Version bump

- Implemented T-SNE via the BHTSNE library as a preliminary dimensionality reduction technique

Author: Ian

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "single_algebra"
-version = "0.8.7"
+version = "0.9.0"
 edition = "2021"
 license-file = "LICENSE.md"
 description = "A linear algebra convenience library for the single-rust library. Can be used externally as well."
@@ -36,6 +36,7 @@ single-utilities = "0.8.5"
 linfa-tsne = {version = "0.7.1"}
 linfa = "0.7.1"
 single-svdlib = "1.0.6"
+bhtsne = "0.5.4"
 
 [dev-dependencies]
 criterion = "0.7.0"

Cargo.toml

@@ -18,4 +18,4 @@
 //! - Use **UMAP** (when available) for preserving both local and global structure in embeddings
 
 pub mod pca;
-//pub mod tsne;
+pub mod tsne;

src/dimred/mod.rs

@@ -1,2 +1,66 @@
 // Still working on that
-// Probably going to implement, either using Linfa_tsne or another approach
+// Probably going to implement, either using Linfa_tsne or another approach
+
+use ndarray::{Array2, ArrayD, ArrayViewD};
+use single_utilities::traits::FloatOpsTS;
+
+pub struct TSNEConfig {
+    output_dim: u8,
+    perplexity: f32,
+    epochs: usize,
+    theta: f32,
+}
+
+pub fn run_f32<T: FloatOpsTS>(
+    x: ArrayViewD<f32>,
+    config: TSNEConfig,
+) -> anyhow::Result<ArrayD<f32>> {
+    let n_obs = x.shape()[0];
+    let n_dim = x.shape()[1];
+    let x_slice = x.as_slice().unwrap();
+
+    let x_chunked_slice: Vec<&[f32]> = x_slice.chunks(n_dim).collect();
+    let tsne_result = bhtsne::tSNE::new(&x_chunked_slice)
+        .embedding_dim(config.output_dim)
+        .perplexity(config.perplexity)
+        .epochs(config.epochs)
+        .barnes_hut(config.theta, |sample_a, sample_b| {
+            sample_a
+                .iter()
+                .zip(sample_b.iter())
+                .map(|(&a, &b)| num_traits::Float::powi(a - b, 2))
+                .sum::<f32>()
+                .sqrt()
+        })
+        .embedding();
+
+    let result = Array2::from_shape_vec((n_obs, config.output_dim as usize), tsne_result)?;
+    Ok(result.into_dyn())
+}
+
+pub fn run_f64<T: FloatOpsTS>(
+    x: ArrayViewD<f64>,
+    config: TSNEConfig,
+) -> anyhow::Result<ArrayD<f64>> {
+    let n_obs = x.shape()[0];
+    let n_dim = x.shape()[1];
+    let x_slice = x.as_slice().unwrap();
+
+    let x_chunked_slice: Vec<&[f64]> = x_slice.chunks(n_dim).collect();
+    let tsne_result = bhtsne::tSNE::new(&x_chunked_slice)
+        .embedding_dim(config.output_dim)
+        .perplexity(config.perplexity as f64)
+        .epochs(config.epochs)
+        .barnes_hut(config.theta as f64, |sample_a, sample_b| {
+            sample_a
+                .iter()
+                .zip(sample_b.iter())
+                .map(|(&a, &b)| num_traits::Float::powi(a - b, 2))
+                .sum::<f64>()
+                .sqrt()
+        })
+        .embedding();
+
+    let result = Array2::from_shape_vec((n_obs, config.output_dim as usize), tsne_result)?;
+    Ok(result.into_dyn())
+}