sim-lib-numbers-tensor 0.4.0

Uniform n-dimensional tensor value, constructors, and specialization hooks for SIM numbers.
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
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155

use sim_kernel::Symbol;

use crate::{
    CanonicalAttrs, CpuTensorExecutor, PadMode, TensorExecutor, TensorMeta, allclose_op_symbol,
    arange_op_symbol, build_tensor_value, execute_canonical_tensor_op, isclose_op_symbol,
    maximum_op_symbol, nonzero_op_symbol, pad_op_symbol, tensor_value_ref,
};

use super::{number, test_cx};

fn tensor(
    cx: &mut sim_kernel::Cx,
    shape: Vec<usize>,
    domain: &str,
    values: &[&str],
) -> crate::Tensor {
    let dtype = Symbol::qualified("numbers", domain);
    let value = build_tensor_value(
        cx,
        shape,
        Some(dtype.clone()),
        values.iter().map(|v| number(domain, v)).collect(),
    )
    .unwrap();
    tensor_value_ref(&value).unwrap().clone()
}
fn text(cx: &mut sim_kernel::Cx, t: &crate::Tensor) -> Vec<String> {
    t.cells()
        .unwrap()
        .iter()
        .map(|v| v.object().display(cx).unwrap())
        .collect()
}

#[test]
fn construction_selection_and_edge_policies_are_explicit() {
    let mut cx = test_cx();
    let range = execute_canonical_tensor_op(
        &mut cx,
        arange_op_symbol(),
        vec![],
        TensorMeta::new(vec![3], Symbol::qualified("numbers", "f64")),
        CanonicalAttrs::Range {
            start: 0.0,
            stop: 1.0,
            step: 0.5,
            inclusive: true,
        },
    )
    .unwrap();
    assert_eq!(text(&mut cx, &range), ["0", "0.5", "1"]);
    let nz = execute_canonical_tensor_op(
        &mut cx,
        nonzero_op_symbol(),
        vec![range],
        TensorMeta::new(vec![2], Symbol::qualified("numbers", "i64")),
        CanonicalAttrs::None,
    )
    .unwrap();
    assert_eq!(text(&mut cx, &nz), ["1", "2"]);
    let empty = execute_canonical_tensor_op(
        &mut cx,
        arange_op_symbol(),
        vec![],
        TensorMeta::new(vec![0], Symbol::qualified("numbers", "f32")),
        CanonicalAttrs::Range {
            start: 1.0,
            stop: 1.0,
            step: 1.0,
            inclusive: false,
        },
    )
    .unwrap();
    assert!(empty.is_empty());
}

#[test]
fn strict_shape_tolerance_nan_and_padding_rules_hold() {
    let mut cx = test_cx();
    let a = tensor(&mut cx, vec![2], "f64", &["NaN", "-0.0"]);
    let b = tensor(&mut cx, vec![2], "f64", &["NaN", "0.0"]);
    let close = execute_canonical_tensor_op(
        &mut cx,
        isclose_op_symbol(),
        vec![a.clone(), b.clone()],
        TensorMeta::new(vec![2], Symbol::qualified("numbers", "f64")),
        CanonicalAttrs::Close {
            relative: 0.0,
            absolute: 0.0,
            equal_nan: true,
        },
    )
    .unwrap();
    assert_eq!(text(&mut cx, &close), ["1", "1"]);
    let bad = execute_canonical_tensor_op(
        &mut cx,
        allclose_op_symbol(),
        vec![a.clone(), b],
        TensorMeta::new(vec![], Symbol::qualified("numbers", "f64")),
        CanonicalAttrs::Close {
            relative: -1.0,
            absolute: 0.0,
            equal_nan: false,
        },
    );
    let error = match bad {
        Err(error) => error,
        Ok(_) => panic!("negative dimension unexpectedly produced a tensor"),
    };
    assert!(error.to_string().contains("non-negative"));
    let finite = tensor(&mut cx, vec![2], "i64", &["1", "2"]);
    let padded = execute_canonical_tensor_op(
        &mut cx,
        pad_op_symbol(),
        vec![finite.clone()],
        TensorMeta::new(vec![4], Symbol::qualified("numbers", "i64")),
        CanonicalAttrs::Pad {
            widths: vec![(1, 1)].into(),
            mode: PadMode::Constant(9.0),
        },
    )
    .unwrap();
    assert_eq!(text(&mut cx, &padded), ["9", "1", "2", "9"]);
    let mismatch = tensor(&mut cx, vec![1], "i64", &["1"]);
    assert!(
        execute_canonical_tensor_op(
            &mut cx,
            maximum_op_symbol(),
            vec![finite, mismatch],
            TensorMeta::new(vec![2], Symbol::qualified("numbers", "i64")),
            CanonicalAttrs::None
        )
        .is_err()
    );
}

#[test]
fn cpu_provider_card_advertises_the_complete_vocabulary() {
    let card = CpuTensorExecutor::new().card();
    assert!(card.operations.contains(&arange_op_symbol()));
    assert!(card.operations.contains(&allclose_op_symbol()));
}

#[test]
fn explicit_sum_modes_expose_catastrophic_cancellation() {
    use crate::{SumMode, cumsum_f64, sum_f64};
    let v = [1.0e16, 1.0, -1.0e16];
    assert_eq!(sum_f64(&v, SumMode::Naive), 0.0);
    assert_eq!(sum_f64(&v, SumMode::Neumaier), 1.0);
    assert_eq!(cumsum_f64(&v, SumMode::Neumaier).last(), Some(&1.0));
    assert_eq!(sum_f64(&[1., 2., 3., 4.], SumMode::Pairwise), 10.0);
    // The exact rational sum is one; the compensated mode preserves it despite
    // the two terms sixteen orders of magnitude larger.
    assert_eq!(sum_f64(&[1.0e16, 1.0, -1.0e16], SumMode::Neumaier), 1.0);
}