benchmark/
benchmark.rs

1// Copyright (C) 2024 Hallvard Høyland Lavik
2
3extern crate neurons;
4
5use neurons::{activation, network, objective, optimizer, tensor};
6
7use std::fs::File;
8use std::io::{BufReader, Read, Result};
9use std::time;
10
11fn read(reader: &mut dyn Read) -> Result<u32> {
12    let mut buffer = [0; 4];
13    reader.read_exact(&mut buffer)?;
14    Ok(u32::from_be_bytes(buffer))
15}
16
17fn load_images(path: &str) -> Result<Vec<tensor::Tensor>> {
18    let mut reader = BufReader::new(File::open(path)?);
19    let mut images: Vec<tensor::Tensor> = Vec::new();
20
21    let _magic_number = read(&mut reader)?;
22    let num_images = read(&mut reader)?;
23    let num_rows = read(&mut reader)?;
24    let num_cols = read(&mut reader)?;
25
26    for _ in 0..num_images {
27        let mut image: Vec<Vec<f32>> = Vec::new();
28        for _ in 0..num_rows {
29            let mut row: Vec<f32> = Vec::new();
30            for _ in 0..num_cols {
31                let mut pixel = [0];
32                reader.read_exact(&mut pixel)?;
33                row.push(pixel[0] as f32 / 255.0);
34            }
35            image.push(row);
36        }
37        images.push(tensor::Tensor::triple(vec![image]).resize(tensor::Shape::Triple(1, 14, 14)));
38    }
39
40    Ok(images)
41}
42
43fn load_labels(file_path: &str, numbers: usize) -> Result<Vec<tensor::Tensor>> {
44    let mut reader = BufReader::new(File::open(file_path)?);
45    let _magic_number = read(&mut reader)?;
46    let num_labels = read(&mut reader)?;
47
48    let mut _labels = vec![0; num_labels as usize];
49    reader.read_exact(&mut _labels)?;
50
51    Ok(_labels
52        .iter()
53        .map(|&x| tensor::Tensor::one_hot(x as usize, numbers))
54        .collect())
55}
56
57fn main() {
58    let x_train = load_images("./examples/datasets/mnist/train-images-idx3-ubyte").unwrap();
59    let y_train = load_labels("./examples/datasets/mnist/train-labels-idx1-ubyte", 10).unwrap();
60
61    let x_train: Vec<&tensor::Tensor> = x_train.iter().collect();
62    let y_train: Vec<&tensor::Tensor> = y_train.iter().collect();
63
64    let mut times: Vec<time::Duration> = Vec::new();
65
66    for iteration in 0..10 {
67        let start = time::Instant::now();
68
69        // Create the network
70        let mut network = network::Network::new(tensor::Shape::Triple(1, 14, 14));
71
72        network.convolution(
73            8,
74            (3, 3),
75            (1, 1),
76            (0, 0),
77            (1, 1),
78            activation::Activation::ReLU,
79            Some(0.05),
80        );
81        network.maxpool((2, 2), (2, 2));
82        network.dense(10, activation::Activation::Softmax, true, None);
83
84        network.set_optimizer(optimizer::Adam::create(
85            0.001,      // Learning rate
86            0.9,        // Beta1
87            0.999,      // Beta2
88            1e-8,       // Epsilon
89            Some(0.01), // Decay
90        ));
91        network.set_objective(
92            objective::Objective::CrossEntropy, // Objective function
93            None,                               // Gradient clipping
94        );
95
96        // Train the network
97        let (train_loss, _, _) = network.learn(&x_train, &y_train, None, 128, 10, None);
98
99        println!("Iteration: {}, Loss: {:?}", iteration, train_loss);
100
101        let duration = start.elapsed();
102        times.push(duration);
103    }
104
105    let sum: time::Duration = times.iter().sum();
106    let avg = sum / times.len() as u32;
107    println!("Average time: {:?}", avg);
108}
benchmark/benchmark.rs

benchmark/
benchmark.rs
