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 "mp.h"
#include <cassert>
#include <cmath>
#include <algorithm>
static
void knn(const arma::mat &dmat, arma::uword i, arma::uword k, arma::uvec &nn, arma::vec &dist)
{
arma::uword n = dist.n_rows;
if (k > n) {
return;
}
for (arma::uword j = 0, l = 0; l < k; j++, l++) {
if (j == i) {
j++;
}
nn[l] = j;
dist[l] = dmat(i, j);
}
double dmax = *std::max_element(dist.begin(), dist.end());
for (arma::uword j = 0; j < n; j++) {
if (j == i) {
continue;
}
if (dmat(i, j) < dmax) {
dmax = dmat(i, j);
arma::uword l;
for (l = 0; dmat(i, j) > dist[l] && l < k; l++);
for (arma::uword m = l + 1; m < k; m++) {
nn[m] = nn[m - 1];
dist[m] = dist[m - 1];
}
nn[l] = j;
dist[l] = dmat(i, j);
}
}
}
arma::vec mp::neighborhoodPreservation(const arma::mat &distA,
const arma::mat &distB,
arma::uword k)
{
arma::uword n = distA.n_rows;
arma::vec np(n);
#pragma omp parallel for shared(np, n)
for (arma::uword i = 0; i < n; i++) {
arma::uvec nnA(k);
arma::uvec nnB(k);
arma::vec dist(k);
knn(distA, i, k, nnA, dist);
knn(distB, i, k, nnB, dist);
std::sort(nnA.begin(), nnA.end());
std::sort(nnB.begin(), nnB.end());
arma::uword l;
for (l = 0; nnA[l] == nnB[l] && l < k; l++);
np[i] = ((double) l) / k;
}
return np;
}
arma::vec mp::silhouette(const arma::mat &distA,
const arma::mat &distB,
const arma::vec &labels)
{
// TODO
return arma::vec(distA.n_rows, arma::fill::zeros);
}
/*
double mp::stress(const arma::mat &Dp, const arma::mat &Dq)
{
assert(Dp.n_rows == Dp.n_cols);
assert(Dq.n_rows == Dq.n_cols);
assert(Dp.n_rows == Dq.n_rows);
arma::uword n = Dp.n_rows;
double sigma = 0, s = 0;
for (arma::uword i = 0; i < n; i++)
for (arma::uword j = i + 1; j < n; j++) {
double delta = Dp(i, j);
double d = Dq(i, j);
sigma += (delta - d) * (delta - d) / delta;
s += delta;
}
return sigma / s;
}
arma::vec mp::klDivergence(const arma::mat &P, const arma::mat &Q)
{
arma::vec diver(P.n_rows);
mp::klDivergence(P, Q, diver);
return diver;
}
void mp::klDivergence(const arma::mat &P, const arma::mat &Q, arma::vec &diver)
{
assert(P.n_rows == P.n_cols);
assert(Q.n_rows == Q.n_cols);
assert(P.n_rows == Q.n_cols);
assert(diver.n_elem == P.n_rows);
arma::uword n = P.n_rows;
for (arma::uword i = 0; i < n; i++)
diver(i) = klDivergence(P.row(i), Q.row(i));
}
double mp::klDivergence(const arma::rowvec &pi, const arma::rowvec &qi)
{
// Pii and Qii should both be 1, zeroing the i-th term in the sum below
return arma::accu(pi % arma::log(pi / qi));
}
arma::mat mp::d2p(const arma::mat &D, const arma::vec &sigmas)
{
arma::mat P(D.n_rows, D.n_cols);
mp::d2p(D, sigmas, P);
return P;
}
void mp::d2p(const arma::mat &D, const arma::vec &sigmas, arma::mat &P)
{
// WARNING: assumes D and sigmas are already squared
assert(D.n_rows == D.n_cols);
assert(P.n_rows == P.n_cols);
assert(D.n_rows == P.n_rows);
arma::uword n = D.n_rows;
for (arma::uword i = 0; i < n; i++) {
double den = -1; // k == i must be skipped
for (arma::uword k = 0; k < n; k++)
den += exp(-D(i, k) / sigmas(i));
for (arma::uword j = 0; j < n; j++)
P(i, j) = exp(-D(i, j) / sigmas(i)) / den;
}
}
*/
|
