klaster 0.2.0

Machine learning library providing modern clusterning algorithms for the Rust programming language
Documentation 
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

// Copyright (C) 2025 Piotr Jabłoński
// Extended copyright information can be found in the LICENSE file.

use crate::sdc::cdist::pairwise_distances_squared;
use burn::{
    nn::loss::{MseLoss, Reduction},
    prelude::{Backend, Tensor},
};
use derive_new::new;

#[derive(new)]
pub struct ClusteringLoss;

impl ClusteringLoss {
    /// Compute the combined reconstruction and clustering loss.
    ///
    /// # Params
    /// - `logits`: Input tensor,
    /// - `targets`: Reconstructed input tensor,
    /// - `embeddings`: Latent embeddings from the encoder,
    /// - `centroids`: Cluster centroids,
    /// - `gamma`: Focal exponent for reconstruction loss,
    /// - `alpha`: Weighting for the clustering loss term.
    ///
    /// # Returns
    /// A scalar loss tensor.
    pub fn forward<B: Backend, const D: usize>(
        &self,
        logits: Tensor<B, D>,
        targets: Tensor<B, D>,
        embeddings: Tensor<B, 2>,
        centroids: Tensor<B, 2>,
        gamma: f64,
        alpha: f64,
    ) -> Tensor<B, 1> {
        let mse_loss = MseLoss::new().forward(logits.clone(), targets.clone(), Reduction::Auto);
        let focal_weight = (logits - targets).abs().powf_scalar(gamma);
        let focal_loss = focal_weight.mul(mse_loss.unsqueeze());
        let recon_loss = focal_loss.mean();

        // Clustering loss
        let dist = pairwise_distances_squared(embeddings, centroids);

        let q: Tensor<B, 2> = 1.0 / (1.0 + dist / alpha);
        let q: Tensor<B, 2> = q.powf_scalar(f64::midpoint(alpha, 1.0));
        let q: Tensor<B, 2> = q.clone() / q.sum_dim(1);

        let p: Tensor<B, 2> = q.clone().powi_scalar(2) / q.clone().sum_dim(0);
        let p: Tensor<B, 2> = p.clone() / p.sum_dim(1);

        let cluster_loss = ((p.clone() / (q + 1e-8)) + 1e-8)
            .log()
            .mul(p)
            .sum_dim(1)
            .mean();

        // Combined loss with weighting coefficients
        recon_loss + alpha * cluster_loss
    }
}