-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathmix.cpp
More file actions
149 lines (130 loc) · 4.86 KB
/
mix.cpp
File metadata and controls
149 lines (130 loc) · 4.86 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
#include "PartialModel.hpp"
#include "aux.hpp"
#include "odds.hpp"
using namespace std;
namespace PoissonFactorization {
namespace Partial {
template <>
void Model<Variable::Mix, Kind::HierGamma>::initialize_factor(size_t t) {
// randomly initialize p of Θ
LOG(debug) << "Initializing P of Θ";
if (true) // TODO make this CLI-switchable
prior.p[t] = prob_to_neg_odds(
sample_beta<Float>(parameters.hyperparameters.theta_p_1,
parameters.hyperparameters.theta_p_2));
else
prior.p[t] = 1;
// initialize r of Θ
LOG(debug) << "Initializing R of Θ";
// NOTE: std::gamma_distribution takes a shape and scale parameter
prior.r[t] = std::gamma_distribution<Float>(
parameters.hyperparameters.theta_r_1,
1 / parameters.hyperparameters.theta_r_2)(EntropySource::rng);
// initialize Θ
LOG(debug) << "Initializing Θ";
#pragma omp parallel for if (DO_PARALLEL)
for (size_t s = 0; s < dim1; ++s)
// NOTE: std::gamma_distribution takes a shape and scale parameter
matrix(s, t) = std::gamma_distribution<Float>(
prior.r(t), 1 / prior.p(t))(EntropySource::rng);
}
template <>
void Model<Variable::Mix, Kind::Dirichlet>::initialize_factor(size_t t) {
assert(false);
throw(std::runtime_error("Not implemented!"));
// TODO implement
std::vector<double> a(dim1);
for (size_t s = 0; s < dim1; ++s)
a[s] = prior.alpha[s];
auto x = sample_dirichlet<Float>(begin(a), end(a),
EntropySource::rngs[omp_get_thread_num()]);
}
template <>
void Model<Variable::Mix, Kind::HierGamma>::initialize() {
// initialize Θ
LOG(debug) << "Initializing Θ from Gamma distribution";
#pragma omp parallel for if (DO_PARALLEL)
for (size_t s = 0; s < dim1; ++s) {
const size_t thread_num = omp_get_thread_num();
for (size_t t = 0; t < dim2; ++t)
// NOTE: std::gamma_distribution takes a shape and scale parameter
matrix(s, t) = std::gamma_distribution<Float>(
prior.r(t), 1 / prior.p(t))(EntropySource::rngs[thread_num]);
}
}
template <>
void Model<Variable::Mix, Kind::Dirichlet>::initialize() {
LOG(debug) << "Initializing Θ from Dirichlet distribution" << std::endl;
#pragma omp parallel for if (DO_PARALLEL)
for (size_t s = 0; s < dim1; ++s) {
const size_t thread_num = omp_get_thread_num();
auto x = sample_dirichlet<Float>(begin(prior.alpha), end(prior.alpha),
EntropySource::rngs[thread_num]);
for (size_t t = 0; t < dim2; ++t)
matrix(s, t) = x[t];
}
}
template <>
// TODO ensure no NaNs or infinities are generated
double Model<Variable::Mix, Kind::HierGamma>::log_likelihood() const {
double l = 0;
for (size_t t = 0; t < dim2; ++t)
l += log_likelihood_factor(t);
return l;
}
template <>
// TODO ensure no NaNs or infinities are generated
double Model<Variable::Mix, Kind::HierGamma>::log_likelihood_factor(
size_t t) const {
double l = 0;
#pragma omp parallel for reduction(+ : l) if (DO_PARALLEL)
for (size_t s = 0; s < dim1; ++s) {
// NOTE: log_gamma takes a shape and scale parameter
auto cur = log_gamma(matrix(s, t), prior.r(t), 1.0 / prior.p(t));
if (false and cur > 0)
LOG(debug) << "ll_cur > 0 for (s,t) = (" << s << ", " << t << "): " << cur
<< " theta = " << matrix(s, t) << " r = " << prior.r(t)
<< " p = " << prior.p(t) << " (r - 1) * log(theta) = "
<< ((prior.r(t) - 1) * log(matrix(s, t)))
<< " - theta / 1/p = " << (-matrix(s, t) / 1 / prior.p(t))
<< " - lgamma(r) = " << (-lgamma(prior.r(t)))
<< " - r * log(1/p) = " << (-prior.r(t) * log(1 / prior.p(t)));
l += cur;
}
if (parameters.respect_theta_prior_likelihood)
// NOTE: log_gamma takes a shape and scale parameter
l += log_gamma(prior.r(t), parameters.hyperparameters.theta_r_1,
1.0 / parameters.hyperparameters.theta_r_2);
if (parameters.respect_theta_prior_likelihood)
l += log_beta_neg_odds(prior.p(t), parameters.hyperparameters.theta_p_1,
parameters.hyperparameters.theta_p_2);
return l;
}
template <>
// TODO ensure no NaNs or infinities are generated
double Model<Variable::Mix, Kind::Dirichlet>::log_likelihood() const {
double l = 0;
#pragma omp parallel for reduction(+ : l) if (DO_PARALLEL)
for (size_t s = 0; s < dim1; ++s) {
vector<double> p(dim2);
for (size_t t = 0; t < dim2; ++t)
p[t] = matrix(s, t);
vector<double> a(dim2, prior.alpha_prior);
l += log_dirichlet(p, a);
}
return l;
}
template <>
// TODO ensure no NaNs or infinities are generated
double Model<Variable::Mix, Kind::Dirichlet>::log_likelihood_factor(
size_t t) const {
// TODO
assert(false);
std::vector<Float> p(dim1);
#pragma omp parallel for if (DO_PARALLEL)
for (size_t s = 0; s < dim1; ++s)
p[s] = matrix(s, t);
return log_dirichlet(p, prior.alpha);
}
}
}