 8206057147010955997160597027453459686201472851741864088919860955 232923048430871432145083976260362799525140798964
Here, sqrt is one of a number of mathematical functions that the
bigfloat package exports. As you can see, these functions operate on
integers and floats as well as BigFloat instances, but always
return a BigFloat instance.
Rounding modes can be controlled similarly. Here are upper and lower bounds for π, accurate to 53 significant bits:
>>> with RoundTowardPositive:
... const_pi()
...
BigFloat.exact('3.14159265358979323846264338327950318', precision=113)
>>> with RoundTowardNegative:
... const_pi()
...
BigFloat.exact('3.14159265358979323846264338327950280', precision=113)
And as you'd expect, with statements like those above can be
nested. Context objects can also be combined using addition:
>>> with RoundTowardPositive + precision(24):
... BigFloat(1) / 3
...
BigFloat.exact('0.333333343', precision=24)
Various Context objects corresponding to IEEE 754 interchange
formats are predefined:
>>> quadruple_precision
Context(precision=113, emax=16384, emin=-16493, subnormalize=True)
>>> half_precision
Context(precision=11, emax=16, emin=-23, subnormalize=True)
>>> with half_precision:
log(2)
...
BigFloat.exact('0.69336', precision=11)
The bigfloat package is available on the Python package index, and can be installed in the usual
way using easy_install or pip. Alternatively, the development sources
may be downloaded from the project's homepage on GitHub.
For more comprehensive installation instructions, please see the full documentation.
Feedback is welcome! Please use the GitHub issue tracker to report issues. Alternatively, you can contact Mark Dickinson directly at [email protected] with suggestions, complaints, bug reports, etc.
The bigfloat package is copyright (C) 2009--2019 Mark Dickinson
The bigfloat package is free software: you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
The bigfloat package is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public License along with the bigfloat package. If not, see <http://www.gnu.org/licenses/>.