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
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
//! FastICA — fixed-point Independent Component Analysis with deflation.
//!
//! Mirrors `sklearn.decomposition.FastICA` with `algorithm='deflation'` and
//! `fun='logcosh'`. Standard pipeline:
//!
//! 1. Centre and whiten X via PCA so that `cov(X) = I`.
//! 2. For each component, iterate the fixed-point update
//! `w ← E[X g(wᵀX)] - E[g'(wᵀX)] w`, orthogonalise against previously
//! extracted components, normalise to unit length.
//! 3. The sources `S = W X_white`; the unmixing matrix in original space is
//! `W K` where `K` is the whitening matrix.
use anofox_ml_core::{FitUnsupervised, Result, RustMlError, Transform};
use faer::linalg::solvers::Svd;
use faer::Mat;
use ndarray::{Array1, Array2};
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
#[derive(Debug, Clone)]
pub struct FastIca {
pub n_components: usize,
pub max_iter: usize,
pub tol: f64,
pub seed: u64,
}
impl FastIca {
pub fn new(n_components: usize) -> Self {
Self {
n_components,
max_iter: 200,
tol: 1e-4,
seed: 0,
}
}
pub fn with_max_iter(mut self, m: usize) -> Self {
self.max_iter = m;
self
}
pub fn with_tol(mut self, t: f64) -> Self {
self.tol = t;
self
}
pub fn with_seed(mut self, s: u64) -> Self {
self.seed = s;
self
}
}
#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct FittedFastIca {
/// Unmixing matrix from whitened space, shape (n_components, n_components).
pub w: Array2<f64>,
/// Whitening matrix `K` such that `X_centered @ K` is whitened, shape
/// (n_features, n_components).
pub whitening: Array2<f64>,
/// Per-feature mean used for centring.
pub mean: Array1<f64>,
pub n_features: usize,
}
/// Logcosh non-linearity: `g(u) = tanh(u)`, `g'(u) = 1 - tanh(u)²`.
fn g_logcosh(u: f64) -> (f64, f64) {
let t = u.tanh();
(t, 1.0 - t * t)
}
impl FitUnsupervised<f64> for FastIca {
type Fitted = FittedFastIca;
fn fit(&self, x: &Array2<f64>) -> Result<Self::Fitted> {
let n = x.nrows();
let d = x.ncols();
let k = self.n_components.min(d.min(n));
if k == 0 {
return Err(RustMlError::InvalidParameter("n_components >= 1".into()));
}
if n < 2 {
return Err(RustMlError::EmptyInput("need at least 2 samples".into()));
}
// 1. Centre.
let mut mean = Array1::<f64>::zeros(d);
for j in 0..d {
mean[j] = x.column(j).sum() / n as f64;
}
let mut xc = x.clone();
for j in 0..d {
for i in 0..n {
xc[[i, j]] -= mean[j];
}
}
// 2. Whiten via SVD: X_centered = U Σ Vᵀ. Whitening matrix K = V Σ⁻¹ √(n-1).
let xm = Mat::from_fn(n, d, |i, j| xc[[i, j]]);
let svd = Svd::new(xm.as_ref())
.map_err(|e| RustMlError::InvalidParameter(format!("SVD failed: {e:?}")))?;
let s = svd.S();
let v = svd.V();
let scale = (n as f64 - 1.0).sqrt();
let mut k_white = Array2::<f64>::zeros((d, k));
for c in 0..k {
let sigma = s.column_vector()[c].max(1e-12);
for j in 0..d {
k_white[[j, c]] = v[(j, c)] * scale / sigma;
}
}
// Whitened data: X1 = X_centered @ K, shape (n, k).
let x1 = xc.dot(&k_white);
// 3. Deflation extraction.
let mut rng = StdRng::seed_from_u64(self.seed);
let mut w = Array2::<f64>::zeros((k, k));
for comp in 0..k {
// Random init.
let mut wi: Array1<f64> = Array1::from_shape_fn(k, |_| rng.gen::<f64>() * 2.0 - 1.0);
// Normalize.
let nrm = wi.iter().map(|v| v * v).sum::<f64>().sqrt().max(1e-12);
wi.mapv_inplace(|v| v / nrm);
for _ in 0..self.max_iter {
// Compute u = X1 @ wi (length n).
let mut u = vec![0.0_f64; n];
for i in 0..n {
let mut s = 0.0;
for c in 0..k {
s += x1[[i, c]] * wi[c];
}
u[i] = s;
}
// g(u) and g'(u).
let mut gu = vec![0.0_f64; n];
let mut g_prime_mean = 0.0_f64;
for i in 0..n {
let (g, gp) = g_logcosh(u[i]);
gu[i] = g;
g_prime_mean += gp;
}
g_prime_mean /= n as f64;
// New wi: E[X1 g(wᵀ X1)] - E[g'(...)] w.
let mut new_wi = Array1::<f64>::zeros(k);
for c in 0..k {
let mut s = 0.0;
for i in 0..n {
s += x1[[i, c]] * gu[i];
}
new_wi[c] = s / n as f64 - g_prime_mean * wi[c];
}
// Deflate: orthogonalise against previously-extracted components.
for prev in 0..comp {
let mut dot = 0.0;
for c in 0..k {
dot += new_wi[c] * w[[prev, c]];
}
for c in 0..k {
new_wi[c] -= dot * w[[prev, c]];
}
}
// Normalise.
let nrm = new_wi.iter().map(|v| v * v).sum::<f64>().sqrt().max(1e-12);
new_wi.mapv_inplace(|v| v / nrm);
// Convergence: |1 - |<w_new, w_old>||.
let mut dot = 0.0;
for c in 0..k {
dot += new_wi[c] * wi[c];
}
let conv = (1.0 - dot.abs()).abs();
wi = new_wi;
if conv < self.tol {
break;
}
}
for c in 0..k {
w[[comp, c]] = wi[c];
}
}
Ok(FittedFastIca {
w,
whitening: k_white,
mean,
n_features: d,
})
}
}
impl Transform<f64> for FittedFastIca {
/// Returns the recovered source signals `S = (X - mean) · K · Wᵀ`.
fn transform(&self, x: &Array2<f64>) -> Result<Array2<f64>> {
if x.ncols() != self.n_features {
return Err(RustMlError::ShapeMismatch(format!(
"expected {} features, got {}",
self.n_features,
x.ncols()
)));
}
let mut xc = x.clone();
for j in 0..self.n_features {
for i in 0..x.nrows() {
xc[[i, j]] -= self.mean[j];
}
}
let x_white = xc.dot(&self.whitening);
Ok(x_white.dot(&self.w.t()))
}
}
#[cfg(test)]
mod tests {
use super::*;
use ndarray::array;
#[test]
fn test_fast_ica_runs() {
// Build a mixture of two simple signals; FastICA should separate them.
let n = 100;
let mut s = Array2::<f64>::zeros((n, 2));
for i in 0..n {
let t = i as f64 * 0.1;
s[[i, 0]] = t.sin(); // sine
s[[i, 1]] = (t * 0.3).sin().signum(); // square
}
// Mixing matrix.
let a = array![[1.0_f64, 0.5], [0.5, 1.0]];
let x = s.dot(&a);
let fitted = FastIca::new(2).with_seed(1).fit(&x).unwrap();
let recovered = fitted.transform(&x).unwrap();
assert_eq!(recovered.shape(), &[n, 2]);
for v in recovered.iter() {
assert!(v.is_finite());
}
}
}