candela-tensor 0.2.0

A lazy, graph-based tensor engine in Rust
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
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

mod common;

use candela::{Dimension, FloatLikeTensorElement, Layout, Tensor, ones};
use common::{assert_approx_eq, tensor_of};
use rstest::rstest;

// ── Layout-level broadcast unit ───────────────────────────────────────────────

#[test]
fn broadcast_layout_zero_stride_on_expanded_dim() {
    // [4] broadcast to [3,4]: new leading dim gets stride 0
    let l = Layout::new(&[4]);
    let b = l.broadcast(&[3, 4]).unwrap();
    assert_eq!(b.shape(), &[3, 4]);
    assert_eq!(b.stride()[0], 0);
    assert_eq!(b.stride()[1], 1);
    assert_eq!(b.len(), 12);
}

// ── Integration: binary ops auto-broadcast ────────────────────────────────────

#[rstest]
#[case::f64(1.0f64, 2.0f64)]
#[case::f32(1.0f32, 2.0f32)]
fn broadcast_col_plus_row<T: FloatLikeTensorElement>(#[case] input1: T, #[case] input2: T) {
    // [3,1] + [1,4] = [3,4], every element = 1 + 2 = 3
    let col = Tensor::from_scalar(input1, &[3, 1]);
    let row = Tensor::from_scalar(input2, &[1, 4]);
    let result = (col + row).materialize();
    assert_eq!(result.shape(), &[3, 4]);
    assert_approx_eq(result.data(), &[3.0; 12]);
}

#[rstest]
#[case::f64(0.0f64)]
#[case::f32(0.0f32)]
fn broadcast_1d_against_2d<T: FloatLikeTensorElement>(#[case] _t: T) {
    // [4] + [3,4]: each row of result should be [1,2,3,4]
    let v = tensor_of::<T>(&[0.0, 1.0, 2.0, 3.0], &[4]);
    let m = Tensor::from_scalar(T::from_f64(1.0), &[3, 4]);
    let result = (v + m).materialize();
    assert_eq!(result.shape(), &[3, 4]);
    let expected: Vec<f64> = [1.0, 2.0, 3.0, 4.0]
        .iter()
        .cycle()
        .take(12)
        .copied()
        .collect();
    assert_approx_eq(result.data(), &expected);
}

#[rstest]
#[case::f64(0.0f64)]
#[case::f32(0.0f32)]
fn broadcast_scalar_tensor_against_matrix<T: FloatLikeTensorElement>(#[case] _t: T) {
    // [1] * [3,3] = [3,3], all 5.0
    let s = Tensor::from_scalar(T::from_f64(5.0), &[1]);
    let m = Tensor::from_scalar(T::from_f64(1.0), &[3, 3]);
    let result = (s * m).materialize();
    assert_approx_eq(result.data(), &[5.0; 9]);
}

#[rstest]
#[case::f64(0.0f64)]
#[case::f32(0.0f32)]
fn broadcast_size_one_dim_in_both<T: FloatLikeTensorElement>(#[case] _t: T) {
    // [3,1] + [1,4]: values differ
    // col: [[0],[1],[2]] + row: [[0,1,2,3]]
    // result[i][j] = i + j
    let col = tensor_of::<T>(&[0.0, 1.0, 2.0], &[3, 1]);
    let row = tensor_of::<T>(&[0.0, 1.0, 2.0, 3.0], &[1, 4]);
    let result = (col + row).materialize();
    assert_eq!(result.shape(), &[3, 4]);
    let expected = vec![
        0.0, 1.0, 2.0, 3.0, // row 0: 0+j
        1.0, 2.0, 3.0, 4.0, // row 1: 1+j
        2.0, 3.0, 4.0, 5.0, // row 2: 2+j
    ];
    assert_approx_eq(result.data(), &expected);
}

#[rstest]
#[case::f64(0.0f64)]
#[case::f32(0.0f32)]
fn broadcast_mul_with_row_vector<T: FloatLikeTensorElement>(#[case] _t: T) {
    // [3,1] * [1,3]: outer product style
    let col = tensor_of::<T>(&[1.0, 2.0, 3.0], &[3, 1]);
    let row = tensor_of::<T>(&[1.0, 2.0, 3.0], &[1, 3]);
    let result = (col * row).materialize();
    assert_eq!(result.shape(), &[3, 3]);
    let expected = vec![
        1.0, 2.0, 3.0, // 1 * [1,2,3]
        2.0, 4.0, 6.0, // 2 * [1,2,3]
        3.0, 6.0, 9.0, // 3 * [1,2,3]
    ];
    assert_approx_eq(result.data(), &expected);
}

#[test]
#[should_panic]
fn broadcast_incompatible_shapes_panics() {
    // [3,4] + [2,4] - dim 0 mismatch, neither is 1
    let a = ones!(&[3, 4]);
    let b = ones!(&[2, 4]);
    let _ = (a + b).materialize();
}