1 """Random variable generators.
11 generate random permutation
13 distributions on the real line:
14 ------------------------------
25 distributions on the circle (angles 0 to 2pi)
26 ---------------------------------------------
30 General notes on the underlying Mersenne Twister core generator:
32 * The period is 2**19937-1.
33 * It is one of the most extensively tested generators in existence.
34 * Without a direct way to compute N steps forward, the semantics of
35 jumpahead(n) are weakened to simply jump to another distant state and rely
36 on the large period to avoid overlapping sequences.
37 * The random() method is implemented in C, executes in a single Python step,
38 and is, therefore, threadsafe.
42 from __future__ import division
43 from warnings import warn as _warn
44 from types import MethodType as _MethodType, BuiltinMethodType as _BuiltinMethodType
45 from math import log as _log, exp as _exp, pi as _pi, e as _e, ceil as _ceil
46 from math import sqrt as _sqrt, acos as _acos, cos as _cos, sin as _sin
47 from os import urandom as _urandom
48 from binascii import hexlify as _hexlify
49 import hashlib as _hashlib
51 __all__ = ["Random","seed","random","uniform","randint","choice","sample",
52 "randrange","shuffle","normalvariate","lognormvariate",
53 "expovariate","vonmisesvariate","gammavariate","triangular",
54 "gauss","betavariate","paretovariate","weibullvariate",
55 "getstate","setstate","jumpahead", "WichmannHill", "getrandbits",
58 NV_MAGICCONST = 4 * _exp(-0.5)/_sqrt(2.0)
61 SG_MAGICCONST = 1.0 + _log(4.5)
62 BPF = 53 # Number of bits in a float
66 # Translated by Guido van Rossum from C source provided by
67 # Adrian Baddeley. Adapted by Raymond Hettinger for use with
68 # the Mersenne Twister and os.urandom() core generators.
72 class Random(_random.Random):
73 """Random number generator base class used by bound module functions.
75 Used to instantiate instances of Random to get generators that don't
76 share state. Especially useful for multi-threaded programs, creating
77 a different instance of Random for each thread, and using the jumpahead()
78 method to ensure that the generated sequences seen by each thread don't
81 Class Random can also be subclassed if you want to use a different basic
82 generator of your own devising: in that case, override the following
83 methods: random(), seed(), getstate(), setstate() and jumpahead().
84 Optionally, implement a getrandbits() method so that randrange() can cover
85 arbitrarily large ranges.
89 VERSION = 3 # used by getstate/setstate
91 def __init__(self, x=None):
92 """Initialize an instance.
94 Optional argument x controls seeding, as for Random.seed().
98 self.gauss_next = None
100 def seed(self, a=None):
101 """Initialize internal state from hashable object.
103 None or no argument seeds from current time or from an operating
104 system specific randomness source if available.
106 If a is not None or an int or long, hash(a) is used instead.
111 a = long(_hexlify(_urandom(16)), 16)
112 except NotImplementedError:
114 a = long(time.time() * 256) # use fractional seconds
116 super(Random, self).seed(a)
117 self.gauss_next = None
120 """Return internal state; can be passed to setstate() later."""
121 return self.VERSION, super(Random, self).getstate(), self.gauss_next
123 def setstate(self, state):
124 """Restore internal state from object returned by getstate()."""
127 version, internalstate, self.gauss_next = state
128 super(Random, self).setstate(internalstate)
130 version, internalstate, self.gauss_next = state
131 # In version 2, the state was saved as signed ints, which causes
132 # inconsistencies between 32/64-bit systems. The state is
133 # really unsigned 32-bit ints, so we convert negative ints from
134 # version 2 to positive longs for version 3.
136 internalstate = tuple( long(x) % (2**32) for x in internalstate )
137 except ValueError, e:
139 super(Random, self).setstate(internalstate)
141 raise ValueError("state with version %s passed to "
142 "Random.setstate() of version %s" %
143 (version, self.VERSION))
145 def jumpahead(self, n):
146 """Change the internal state to one that is likely far away
147 from the current state. This method will not be in Py3.x,
148 so it is better to simply reseed.
150 # The super.jumpahead() method uses shuffling to change state,
151 # so it needs a large and "interesting" n to work with. Here,
152 # we use hashing to create a large n for the shuffle.
153 s = repr(n) + repr(self.getstate())
154 n = int(_hashlib.new('sha512', s).hexdigest(), 16)
155 super(Random, self).jumpahead(n)
157 ## ---- Methods below this point do not need to be overridden when
158 ## ---- subclassing for the purpose of using a different core generator.
160 ## -------------------- pickle support -------------------
162 def __getstate__(self): # for pickle
163 return self.getstate()
165 def __setstate__(self, state): # for pickle
168 def __reduce__(self):
169 return self.__class__, (), self.getstate()
171 ## -------------------- integer methods -------------------
173 def randrange(self, start, stop=None, step=1, int=int, default=None,
175 """Choose a random item from range(start, stop[, step]).
177 This fixes the problem with randint() which includes the
178 endpoint; in Python this is usually not what you want.
179 Do not supply the 'int', 'default', and 'maxwidth' arguments.
182 # This code is a bit messy to make it fast for the
183 # common case while still doing adequate error checking.
186 raise ValueError, "non-integer arg 1 for randrange()"
189 if istart >= maxwidth:
190 return self._randbelow(istart)
191 return int(self.random() * istart)
192 raise ValueError, "empty range for randrange()"
194 # stop argument supplied.
197 raise ValueError, "non-integer stop for randrange()"
198 width = istop - istart
199 if step == 1 and width > 0:
201 # int(istart + self.random()*width)
202 # instead would be incorrect. For example, consider istart
203 # = -2 and istop = 0. Then the guts would be in
204 # -2.0 to 0.0 exclusive on both ends (ignoring that random()
205 # might return 0.0), and because int() truncates toward 0, the
206 # final result would be -1 or 0 (instead of -2 or -1).
207 # istart + int(self.random()*width)
208 # would also be incorrect, for a subtler reason: the RHS
209 # can return a long, and then randrange() would also return
210 # a long, but we're supposed to return an int (for backward
213 if width >= maxwidth:
214 return int(istart + self._randbelow(width))
215 return int(istart + int(self.random()*width))
217 raise ValueError, "empty range for randrange() (%d,%d, %d)" % (istart, istop, width)
219 # Non-unit step argument supplied.
222 raise ValueError, "non-integer step for randrange()"
224 n = (width + istep - 1) // istep
226 n = (width + istep + 1) // istep
228 raise ValueError, "zero step for randrange()"
231 raise ValueError, "empty range for randrange()"
234 return istart + istep*self._randbelow(n)
235 return istart + istep*int(self.random() * n)
237 def randint(self, a, b):
238 """Return random integer in range [a, b], including both end points.
241 return self.randrange(a, b+1)
243 def _randbelow(self, n, _log=_log, int=int, _maxwidth=1L<<BPF,
244 _Method=_MethodType, _BuiltinMethod=_BuiltinMethodType):
245 """Return a random int in the range [0,n)
247 Handles the case where n has more bits than returned
248 by a single call to the underlying generator.
252 getrandbits = self.getrandbits
253 except AttributeError:
256 # Only call self.getrandbits if the original random() builtin method
257 # has not been overridden or if a new getrandbits() was supplied.
258 # This assures that the two methods correspond.
259 if type(self.random) is _BuiltinMethod or type(getrandbits) is _Method:
260 k = int(1.00001 + _log(n-1, 2.0)) # 2**k > n-1 > 2**(k-2)
266 _warn("Underlying random() generator does not supply \n"
267 "enough bits to choose from a population range this large")
268 return int(self.random() * n)
270 ## -------------------- sequence methods -------------------
272 def choice(self, seq):
273 """Choose a random element from a non-empty sequence."""
274 return seq[int(self.random() * len(seq))] # raises IndexError if seq is empty
276 def shuffle(self, x, random=None, int=int):
277 """x, random=random.random -> shuffle list x in place; return None.
279 Optional arg random is a 0-argument function returning a random
280 float in [0.0, 1.0); by default, the standard random.random.
285 for i in reversed(xrange(1, len(x))):
286 # pick an element in x[:i+1] with which to exchange x[i]
287 j = int(random() * (i+1))
288 x[i], x[j] = x[j], x[i]
290 def sample(self, population, k):
291 """Chooses k unique random elements from a population sequence.
293 Returns a new list containing elements from the population while
294 leaving the original population unchanged. The resulting list is
295 in selection order so that all sub-slices will also be valid random
296 samples. This allows raffle winners (the sample) to be partitioned
297 into grand prize and second place winners (the subslices).
299 Members of the population need not be hashable or unique. If the
300 population contains repeats, then each occurrence is a possible
301 selection in the sample.
303 To choose a sample in a range of integers, use xrange as an argument.
304 This is especially fast and space efficient for sampling from a
305 large population: sample(xrange(10000000), 60)
308 # Sampling without replacement entails tracking either potential
309 # selections (the pool) in a list or previous selections in a set.
311 # When the number of selections is small compared to the
312 # population, then tracking selections is efficient, requiring
313 # only a small set and an occasional reselection. For
314 # a larger number of selections, the pool tracking method is
315 # preferred since the list takes less space than the
316 # set and it doesn't suffer from frequent reselections.
320 raise ValueError, "sample larger than population"
324 setsize = 21 # size of a small set minus size of an empty list
326 setsize += 4 ** _ceil(_log(k * 3, 4)) # table size for big sets
327 if n <= setsize or hasattr(population, "keys"):
328 # An n-length list is smaller than a k-length set, or this is a
329 # mapping type so the other algorithm wouldn't work.
330 pool = list(population)
331 for i in xrange(k): # invariant: non-selected at [0,n-i)
332 j = _int(random() * (n-i))
334 pool[j] = pool[n-i-1] # move non-selected item into vacancy
338 selected_add = selected.add
340 j = _int(random() * n)
342 j = _int(random() * n)
344 result[i] = population[j]
345 except (TypeError, KeyError): # handle (at least) sets
346 if isinstance(population, list):
348 return self.sample(tuple(population), k)
351 ## -------------------- real-valued distributions -------------------
353 ## -------------------- uniform distribution -------------------
355 def uniform(self, a, b):
356 "Get a random number in the range [a, b) or [a, b] depending on rounding."
357 return a + (b-a) * self.random()
359 ## -------------------- triangular --------------------
361 def triangular(self, low=0.0, high=1.0, mode=None):
362 """Triangular distribution.
364 Continuous distribution bounded by given lower and upper limits,
365 and having a given mode value in-between.
367 http://en.wikipedia.org/wiki/Triangular_distribution
371 c = 0.5 if mode is None else (mode - low) / (high - low)
375 low, high = high, low
376 return low + (high - low) * (u * c) ** 0.5
378 ## -------------------- normal distribution --------------------
380 def normalvariate(self, mu, sigma):
381 """Normal distribution.
383 mu is the mean, and sigma is the standard deviation.
386 # mu = mean, sigma = standard deviation
388 # Uses Kinderman and Monahan method. Reference: Kinderman,
389 # A.J. and Monahan, J.F., "Computer generation of random
390 # variables using the ratio of uniform deviates", ACM Trans
391 # Math Software, 3, (1977), pp257-260.
397 z = NV_MAGICCONST*(u1-0.5)/u2
403 ## -------------------- lognormal distribution --------------------
405 def lognormvariate(self, mu, sigma):
406 """Log normal distribution.
408 If you take the natural logarithm of this distribution, you'll get a
409 normal distribution with mean mu and standard deviation sigma.
410 mu can have any value, and sigma must be greater than zero.
413 return _exp(self.normalvariate(mu, sigma))
415 ## -------------------- exponential distribution --------------------
417 def expovariate(self, lambd):
418 """Exponential distribution.
420 lambd is 1.0 divided by the desired mean. It should be
421 nonzero. (The parameter would be called "lambda", but that is
422 a reserved word in Python.) Returned values range from 0 to
423 positive infinity if lambd is positive, and from negative
424 infinity to 0 if lambd is negative.
427 # lambd: rate lambd = 1/mean
428 # ('lambda' is a Python reserved word)
434 return -_log(u)/lambd
436 ## -------------------- von Mises distribution --------------------
438 def vonmisesvariate(self, mu, kappa):
439 """Circular data distribution.
441 mu is the mean angle, expressed in radians between 0 and 2*pi, and
442 kappa is the concentration parameter, which must be greater than or
443 equal to zero. If kappa is equal to zero, this distribution reduces
444 to a uniform random angle over the range 0 to 2*pi.
447 # mu: mean angle (in radians between 0 and 2*pi)
448 # kappa: concentration parameter kappa (>= 0)
449 # if kappa = 0 generate uniform random angle
451 # Based upon an algorithm published in: Fisher, N.I.,
452 # "Statistical Analysis of Circular Data", Cambridge
453 # University Press, 1993.
455 # Thanks to Magnus Kessler for a correction to the
456 # implementation of step 4.
460 return TWOPI * random()
462 a = 1.0 + _sqrt(1.0 + 4.0 * kappa * kappa)
463 b = (a - _sqrt(2.0 * a))/(2.0 * kappa)
464 r = (1.0 + b * b)/(2.0 * b)
470 f = (1.0 + r * z)/(r + z)
475 if u2 < c * (2.0 - c) or u2 <= c * _exp(1.0 - c):
480 theta = (mu % TWOPI) + _acos(f)
482 theta = (mu % TWOPI) - _acos(f)
486 ## -------------------- gamma distribution --------------------
488 def gammavariate(self, alpha, beta):
489 """Gamma distribution. Not the gamma function!
491 Conditions on the parameters are alpha > 0 and beta > 0.
495 # alpha > 0, beta > 0, mean is alpha*beta, variance is alpha*beta**2
497 # Warning: a few older sources define the gamma distribution in terms
499 if alpha <= 0.0 or beta <= 0.0:
500 raise ValueError, 'gammavariate: alpha and beta must be > 0.0'
505 # Uses R.C.H. Cheng, "The generation of Gamma
506 # variables with non-integral shape parameters",
507 # Applied Statistics, (1977), 26, No. 1, p71-74
509 ainv = _sqrt(2.0 * alpha - 1.0)
515 if not 1e-7 < u1 < .9999999:
518 v = _log(u1/(1.0-u1))/ainv
522 if r + SG_MAGICCONST - 4.5*z >= 0.0 or r >= _log(z):
530 return -_log(u) * beta
532 else: # alpha is between 0 and 1 (exclusive)
534 # Uses ALGORITHM GS of Statistical Computing - Kennedy & Gentle
543 x = -_log((b-p)/alpha)
546 if u1 <= x ** (alpha - 1.0):
552 ## -------------------- Gauss (faster alternative) --------------------
554 def gauss(self, mu, sigma):
555 """Gaussian distribution.
557 mu is the mean, and sigma is the standard deviation. This is
558 slightly faster than the normalvariate() function.
560 Not thread-safe without a lock around calls.
564 # When x and y are two variables from [0, 1), uniformly
567 # cos(2*pi*x)*sqrt(-2*log(1-y))
568 # sin(2*pi*x)*sqrt(-2*log(1-y))
570 # are two *independent* variables with normal distribution
571 # (mu = 0, sigma = 1).
573 # (corrected version; bug discovered by Mike Miller, fixed by LM)
575 # Multithreading note: When two threads call this function
576 # simultaneously, it is possible that they will receive the
577 # same return value. The window is very small though. To
578 # avoid this, you have to use a lock around all calls. (I
579 # didn't want to slow this down in the serial case by using a
584 self.gauss_next = None
586 x2pi = random() * TWOPI
587 g2rad = _sqrt(-2.0 * _log(1.0 - random()))
588 z = _cos(x2pi) * g2rad
589 self.gauss_next = _sin(x2pi) * g2rad
593 ## -------------------- beta --------------------
595 ## http://sourceforge.net/bugs/?func=detailbug&bug_id=130030&group_id=5470
596 ## for Ivan Frohne's insightful analysis of why the original implementation:
598 ## def betavariate(self, alpha, beta):
599 ## # Discrete Event Simulation in C, pp 87-88.
601 ## y = self.expovariate(alpha)
602 ## z = self.expovariate(1.0/beta)
605 ## was dead wrong, and how it probably got that way.
607 def betavariate(self, alpha, beta):
608 """Beta distribution.
610 Conditions on the parameters are alpha > 0 and beta > 0.
611 Returned values range between 0 and 1.
615 # This version due to Janne Sinkkonen, and matches all the std
616 # texts (e.g., Knuth Vol 2 Ed 3 pg 134 "the beta distribution").
617 y = self.gammavariate(alpha, 1.)
621 return y / (y + self.gammavariate(beta, 1.))
623 ## -------------------- Pareto --------------------
625 def paretovariate(self, alpha):
626 """Pareto distribution. alpha is the shape parameter."""
629 u = 1.0 - self.random()
630 return 1.0 / pow(u, 1.0/alpha)
632 ## -------------------- Weibull --------------------
634 def weibullvariate(self, alpha, beta):
635 """Weibull distribution.
637 alpha is the scale parameter and beta is the shape parameter.
640 # Jain, pg. 499; bug fix courtesy Bill Arms
642 u = 1.0 - self.random()
643 return alpha * pow(-_log(u), 1.0/beta)
645 ## -------------------- Wichmann-Hill -------------------
647 class WichmannHill(Random):
649 VERSION = 1 # used by getstate/setstate
651 def seed(self, a=None):
652 """Initialize internal state from hashable object.
654 None or no argument seeds from current time or from an operating
655 system specific randomness source if available.
657 If a is not None or an int or long, hash(a) is used instead.
659 If a is an int or long, a is used directly. Distinct values between
660 0 and 27814431486575L inclusive are guaranteed to yield distinct
661 internal states (this guarantee is specific to the default
662 Wichmann-Hill generator).
667 a = long(_hexlify(_urandom(16)), 16)
668 except NotImplementedError:
670 a = long(time.time() * 256) # use fractional seconds
672 if not isinstance(a, (int, long)):
675 a, x = divmod(a, 30268)
676 a, y = divmod(a, 30306)
677 a, z = divmod(a, 30322)
678 self._seed = int(x)+1, int(y)+1, int(z)+1
680 self.gauss_next = None
683 """Get the next random number in the range [0.0, 1.0)."""
685 # Wichman-Hill random number generator.
687 # Wichmann, B. A. & Hill, I. D. (1982)
689 # An efficient and portable pseudo-random number generator
690 # Applied Statistics 31 (1982) 188-190
693 # Correction to Algorithm AS 183
694 # Applied Statistics 33 (1984) 123
696 # McLeod, A. I. (1985)
697 # A remark on Algorithm AS 183
698 # Applied Statistics 34 (1985),198-200
700 # This part is thread-unsafe:
701 # BEGIN CRITICAL SECTION
703 x = (171 * x) % 30269
704 y = (172 * y) % 30307
705 z = (170 * z) % 30323
707 # END CRITICAL SECTION
709 # Note: on a platform using IEEE-754 double arithmetic, this can
710 # never return 0.0 (asserted by Tim; proof too long for a comment).
711 return (x/30269.0 + y/30307.0 + z/30323.0) % 1.0
714 """Return internal state; can be passed to setstate() later."""
715 return self.VERSION, self._seed, self.gauss_next
717 def setstate(self, state):
718 """Restore internal state from object returned by getstate()."""
721 version, self._seed, self.gauss_next = state
723 raise ValueError("state with version %s passed to "
724 "Random.setstate() of version %s" %
725 (version, self.VERSION))
727 def jumpahead(self, n):
728 """Act as if n calls to random() were made, but quickly.
730 n is an int, greater than or equal to 0.
732 Example use: If you have 2 threads and know that each will
733 consume no more than a million random numbers, create two Random
734 objects r1 and r2, then do
735 r2.setstate(r1.getstate())
736 r2.jumpahead(1000000)
737 Then r1 and r2 will use guaranteed-disjoint segments of the full
742 raise ValueError("n must be >= 0")
744 x = int(x * pow(171, n, 30269)) % 30269
745 y = int(y * pow(172, n, 30307)) % 30307
746 z = int(z * pow(170, n, 30323)) % 30323
749 def __whseed(self, x=0, y=0, z=0):
750 """Set the Wichmann-Hill seed from (x, y, z).
752 These must be integers in the range [0, 256).
755 if not type(x) == type(y) == type(z) == int:
756 raise TypeError('seeds must be integers')
757 if not (0 <= x < 256 and 0 <= y < 256 and 0 <= z < 256):
758 raise ValueError('seeds must be in range(0, 256)')
760 # Initialize from current time
762 t = long(time.time() * 256)
763 t = int((t&0xffffff) ^ (t>>24))
764 t, x = divmod(t, 256)
765 t, y = divmod(t, 256)
766 t, z = divmod(t, 256)
767 # Zero is a poor seed, so substitute 1
768 self._seed = (x or 1, y or 1, z or 1)
770 self.gauss_next = None
772 def whseed(self, a=None):
773 """Seed from hashable object's hash code.
775 None or no argument seeds from current time. It is not guaranteed
776 that objects with distinct hash codes lead to distinct internal
779 This is obsolete, provided for compatibility with the seed routine
780 used prior to Python 2.1. Use the .seed() method instead.
787 a, x = divmod(a, 256)
788 a, y = divmod(a, 256)
789 a, z = divmod(a, 256)
790 x = (x + a) % 256 or 1
791 y = (y + a) % 256 or 1
792 z = (z + a) % 256 or 1
793 self.__whseed(x, y, z)
795 ## --------------- Operating System Random Source ------------------
797 class SystemRandom(Random):
798 """Alternate random number generator using sources provided
799 by the operating system (such as /dev/urandom on Unix or
800 CryptGenRandom on Windows).
802 Not available on all systems (see os.urandom() for details).
806 """Get the next random number in the range [0.0, 1.0)."""
807 return (long(_hexlify(_urandom(7)), 16) >> 3) * RECIP_BPF
809 def getrandbits(self, k):
810 """getrandbits(k) -> x. Generates a long int with k random bits."""
812 raise ValueError('number of bits must be greater than zero')
814 raise TypeError('number of bits should be an integer')
815 bytes = (k + 7) // 8 # bits / 8 and rounded up
816 x = long(_hexlify(_urandom(bytes)), 16)
817 return x >> (bytes * 8 - k) # trim excess bits
819 def _stub(self, *args, **kwds):
820 "Stub method. Not used for a system random number generator."
822 seed = jumpahead = _stub
824 def _notimplemented(self, *args, **kwds):
825 "Method should not be called for a system random number generator."
826 raise NotImplementedError('System entropy source does not have state.')
827 getstate = setstate = _notimplemented
829 ## -------------------- test program --------------------
831 def _test_generator(n, func, args):
833 print n, 'times', func.__name__
843 smallest = min(x, smallest)
844 largest = max(x, largest)
846 print round(t1-t0, 3), 'sec,',
848 stddev = _sqrt(sqsum/n - avg*avg)
849 print 'avg %g, stddev %g, min %g, max %g' % \
850 (avg, stddev, smallest, largest)
854 _test_generator(N, random, ())
855 _test_generator(N, normalvariate, (0.0, 1.0))
856 _test_generator(N, lognormvariate, (0.0, 1.0))
857 _test_generator(N, vonmisesvariate, (0.0, 1.0))
858 _test_generator(N, gammavariate, (0.01, 1.0))
859 _test_generator(N, gammavariate, (0.1, 1.0))
860 _test_generator(N, gammavariate, (0.1, 2.0))
861 _test_generator(N, gammavariate, (0.5, 1.0))
862 _test_generator(N, gammavariate, (0.9, 1.0))
863 _test_generator(N, gammavariate, (1.0, 1.0))
864 _test_generator(N, gammavariate, (2.0, 1.0))
865 _test_generator(N, gammavariate, (20.0, 1.0))
866 _test_generator(N, gammavariate, (200.0, 1.0))
867 _test_generator(N, gauss, (0.0, 1.0))
868 _test_generator(N, betavariate, (3.0, 3.0))
869 _test_generator(N, triangular, (0.0, 1.0, 1.0/3.0))
871 # Create one instance, seeded from current time, and export its methods
872 # as module-level functions. The functions share state across all uses
873 #(both in the user's code and in the Python libraries), but that's fine
874 # for most programs and is easier for the casual user than making them
875 # instantiate their own Random() instance.
879 random = _inst.random
880 uniform = _inst.uniform
881 triangular = _inst.triangular
882 randint = _inst.randint
883 choice = _inst.choice
884 randrange = _inst.randrange
885 sample = _inst.sample
886 shuffle = _inst.shuffle
887 normalvariate = _inst.normalvariate
888 lognormvariate = _inst.lognormvariate
889 expovariate = _inst.expovariate
890 vonmisesvariate = _inst.vonmisesvariate
891 gammavariate = _inst.gammavariate
893 betavariate = _inst.betavariate
894 paretovariate = _inst.paretovariate
895 weibullvariate = _inst.weibullvariate
896 getstate = _inst.getstate
897 setstate = _inst.setstate
898 jumpahead = _inst.jumpahead
899 getrandbits = _inst.getrandbits
901 if __name__ == '__main__':