shap-rs 0.1.0

Native Rust implementations of model-agnostic, linear, and TreeSHAP explainers
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

#![cfg(feature = "json-adapters")]

use ndarray::{Array1, Array2, Array3, ArrayView2};
use serde::Deserialize;
use shap_rs::{
    explainers::{
        ExactExplainer, HierarchicalPartitionExplainer, KernelExplainer, LinearExplainer,
        PartitionNode, PartitionTree, PermutationExplainer,
    },
    Background, Explainer, Explanation, FnModel, Link,
};

#[derive(Deserialize)]
struct Expected {
    values: Vec<Vec<Vec<f64>>>,
    base_values: Vec<Vec<f64>>,
}

#[derive(Deserialize)]
struct Fixture {
    background: Vec<Vec<f64>>,
    samples: Vec<Vec<f64>>,
    coefficients: Vec<Vec<f64>>,
    intercept: Vec<f64>,
    exact: Expected,
    permutation: Expected,
    kernel: Expected,
    linear: Expected,
    partition: Expected,
    exact_logit: Expected,
}

fn matrix(rows: &[Vec<f64>]) -> Array2<f64> {
    Array2::from_shape_vec(
        (rows.len(), rows.first().unwrap().len()),
        rows.iter().flatten().copied().collect(),
    )
    .unwrap()
}

fn assert_matches(actual: &Explanation, expected: &Expected, tolerance: f64) {
    let dimensions = (
        expected.values.len(),
        expected.values[0].len(),
        expected.values[0][0].len(),
    );
    let values = Array3::from_shape_vec(
        dimensions,
        expected
            .values
            .iter()
            .flatten()
            .flatten()
            .copied()
            .collect(),
    )
    .unwrap();
    let bases = matrix(&expected.base_values);
    assert_eq!(actual.values().dim(), values.dim());
    assert_eq!(actual.base_values().dim(), bases.dim());
    for (left, right) in actual.values().iter().zip(values.iter()) {
        assert!((left - right).abs() <= tolerance, "{left} != {right}");
    }
    for (left, right) in actual.base_values().iter().zip(bases.iter()) {
        assert!((left - right).abs() <= tolerance, "{left} != {right}");
    }
}

#[test]
fn core_explainers_match_python_shap_046() {
    let fixture: Fixture =
        serde_json::from_str(include_str!("fixtures/python_shap_linear.json")).unwrap();
    let background_data = matrix(&fixture.background);
    let samples = matrix(&fixture.samples);
    let coefficients = matrix(&fixture.coefficients);
    let intercept = Array1::from(fixture.intercept.clone());
    let make_model = || {
        let coefficients = coefficients.clone();
        let intercept = intercept.clone();
        FnModel::new(move |x: ArrayView2<'_, f64>| Ok(x.dot(&coefficients) + &intercept))
    };
    let make_background = || Background::new(background_data.clone()).unwrap();

    assert_matches(
        &ExactExplainer::new(make_model(), make_background())
            .explain(samples.view())
            .unwrap(),
        &fixture.exact,
        1e-10,
    );
    assert_matches(
        &PermutationExplainer::new(make_model(), make_background())
            .with_n_permutations(300)
            .with_seed(7)
            .explain(samples.view())
            .unwrap(),
        &fixture.permutation,
        1e-10,
    );
    assert_matches(
        &KernelExplainer::new(make_model(), make_background())
            .with_nsamples(256)
            .with_seed(7)
            .explain(samples.view())
            .unwrap(),
        &fixture.kernel,
        1e-8,
    );
    assert_matches(
        &LinearExplainer::new(coefficients.clone(), intercept.clone(), make_background())
            .unwrap()
            .explain(samples.view())
            .unwrap(),
        &fixture.linear,
        1e-10,
    );
    let hierarchy = PartitionTree::new(
        PartitionNode::Group(
            Box::new(PartitionNode::Group(
                Box::new(PartitionNode::Feature(0)),
                Box::new(PartitionNode::Feature(1)),
            )),
            Box::new(PartitionNode::Feature(2)),
        ),
        3,
    )
    .unwrap();
    assert_matches(
        &HierarchicalPartitionExplainer::new(make_model(), make_background(), hierarchy)
            .explain(samples.view())
            .unwrap(),
        &fixture.partition,
        1e-10,
    );
    let probability_model = FnModel::new(|x: ArrayView2<'_, f64>| {
        Ok(Array2::from_shape_fn((x.nrows(), 1), |(row, _)| {
            let margin = 0.7 * x[[row, 0]] - 0.4 * x[[row, 1]] + 0.2 * x[[row, 2]] + 0.1;
            1.0 / (1.0 + (-margin).exp())
        }))
    });
    assert_matches(
        &ExactExplainer::new(probability_model, make_background())
            .with_link(Link::Logit)
            .explain(samples.view())
            .unwrap(),
        &fixture.exact_logit,
        1e-10,
    );
}