1 | """ |
---|
2 | Tests for allmydata.util.statistics. |
---|
3 | |
---|
4 | Ported to Python 3. |
---|
5 | """ |
---|
6 | |
---|
7 | from io import StringIO |
---|
8 | |
---|
9 | from twisted.trial import unittest |
---|
10 | |
---|
11 | from allmydata.util import statistics |
---|
12 | |
---|
13 | |
---|
14 | class Statistics(unittest.TestCase): |
---|
15 | def should_assert(self, msg, func, *args, **kwargs): |
---|
16 | try: |
---|
17 | func(*args, **kwargs) |
---|
18 | self.fail(msg) |
---|
19 | except AssertionError: |
---|
20 | pass |
---|
21 | |
---|
22 | def failUnlessListEqual(self, a, b, msg = None): |
---|
23 | self.failUnlessEqual(len(a), len(b)) |
---|
24 | for i in range(len(a)): |
---|
25 | self.failUnlessEqual(a[i], b[i], msg) |
---|
26 | |
---|
27 | def failUnlessListAlmostEqual(self, a, b, places = 7, msg = None): |
---|
28 | self.failUnlessEqual(len(a), len(b)) |
---|
29 | for i in range(len(a)): |
---|
30 | self.failUnlessAlmostEqual(a[i], b[i], places, msg) |
---|
31 | |
---|
32 | def test_binomial_coeff(self): |
---|
33 | f = statistics.binomial_coeff |
---|
34 | self.failUnlessEqual(f(20, 0), 1) |
---|
35 | self.failUnlessEqual(f(20, 1), 20) |
---|
36 | self.failUnlessEqual(f(20, 2), 190) |
---|
37 | self.failUnlessEqual(f(20, 8), f(20, 12)) |
---|
38 | self.should_assert("Should assert if n < k", f, 2, 3) |
---|
39 | self.assertEqual(f(5, 3), f(5, 2)) |
---|
40 | |
---|
41 | def test_binomial_distribution_pmf(self): |
---|
42 | f = statistics.binomial_distribution_pmf |
---|
43 | |
---|
44 | pmf_comp = f(2, .1) |
---|
45 | pmf_stat = [0.81, 0.18, 0.01] |
---|
46 | self.failUnlessListAlmostEqual(pmf_comp, pmf_stat) |
---|
47 | |
---|
48 | # Summing across a PMF should give the total probability 1 |
---|
49 | self.failUnlessAlmostEqual(sum(pmf_comp), 1) |
---|
50 | self.should_assert("Should assert if not 0<=p<=1", f, 1, -1) |
---|
51 | self.should_assert("Should assert if n < 1", f, 0, .1) |
---|
52 | |
---|
53 | out = StringIO() |
---|
54 | statistics.print_pmf(pmf_comp, out=out) |
---|
55 | lines = out.getvalue().splitlines() |
---|
56 | self.failUnlessEqual(lines[0], "i=0: 0.81") |
---|
57 | self.failUnlessEqual(lines[1], "i=1: 0.18") |
---|
58 | self.failUnlessEqual(lines[2], "i=2: 0.01") |
---|
59 | |
---|
60 | def test_survival_pmf(self): |
---|
61 | f = statistics.survival_pmf |
---|
62 | # Cross-check binomial-distribution method against convolution |
---|
63 | # method. |
---|
64 | p_list = [.9999] * 100 + [.99] * 50 + [.8] * 20 |
---|
65 | pmf1 = statistics.survival_pmf_via_conv(p_list) |
---|
66 | pmf2 = statistics.survival_pmf_via_bd(p_list) |
---|
67 | self.failUnlessListAlmostEqual(pmf1, pmf2) |
---|
68 | self.failUnlessTrue(statistics.valid_pmf(pmf1)) |
---|
69 | self.should_assert("Should assert if p_i > 1", f, [1.1]); |
---|
70 | self.should_assert("Should assert if p_i < 0", f, [-.1]); |
---|
71 | |
---|
72 | def test_repair_count_pmf(self): |
---|
73 | survival_pmf = statistics.binomial_distribution_pmf(5, .9) |
---|
74 | repair_pmf = statistics.repair_count_pmf(survival_pmf, 3) |
---|
75 | # repair_pmf[0] == sum(survival_pmf[0,1,2,5]) |
---|
76 | # repair_pmf[1] == survival_pmf[4] |
---|
77 | # repair_pmf[2] = survival_pmf[3] |
---|
78 | self.failUnlessListAlmostEqual(repair_pmf, |
---|
79 | [0.00001 + 0.00045 + 0.0081 + 0.59049, |
---|
80 | .32805, |
---|
81 | .0729, |
---|
82 | 0, 0, 0]) |
---|
83 | |
---|
84 | def test_repair_cost(self): |
---|
85 | survival_pmf = statistics.binomial_distribution_pmf(5, .9) |
---|
86 | bwcost = statistics.bandwidth_cost_function |
---|
87 | cost = statistics.mean_repair_cost(bwcost, 1000, |
---|
88 | survival_pmf, 3, ul_dl_ratio=1.0) |
---|
89 | self.failUnlessAlmostEqual(cost, 558.90) |
---|
90 | cost = statistics.mean_repair_cost(bwcost, 1000, |
---|
91 | survival_pmf, 3, ul_dl_ratio=8.0) |
---|
92 | self.failUnlessAlmostEqual(cost, 1664.55) |
---|
93 | |
---|
94 | # I haven't manually checked the math beyond here -warner |
---|
95 | cost = statistics.eternal_repair_cost(bwcost, 1000, |
---|
96 | survival_pmf, 3, |
---|
97 | discount_rate=0, ul_dl_ratio=1.0) |
---|
98 | self.failUnlessAlmostEqual(cost, 65292.056074766246) |
---|
99 | cost = statistics.eternal_repair_cost(bwcost, 1000, |
---|
100 | survival_pmf, 3, |
---|
101 | discount_rate=0.05, |
---|
102 | ul_dl_ratio=1.0) |
---|
103 | self.failUnlessAlmostEqual(cost, 9133.6097158191551) |
---|
104 | |
---|
105 | def test_convolve(self): |
---|
106 | f = statistics.convolve |
---|
107 | v1 = [ 1, 2, 3 ] |
---|
108 | v2 = [ 4, 5, 6 ] |
---|
109 | v3 = [ 7, 8 ] |
---|
110 | v1v2result = [ 4, 13, 28, 27, 18 ] |
---|
111 | # Convolution is commutative |
---|
112 | r1 = f(v1, v2) |
---|
113 | r2 = f(v2, v1) |
---|
114 | self.failUnlessListEqual(r1, r2, "Convolution should be commutative") |
---|
115 | self.failUnlessListEqual(r1, v1v2result, "Didn't match known result") |
---|
116 | # Convolution is associative |
---|
117 | r1 = f(f(v1, v2), v3) |
---|
118 | r2 = f(v1, f(v2, v3)) |
---|
119 | self.failUnlessListEqual(r1, r2, "Convolution should be associative") |
---|
120 | # Convolution is distributive |
---|
121 | r1 = f(v3, [ a + b for a, b in zip(v1, v2) ]) |
---|
122 | tmp1 = f(v3, v1) |
---|
123 | tmp2 = f(v3, v2) |
---|
124 | r2 = [ a + b for a, b in zip(tmp1, tmp2) ] |
---|
125 | self.failUnlessListEqual(r1, r2, "Convolution should be distributive") |
---|
126 | # Convolution is scalar multiplication associative |
---|
127 | tmp1 = f(v1, v2) |
---|
128 | r1 = [ a * 4 for a in tmp1 ] |
---|
129 | tmp2 = [ a * 4 for a in v1 ] |
---|
130 | r2 = f(tmp2, v2) |
---|
131 | self.failUnlessListEqual(r1, r2, "Convolution should be scalar multiplication associative") |
---|
132 | |
---|
133 | def test_find_k(self): |
---|
134 | f = statistics.find_k |
---|
135 | g = statistics.pr_file_loss |
---|
136 | plist = [.9] * 10 + [.8] * 10 # N=20 |
---|
137 | t = .0001 |
---|
138 | k = f(plist, t) |
---|
139 | self.failUnlessEqual(k, 10) |
---|
140 | self.failUnless(g(plist, k) < t) |
---|
141 | |
---|
142 | def test_pr_file_loss(self): |
---|
143 | f = statistics.pr_file_loss |
---|
144 | plist = [.5] * 10 |
---|
145 | self.failUnlessEqual(f(plist, 3), .0546875) |
---|
146 | |
---|
147 | def test_pr_backup_file_loss(self): |
---|
148 | f = statistics.pr_backup_file_loss |
---|
149 | plist = [.5] * 10 |
---|
150 | self.failUnlessEqual(f(plist, .5, 3), .02734375) |
---|