diff options
Diffstat (limited to 'Lib/test/test_random.py')
| -rw-r--r-- | Lib/test/test_random.py | 207 |
1 files changed, 207 insertions, 0 deletions
diff --git a/Lib/test/test_random.py b/Lib/test/test_random.py index facddb1..49a3f7b 100644 --- a/Lib/test/test_random.py +++ b/Lib/test/test_random.py @@ -1,10 +1,12 @@ #!/usr/bin/env python3 import unittest +import unittest.mock import random import time import pickle import warnings +from functools import partial from math import log, exp, pi, fsum, sin from test import support @@ -46,6 +48,48 @@ class TestBasicOps: self.assertRaises(TypeError, self.gen.seed, 1, 2, 3, 4) self.assertRaises(TypeError, type(self.gen), []) + @unittest.mock.patch('random._urandom') # os.urandom + def test_seed_when_randomness_source_not_found(self, urandom_mock): + # Random.seed() uses time.time() when an operating system specific + # randomness source is not found. To test this on machines were it + # exists, run the above test, test_seedargs(), again after mocking + # os.urandom() so that it raises the exception expected when the + # randomness source is not available. + urandom_mock.side_effect = NotImplementedError + self.test_seedargs() + + def test_shuffle(self): + shuffle = self.gen.shuffle + lst = [] + shuffle(lst) + self.assertEqual(lst, []) + lst = [37] + shuffle(lst) + self.assertEqual(lst, [37]) + seqs = [list(range(n)) for n in range(10)] + shuffled_seqs = [list(range(n)) for n in range(10)] + for shuffled_seq in shuffled_seqs: + shuffle(shuffled_seq) + for (seq, shuffled_seq) in zip(seqs, shuffled_seqs): + self.assertEqual(len(seq), len(shuffled_seq)) + self.assertEqual(set(seq), set(shuffled_seq)) + # The above tests all would pass if the shuffle was a + # no-op. The following non-deterministic test covers that. It + # asserts that the shuffled sequence of 1000 distinct elements + # must be different from the original one. Although there is + # mathematically a non-zero probability that this could + # actually happen in a genuinely random shuffle, it is + # completely negligible, given that the number of possible + # permutations of 1000 objects is 1000! (factorial of 1000), + # which is considerably larger than the number of atoms in the + # universe... + lst = list(range(1000)) + shuffled_lst = list(range(1000)) + shuffle(shuffled_lst) + self.assertTrue(lst != shuffled_lst) + shuffle(lst) + self.assertTrue(lst != shuffled_lst) + def test_choice(self): choice = self.gen.choice with self.assertRaises(IndexError): @@ -65,6 +109,8 @@ class TestBasicOps: self.assertEqual(len(uniq), k) self.assertTrue(uniq <= set(population)) self.assertEqual(self.gen.sample([], 0), []) # test edge case N==k==0 + # Exception raised if size of sample exceeds that of population + self.assertRaises(ValueError, self.gen.sample, population, N+1) def test_sample_distribution(self): # For the entire allowable range of 0 <= k <= N, validate that @@ -205,6 +251,25 @@ class SystemRandom_TestBasicOps(TestBasicOps, unittest.TestCase): self.assertEqual(set(range(start,stop)), set([self.gen.randrange(start,stop) for i in range(100)])) + def test_randrange_nonunit_step(self): + rint = self.gen.randrange(0, 10, 2) + self.assertIn(rint, (0, 2, 4, 6, 8)) + rint = self.gen.randrange(0, 2, 2) + self.assertEqual(rint, 0) + + def test_randrange_errors(self): + raises = partial(self.assertRaises, ValueError, self.gen.randrange) + # Empty range + raises(3, 3) + raises(-721) + raises(0, 100, -12) + # Non-integer start/stop + raises(3.14159) + raises(0, 2.71828) + # Zero and non-integer step + raises(0, 42, 0) + raises(0, 42, 3.14159) + def test_genrandbits(self): # Verify ranges for k in range(1, 1000): @@ -274,6 +339,16 @@ class MersenneTwister_TestBasicOps(TestBasicOps, unittest.TestCase): # Last element s/b an int also self.assertRaises(TypeError, self.gen.setstate, (2, (0,)*624+('a',), None)) + # Little trick to make "tuple(x % (2**32) for x in internalstate)" + # raise ValueError. I cannot think of a simple way to achieve this, so + # I am opting for using a generator as the middle argument of setstate + # which attempts to cast a NaN to integer. + state_values = self.gen.getstate()[1] + state_values = list(state_values) + state_values[-1] = float('nan') + state = (int(x) for x in state_values) + self.assertRaises(TypeError, self.gen.setstate, (2, state, None)) + def test_referenceImplementation(self): # Compare the python implementation with results from the original # code. Create 2000 53-bit precision random floats. Compare only @@ -413,6 +488,38 @@ class MersenneTwister_TestBasicOps(TestBasicOps, unittest.TestCase): self.assertEqual(k, numbits) # note the stronger assertion self.assertTrue(2**k > n > 2**(k-1)) # note the stronger assertion + @unittest.mock.patch('random.Random.random') + def test_randbelow_overriden_random(self, random_mock): + # Random._randbelow() can only use random() when the built-in one + # has been overridden but no new getrandbits() method was supplied. + random_mock.side_effect = random.SystemRandom().random + maxsize = 1<<random.BPF + with warnings.catch_warnings(): + warnings.simplefilter("ignore", UserWarning) + # Population range too large (n >= maxsize) + self.gen._randbelow(maxsize+1, maxsize = maxsize) + self.gen._randbelow(5640, maxsize = maxsize) + + # This might be going too far to test a single line, but because of our + # noble aim of achieving 100% test coverage we need to write a case in + # which the following line in Random._randbelow() gets executed: + # + # rem = maxsize % n + # limit = (maxsize - rem) / maxsize + # r = random() + # while r >= limit: + # r = random() # <== *This line* <==< + # + # Therefore, to guarantee that the while loop is executed at least + # once, we need to mock random() so that it returns a number greater + # than 'limit' the first time it gets called. + + n = 42 + epsilon = 0.01 + limit = (maxsize - (maxsize % n)) / maxsize + random_mock.side_effect = [limit + epsilon, limit - epsilon] + self.gen._randbelow(n, maxsize = maxsize) + def test_randrange_bug_1590891(self): start = 1000000000000 stop = -100000000000000000000 @@ -530,6 +637,106 @@ class TestDistributions(unittest.TestCase): random.vonmisesvariate(0, 1e15) random.vonmisesvariate(0, 1e100) + def test_gammavariate_errors(self): + # Both alpha and beta must be > 0.0 + self.assertRaises(ValueError, random.gammavariate, -1, 3) + self.assertRaises(ValueError, random.gammavariate, 0, 2) + self.assertRaises(ValueError, random.gammavariate, 2, 0) + self.assertRaises(ValueError, random.gammavariate, 1, -3) + + @unittest.mock.patch('random.Random.random') + def test_gammavariate_full_code_coverage(self, random_mock): + # There are three different possibilities in the current implementation + # of random.gammavariate(), depending on the value of 'alpha'. What we + # are going to do here is to fix the values returned by random() to + # generate test cases that provide 100% line coverage of the method. + + # #1: alpha > 1.0: we want the first random number to be outside the + # [1e-7, .9999999] range, so that the continue statement executes + # once. The values of u1 and u2 will be 0.5 and 0.3, respectively. + random_mock.side_effect = [1e-8, 0.5, 0.3] + returned_value = random.gammavariate(1.1, 2.3) + self.assertAlmostEqual(returned_value, 2.53) + + # #2: alpha == 1: first random number less than 1e-7 to that the body + # of the while loop executes once. Then random.random() returns 0.45, + # which causes while to stop looping and the algorithm to terminate. + random_mock.side_effect = [1e-8, 0.45] + returned_value = random.gammavariate(1.0, 3.14) + self.assertAlmostEqual(returned_value, 2.507314166123803) + + # #3: 0 < alpha < 1. This is the most complex region of code to cover, + # as there are multiple if-else statements. Let's take a look at the + # source code, and determine the values that we need accordingly: + # + # while 1: + # u = random() + # b = (_e + alpha)/_e + # p = b*u + # if p <= 1.0: # <=== (A) + # x = p ** (1.0/alpha) + # else: # <=== (B) + # x = -_log((b-p)/alpha) + # u1 = random() + # if p > 1.0: # <=== (C) + # if u1 <= x ** (alpha - 1.0): # <=== (D) + # break + # elif u1 <= _exp(-x): # <=== (E) + # break + # return x * beta + # + # First, we want (A) to be True. For that we need that: + # b*random() <= 1.0 + # r1 = random() <= 1.0 / b + # + # We now get to the second if-else branch, and here, since p <= 1.0, + # (C) is False and we take the elif branch, (E). For it to be True, + # so that the break is executed, we need that: + # r2 = random() <= _exp(-x) + # r2 <= _exp(-(p ** (1.0/alpha))) + # r2 <= _exp(-((b*r1) ** (1.0/alpha))) + + _e = random._e + _exp = random._exp + _log = random._log + alpha = 0.35 + beta = 1.45 + b = (_e + alpha)/_e + epsilon = 0.01 + + r1 = 0.8859296441566 # 1.0 / b + r2 = 0.3678794411714 # _exp(-((b*r1) ** (1.0/alpha))) + + # These four "random" values result in the following trace: + # (A) True, (E) False --> [next iteration of while] + # (A) True, (E) True --> [while loop breaks] + random_mock.side_effect = [r1, r2 + epsilon, r1, r2] + returned_value = random.gammavariate(alpha, beta) + self.assertAlmostEqual(returned_value, 1.4499999999997544) + + # Let's now make (A) be False. If this is the case, when we get to the + # second if-else 'p' is greater than 1, so (C) evaluates to True. We + # now encounter a second if statement, (D), which in order to execute + # must satisfy the following condition: + # r2 <= x ** (alpha - 1.0) + # r2 <= (-_log((b-p)/alpha)) ** (alpha - 1.0) + # r2 <= (-_log((b-(b*r1))/alpha)) ** (alpha - 1.0) + r1 = 0.8959296441566 # (1.0 / b) + epsilon -- so that (A) is False + r2 = 0.9445400408898141 + + # And these four values result in the following trace: + # (B) and (C) True, (D) False --> [next iteration of while] + # (B) and (C) True, (D) True [while loop breaks] + random_mock.side_effect = [r1, r2 + epsilon, r1, r2] + returned_value = random.gammavariate(alpha, beta) + self.assertAlmostEqual(returned_value, 1.5830349561760781) + + @unittest.mock.patch('random.Random.gammavariate') + def test_betavariate_return_zero(self, gammavariate_mock): + # betavariate() returns zero when the Gamma distribution + # that it uses internally returns this same value. + gammavariate_mock.return_value = 0.0 + self.assertEqual(0.0, random.betavariate(2.71828, 3.14159)) class TestModule(unittest.TestCase): def testMagicConstants(self): |