1
|
|
|
# Copyright (c) 2014, Salesforce.com, Inc. All rights reserved. |
2
|
|
|
# |
3
|
|
|
# Redistribution and use in source and binary forms, with or without |
4
|
|
|
# modification, are permitted provided that the following conditions |
5
|
|
|
# are met: |
6
|
|
|
# |
7
|
|
|
# - Redistributions of source code must retain the above copyright |
8
|
|
|
# notice, this list of conditions and the following disclaimer. |
9
|
|
|
# - Redistributions in binary form must reproduce the above copyright |
10
|
|
|
# notice, this list of conditions and the following disclaimer in the |
11
|
|
|
# documentation and/or other materials provided with the distribution. |
12
|
|
|
# - Neither the name of Salesforce.com nor the names of its contributors |
13
|
|
|
# may be used to endorse or promote products derived from this |
14
|
|
|
# software without specific prior written permission. |
15
|
|
|
# |
16
|
|
|
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
17
|
|
|
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
18
|
|
|
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
19
|
|
|
# FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
20
|
|
|
# COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
21
|
|
|
# INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
22
|
|
|
# BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS |
23
|
|
|
# OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
24
|
|
|
# ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR |
25
|
|
|
# TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE |
26
|
|
|
# USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
27
|
|
|
|
28
|
|
|
import numpy as np |
29
|
|
|
from scipy.special import betaln, binom |
30
|
|
|
from matplotlib import pyplot |
31
|
|
|
|
32
|
|
|
|
33
|
|
|
def binomln(n, k): |
34
|
|
|
return np.log(binom(n, k)) |
35
|
|
|
|
36
|
|
|
|
37
|
|
|
def simone(alpha, beta, n, ITERS): |
38
|
|
|
ks = [] |
39
|
|
|
for i in range(ITERS): |
40
|
|
|
theta = np.random.beta(alpha, beta) |
41
|
|
|
ks.append(np.random.binomial(n, theta)) |
42
|
|
|
return np.array(ks) |
43
|
|
|
|
44
|
|
|
|
45
|
|
|
def simtwo(alpha, beta, n, ITERS): |
46
|
|
|
counts = np.zeros((ITERS, 2), dtype=np.uint32) |
47
|
|
|
thetas = np.random.beta(alpha, beta, size=ITERS) |
48
|
|
|
for i in range(ITERS): |
49
|
|
|
if i % 1000000 == 0: |
50
|
|
|
print i |
51
|
|
|
|
52
|
|
|
counts[i] = np.random.binomial(n, thetas[i], size=2) |
53
|
|
|
return counts |
54
|
|
|
|
55
|
|
|
|
56
|
|
|
class Hypers(object): |
57
|
|
|
|
58
|
|
|
def __init__(self): |
59
|
|
|
self.alpha = 0.0 |
60
|
|
|
self.beta = 0.0 |
61
|
|
|
self.N = 0 |
62
|
|
|
|
63
|
|
|
def __str__(self): |
64
|
|
|
return "alpha=%f, beta=%f, N=%d" % (self.alpha, |
65
|
|
|
self.beta, |
66
|
|
|
self.N) |
67
|
|
|
|
68
|
|
|
|
69
|
|
|
class SuffStats(object): |
70
|
|
|
|
71
|
|
|
def __init__(self): |
72
|
|
|
self.heads = 0 |
73
|
|
|
self.tails = 0 |
74
|
|
|
self.binomln_accum = 0 |
75
|
|
|
self.dpcount = 0 |
76
|
|
|
|
77
|
|
|
def __str__(self): |
78
|
|
|
return "heads=%d, tails=%d" % (self.heads, self.tails) |
79
|
|
|
|
80
|
|
|
|
81
|
|
|
def add_dp(k, hps, suffs): |
82
|
|
|
suffs.heads += k |
83
|
|
|
suffs.tails += hps.N - k |
84
|
|
|
suffs.binomln_accum += binomln(hps.N, k) |
85
|
|
|
|
86
|
|
|
suffs.dpcount += 1 |
87
|
|
|
|
88
|
|
|
|
89
|
|
|
def compute_post_pred(k, hps, suffs): |
90
|
|
|
|
91
|
|
|
a = hps.alpha + suffs.heads |
92
|
|
|
b = hps.beta + suffs.tails |
93
|
|
|
|
94
|
|
|
return binomln(hps.N, k) + betaln(a + k, b + hps.N - k) - betaln(a, b) |
95
|
|
|
|
96
|
|
|
|
97
|
|
|
def compute_total_likelihood(hps, suffs): |
98
|
|
|
|
99
|
|
|
a = hps.alpha + suffs.heads |
100
|
|
|
b = hps.beta + suffs.tails |
101
|
|
|
|
102
|
|
|
r = betaln(a, b) - betaln(hps.alpha, hps.beta) |
103
|
|
|
|
104
|
|
|
r += + suffs.binomln_accum |
105
|
|
|
|
106
|
|
|
return r |
107
|
|
|
|
108
|
|
|
|
109
|
|
|
def two_d_hist(alpha, beta, N, ITERS): |
110
|
|
|
|
111
|
|
|
r = simtwo(alpha, beta, N, ITERS) |
112
|
|
|
h, b1, b2 = np.histogram2d(r[:, 0], r[:, 1], bins=N + 1) |
113
|
|
|
h[:, :] += 1 |
114
|
|
|
pyplot.imshow(h) |
115
|
|
|
pyplot.show() |
116
|
|
|
|
117
|
|
|
logp = np.log(h.astype(float) / (ITERS)) |
118
|
|
|
|
119
|
|
|
return logp |
120
|
|
|
|
121
|
|
|
|
122
|
|
|
def test_post_pred_equal_likelihood(): |
123
|
|
|
hps = Hypers() |
124
|
|
|
hps.N = 10 |
125
|
|
|
|
126
|
|
|
for alpha in [5.0]: |
127
|
|
|
for beta in [5.0]: |
128
|
|
|
for ks in [[9, 1], [1, 9]]: |
129
|
|
|
print "-" * 60 |
130
|
|
|
hps.alpha = alpha |
131
|
|
|
hps.beta = beta |
132
|
|
|
|
133
|
|
|
ss = SuffStats() |
134
|
|
|
|
135
|
|
|
pred_score = 0.0 |
136
|
|
|
for ki, k in enumerate(ks): |
137
|
|
|
pred_score += compute_post_pred(k, hps, ss) |
138
|
|
|
add_dp(k, hps, ss) |
139
|
|
|
|
140
|
|
|
total_score = compute_total_likelihood(hps, ss) |
141
|
|
|
res = two_d_hist(alpha, beta, hps.N, 10000000) |
142
|
|
|
print res |
143
|
|
|
print res.shape |
144
|
|
|
emp_score = res[ks[0], ks[1]] |
145
|
|
|
d1 = abs(pred_score - emp_score) |
146
|
|
|
d2 = abs(emp_score - total_score) |
147
|
|
|
d3 = abs(total_score - pred_score) |
148
|
|
|
print "alpha=", alpha, "beta=", beta, "ks =", ks |
149
|
|
|
print pred_score, total_score, emp_score |
150
|
|
|
print "error=", d1 + d2 + d3 |
151
|
|
|
|