Warning
End-of-life notification
This library was designed to bring alternative generators to the NumPy infrastructure. It as been successful in advancing the conversation for a future implementation of a new random number API in NumPy which will allow new algorithms and/or generators. The next step in this process is to separate the basic (or core RNG) from the functions that transform random bits into useful random numbers. This has been implemented in a successor project randomgen available on GitHub or PyPi.
randomgen has a slightly different API, so please see the randomgen documentation.
This package contains drop-in replacements for the NumPy RandomState object that change the core random number generator.
randomstate.entropy.random_entropy() provides access to the system
source of randomness that is used in cryptographic applications (e.g.,
/dev/urandom on Unix).complex_normal())standard_normal(),
standard_exponential() or
standard_gamma().
The Ziggurat generator can be accessed by passing the keyword
argument method='zig'.In [1]: from randomstate.prng.xoroshiro128plus import standard_normal
In [2]: %timeit standard_normal(1000000, method='bm')
...: %timeit standard_normal(1000000, method='zig')
...:
43.4 ms +- 301 us per loop (mean +- std. dev. of 7 runs, 10 loops each)
9.89 ms +- 60.5 us per loop (mean +- std. dev. of 7 runs, 100 loops each)
In [3]: from randomstate.prng.xoroshiro128plus import standard_exponential
In [4]: %timeit standard_exponential(1000000, method='inv')
...: %timeit standard_exponential(1000000, method='zig')
...:
57.1 ms +- 458 us per loop (mean +- std. dev. of 7 runs, 10 loops each)
5.32 ms +- 68 us per loop (mean +- std. dev. of 7 runs, 100 loops each)
In [5]: from randomstate.prng.xoroshiro128plus import standard_gamma
In [6]: %timeit standard_gamma(3.0, 1000000, method='inv')
...: %timeit standard_gamma(3.0, 1000000, method='zig')
...:
57.7 ms +- 263 us per loop (mean +- std. dev. of 7 runs, 10 loops each)
27.5 ms +- 346 us per loop (mean +- std. dev. of 7 runs, 10 loops each)
dtype argument that accepts np.float32 or np.float64
to produce either single or double prevision uniform random variables for
select distributionsrandom_sample() and
rand())standard_normal() and
randn())standard_gamma())standard_exponential())In [7]: import randomstate as rs
In [8]: rs.seed(0)
In [9]: rs.random_sample(3, dtype='d')
Out[9]: array([0.5488135 , 0.71518937, 0.60276338])
In [10]: rs.seed(0)
In [11]: rs.random_sample(3, dtype='f')
Out[11]: array([0.54881346, 0.5928446 , 0.71518934], dtype=float32)
Optional out argument that allows existing arrays to be filled for
select distributions
random_sample())standard_normal())standard_gamma())standard_exponential())This allows multithreading to fill large arrays in chunks using suitable PRNGs in parallel.
In [12]: import numpy as np
In [13]: import randomstate as rs
In [14]: existing = np.zeros(4)
In [15]: rs.seed(0)
In [16]: rs.random_sample(out=existing[:2])
Out[16]: array([0.5488135 , 0.71518937])
In [17]: print(existing)
[0.5488135 0.71518937 0. 0. ]
The included generators can be used in parallel, distributed applications in one of two ways:
The main innovation is the inclusion of a number of alternative pseudo-random number generators, ‘in addition’ to the standard PRNG in NumPy. The included PRNGs are:
randomstate.prng.mt19937.jump()).
See NumPy’s documentation.jump and so can
be used in parallel applications. See the dSFMT authors’ page.randomstate.prng.xoroshiro128plus.jump() for details.
More information about this PRNG is available at the
xorshift and xoroshiro authors’ page.jump and so can be used in
parallel applications. See the documentation for
randomstate.prng.xorshift1024.jump() for details. More information
about these PRNGs is available at the
xorshift and xoroshiro authors’ page.randomstate.prng.pcg64.advance(). PCG-32 only as a period of
\(2^{64}\) while PCG-64 has a period of \(2^{128}\). See the
PCG author’s page for more details about this class of PRNG.