Server-side WebAssembly
Function as a Service
Machine learning
Use k-means clustering algorithm to classify data points
The lingua franca of machine learning is Python. However, Python relies on C/C++ based native modules to perform the actual computationally intensive tasks of machine learning. It is similar to Node.js relying on C++ to perform computing tasks.
For new machine learning algorithms, developers can choose to implement them in Python for developer productivity or in C++ for runtime efficiency. Now, there is a third choice. Implementing machine learning algorithms in Rust could provide a 25x performance gain over Python as well as safety over C++. In this tutorial, we will demonstrate how to do k-means clustering computation in Rust, and make the function available in Node.js.
The Rust function
fit() is as follows. It reads content from a CSV data file, and group the points into clusters based on the dimensions for the points and the number of estimated clusters.1
use wasm_bindgen::prelude::*;
2
use ndarray::{Array2};
3
use std::str::FromStr;
4
5
#[wasm_bindgen]
6
pub fn fit (csv_content: &[u8], dim: i32, num_clusters: i32) -> String {
7
let data = read_data(csv_content, dim as usize);
8
let (means, _clusters) = rkm::kmeans_lloyd(&data.view(), num_clusters as usize);
9
return serde_json::to_string(&means).unwrap();
10
}
11
12
fn read_data(csv_content: &[u8], dim: usize) -> Array2<f32> {
13
let mut data_reader = csv::Reader::from_reader(csv_content);
14
let mut data: Vec<f32> = Vec::new();
15
for record in data_reader.records() {
16
for field in record.unwrap().iter() {
17
let value = f32::from_str(field);
18
data.push(value.unwrap());
19
}
20
}
21
Array2::from_shape_vec((data.len() / dim, dim), data).unwrap()
22
}
Copied!
The Javascript host application reads the CSV file, and calls the Rust function to perform the computation. The results are returned as a multi-dimensional array for the cluster centers.
1
const { fit } = require('../pkg/kmeans_lib.js');
2
const fs = require('fs');
3
4
var birch3_csv = fs.readFileSync("birch3.data.csv");
5
var means = JSON.parse( fit(birch3_csv, 2, 100) );
6
console.log(means);
Copied!
Rust and WebAssembly made it easy to make high performance machine learning algorithms available as web services.
Last modified 1yr ago
Copy link