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
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
//! Affinity Propagation (Frey-Dueck 2007).
//!
//! Iteratively passes "responsibility" and "availability" messages over
//! a similarity graph until a stable set of exemplars emerges.
use ndarray::ArrayView2;
use solow_core::{Error, Result};
/// Fitted AffinityPropagation.
#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
#[derive(Clone, Debug, PartialEq)]
pub struct AffinityPropagation {
/// Cluster labels (one per input row).
pub labels: Vec<i64>,
/// Exemplar indices for each cluster.
pub cluster_centers_indices: Vec<usize>,
/// Iterations run.
pub n_iter: usize,
/// Whether the fixed point was reached before `max_iter`.
pub converged: bool,
/// Damping factor used.
pub damping: f64,
}
impl AffinityPropagation {
/// Fit with the sklearn defaults `damping = 0.5`, `max_iter = 200`,
/// `convergence_iter = 15`, and preference = median similarity.
pub fn fit(x: ArrayView2<'_, f64>) -> Result<Self> {
Self::fit_with(x, 0.5, 200, 15, None)
}
/// Full-configuration fit.
pub fn fit_with(
x: ArrayView2<'_, f64>,
damping: f64,
max_iter: usize,
convergence_iter: usize,
preference: Option<f64>,
) -> Result<Self> {
let n = x.nrows();
if n == 0 {
return Err(Error::Value("AffinityPropagation: empty input".into()));
}
if !(0.5..1.0).contains(&damping) {
return Err(Error::Value(format!(
"AffinityPropagation: damping must be in [0.5, 1) (got {damping})"
)));
}
let d = x.ncols();
// Similarity = -‖xᵢ − xⱼ‖² (negative squared Euclidean, sklearn default).
let mut s = vec![vec![0.0_f64; n]; n];
for i in 0..n {
for j in i..n {
let mut acc = 0.0_f64;
for k in 0..d {
let e = x[[i, k]] - x[[j, k]];
acc += e * e;
}
s[i][j] = -acc;
s[j][i] = -acc;
}
}
let pref = preference.unwrap_or_else(|| {
let mut off: Vec<f64> = Vec::new();
for i in 0..n {
for j in (i + 1)..n {
off.push(s[i][j]);
}
}
off.sort_by(|a, b| a.partial_cmp(b).unwrap());
if off.is_empty() { 0.0 } else { off[off.len() / 2] }
});
for i in 0..n {
s[i][i] = pref;
}
let mut r = vec![vec![0.0_f64; n]; n];
let mut a = vec![vec![0.0_f64; n]; n];
let mut ex_history: Vec<Vec<bool>> = Vec::new();
let mut used = 0_usize;
let mut converged = false;
for it in 0..max_iter {
used = it + 1;
// Responsibility update: r(i, k) ← s(i, k) − max_{k' ≠ k} (a(i, k') + s(i, k'))
let mut r_new = vec![vec![0.0_f64; n]; n];
for i in 0..n {
let mut m1 = f64::NEG_INFINITY;
let mut m2 = f64::NEG_INFINITY;
let mut m1_idx = 0_usize;
for k in 0..n {
let val = a[i][k] + s[i][k];
if val > m1 {
m2 = m1;
m1 = val;
m1_idx = k;
} else if val > m2 {
m2 = val;
}
}
for k in 0..n {
let max_other = if k == m1_idx { m2 } else { m1 };
r_new[i][k] = s[i][k] - max_other;
}
}
for i in 0..n {
for k in 0..n {
r[i][k] = damping * r[i][k] + (1.0 - damping) * r_new[i][k];
}
}
// Availability update
let mut a_new = vec![vec![0.0_f64; n]; n];
for k in 0..n {
let mut col_sum_pos = 0.0_f64;
for i in 0..n {
if i != k {
col_sum_pos += r[i][k].max(0.0);
}
}
a_new[k][k] = col_sum_pos;
for i in 0..n {
if i == k {
continue;
}
let mut inner = r[k][k] + col_sum_pos - r[i][k].max(0.0);
inner = inner.min(0.0);
a_new[i][k] = inner;
}
}
for i in 0..n {
for k in 0..n {
a[i][k] = damping * a[i][k] + (1.0 - damping) * a_new[i][k];
}
}
// Extract exemplars.
let ex: Vec<bool> = (0..n).map(|i| a[i][i] + r[i][i] > 0.0).collect();
ex_history.push(ex.clone());
if ex_history.len() > convergence_iter {
ex_history.remove(0);
}
if ex_history.len() == convergence_iter {
let stable = ex_history
.iter()
.all(|snap| snap == ex_history.last().unwrap());
if stable {
converged = true;
break;
}
}
}
let exemplars: Vec<usize> = (0..n).filter(|&i| a[i][i] + r[i][i] > 0.0).collect();
let labels: Vec<i64> = (0..n)
.map(|i| {
if exemplars.is_empty() {
-1
} else {
let mut best = 0;
let mut best_s = f64::NEG_INFINITY;
for (k, &e) in exemplars.iter().enumerate() {
if s[i][e] > best_s {
best_s = s[i][e];
best = k;
}
}
best as i64
}
})
.collect();
Ok(Self {
labels,
cluster_centers_indices: exemplars,
n_iter: used,
converged,
damping,
})
}
}
#[cfg(test)]
mod tests {
use super::*;
use ndarray::array;
#[test]
fn affinity_propagation_finds_a_positive_number_of_exemplars() {
let x = array![
[0.0_f64, 0.0], [0.1, 0.1], [0.05, 0.15],
[5.0, 5.0], [5.1, 5.1], [5.05, 5.15]
];
let m = AffinityPropagation::fit(x.view()).unwrap();
assert!(!m.cluster_centers_indices.is_empty());
}
}