ndarrow 0.0.3

Zero-copy bridge between Apache Arrow and ndarray
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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249

//! Integration tests verifying round-trip correctness.
//!
//! Every value that goes Arrow → ndarray → Arrow (or vice versa) must come back identical.
//! These tests exercise the public API exactly as a user would.

use approx::assert_abs_diff_eq;
use arrow_array::{
    Array, Float32Array, Float64Array,
    types::{Float32Type, Float64Type},
};
use ndarray::{Array1, Array2, array};
use ndarrow::{AsNdarray, IntoArrow};
use num_traits::FromPrimitive;

// ─── Scalar round-trips ───

#[test]
fn f64_arrow_to_ndarray_to_arrow() {
    let original = Float64Array::from(vec![1.0, 2.0, 3.0, 4.0, 5.0]);

    // Arrow → ndarray view
    let view = original.as_ndarray().unwrap();
    assert_eq!(view.len(), 5);

    // Copy view into owned array (simulating a computation result)
    let owned = view.to_owned();

    // ndarray → Arrow
    let result: Float64Array = owned.into_arrow().unwrap();

    assert_eq!(result.len(), original.len());
    for i in 0..result.len() {
        assert_abs_diff_eq!(result.value(i), original.value(i));
    }
}

#[test]
fn f32_arrow_to_ndarray_to_arrow() {
    let original = Float32Array::from(vec![1.0_f32, 2.0, 3.0]);

    let view = original.as_ndarray().unwrap();
    let owned = view.to_owned();
    let result: Float32Array = owned.into_arrow().unwrap();

    assert_eq!(result.len(), original.len());
    for i in 0..result.len() {
        assert_abs_diff_eq!(result.value(i), original.value(i));
    }
}

#[test]
fn f64_ndarray_to_arrow_to_ndarray() {
    let original = Array1::from_vec(vec![10.0_f64, 20.0, 30.0, 40.0]);
    let expected = original.clone();

    // ndarray → Arrow
    let arrow: Float64Array = original.into_arrow().unwrap();

    // Arrow → ndarray view
    let view = arrow.as_ndarray().unwrap();

    for (actual, expected) in view.iter().zip(expected.iter()) {
        assert_abs_diff_eq!(*actual, *expected);
    }
}

#[test]
fn f32_ndarray_to_arrow_to_ndarray() {
    let original = Array1::from_vec(vec![1.5_f32, 2.5, 3.5]);
    let expected = original.clone();

    let arrow: Float32Array = original.into_arrow().unwrap();
    let view = arrow.as_ndarray().unwrap();

    for (actual, expected) in view.iter().zip(expected.iter()) {
        assert_abs_diff_eq!(*actual, *expected);
    }
}

// ─── Matrix round-trips ───

#[test]
fn f64_matrix_arrow_to_ndarray_to_arrow() {
    use std::sync::Arc;

    use arrow_schema::Field;

    // Build a FixedSizeListArray: 3 rows of 4-dimensional vectors
    let values = Float64Array::from(vec![
        1.0, 2.0, 3.0, 4.0, // row 0
        5.0, 6.0, 7.0, 8.0, // row 1
        9.0, 10.0, 11.0, 12.0, // row 2
    ]);
    let field = Arc::new(Field::new("item", arrow_schema::DataType::Float64, false));
    let fsl = arrow_array::FixedSizeListArray::new(field, 4, Arc::new(values), None);

    // Arrow → ndarray
    let view = ndarrow::fixed_size_list_as_array2::<Float64Type>(&fsl).unwrap();
    assert_eq!(view.dim(), (3, 4));

    // Simulate computation: double all values
    let doubled = &view * 2.0;

    // ndarray → Arrow
    let result = doubled.into_arrow().unwrap();
    assert_eq!(result.len(), 3);
    assert_eq!(result.value_length(), 4);

    // Verify values
    let inner = result.values().as_any().downcast_ref::<Float64Array>().unwrap();
    let result_values: Vec<f64> = inner.values().iter().copied().collect();
    let expected = [2.0, 4.0, 6.0, 8.0, 10.0, 12.0, 14.0, 16.0, 18.0, 20.0, 22.0, 24.0];
    assert_eq!(result_values.len(), expected.len());
    for (actual, expected) in result_values.iter().zip(expected.iter()) {
        assert_abs_diff_eq!(*actual, *expected);
    }
}

#[test]
fn f64_matrix_ndarray_to_arrow_to_ndarray() {
    let original = array![[1.0_f64, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]];
    let expected = original.clone();

    let fsl = original.into_arrow().unwrap();
    let view = ndarrow::fixed_size_list_as_array2::<Float64Type>(&fsl).unwrap();

    assert_eq!(view.dim(), expected.dim());
    for (actual, expected) in view.iter().zip(expected.iter()) {
        assert_abs_diff_eq!(*actual, *expected);
    }
}

#[test]
fn f32_matrix_ndarray_to_arrow_to_ndarray() {
    let original = array![[1.0_f32, 2.0], [3.0, 4.0]];
    let expected = original.clone();

    let fsl = original.into_arrow().unwrap();
    let view = ndarrow::fixed_size_list_as_array2::<Float32Type>(&fsl).unwrap();

    assert_eq!(view.dim(), expected.dim());
    for (actual, expected) in view.iter().zip(expected.iter()) {
        assert_abs_diff_eq!(*actual, *expected);
    }
}

// ─── Edge cases ───

#[test]
fn empty_scalar_roundtrip() {
    let arr = Float64Array::from(Vec::<f64>::new());
    let view = arr.as_ndarray().unwrap();
    assert_eq!(view.len(), 0);

    let owned = view.to_owned();
    let result: Float64Array = owned.into_arrow().unwrap();
    assert_eq!(result.len(), 0);
}

#[test]
fn single_element_roundtrip() {
    let arr = Float64Array::from(vec![42.0]);
    let view = arr.as_ndarray().unwrap();
    let result: Float64Array = view.to_owned().into_arrow().unwrap();
    assert_abs_diff_eq!(result.value(0), 42.0);
}

#[test]
fn large_array_roundtrip() {
    let data: Vec<f64> = (0..10_000).map(f64::from).collect();
    let arr = Float64Array::from(data.clone());

    let view = arr.as_ndarray().unwrap();
    let result: Float64Array = view.to_owned().into_arrow().unwrap();

    assert_eq!(result.len(), 10_000);
    for (i, expected) in data.iter().enumerate() {
        assert_abs_diff_eq!(result.value(i), *expected);
    }
}

#[test]
fn large_matrix_roundtrip() {
    // 1000 rows of 128-dimensional vectors (typical embedding size)
    let data: Vec<f32> = (0_u32..128_000_u32)
        .map(|i| f32::from_u32(i).expect("u32 to f32 conversion should succeed for test range"))
        .collect();
    let matrix = Array2::from_shape_vec((1000, 128), data).unwrap();
    let expected = matrix.clone();

    let fsl = matrix.into_arrow().unwrap();
    assert_eq!(fsl.len(), 1000);
    assert_eq!(fsl.value_length(), 128);

    let view = ndarrow::fixed_size_list_as_array2::<Float32Type>(&fsl).unwrap();
    for (actual, expected) in view.iter().zip(expected.iter()) {
        assert_abs_diff_eq!(*actual, *expected);
    }
}

// ─── Computation pipeline ───

#[test]
fn realistic_computation_pipeline() {
    // Simulate: receive vectors from Arrow, normalize, return to Arrow.
    let raw_data: Vec<f32> = vec![
        3.0, 4.0, // row 0: norm = 5
        0.0, 1.0, // row 1: norm = 1
        1.0, 0.0, // row 2: norm = 1
    ];
    let values = Float32Array::from(raw_data);
    let field = std::sync::Arc::new(arrow_schema::Field::new(
        "item",
        arrow_schema::DataType::Float32,
        false,
    ));
    let fsl = arrow_array::FixedSizeListArray::new(field, 2, std::sync::Arc::new(values), None);

    // Arrow → ndarray
    let view = ndarrow::fixed_size_list_as_array2::<Float32Type>(&fsl).unwrap();

    // Normalize each row (L2)
    let mut result = view.to_owned();
    for mut row in result.rows_mut() {
        let norm: f32 = row.iter().map(|x| x * x).sum::<f32>().sqrt();
        if norm > 0.0 {
            row /= norm;
        }
    }

    // ndarray → Arrow
    let output = result.into_arrow().unwrap();
    assert_eq!(output.len(), 3);
    assert_eq!(output.value_length(), 2);

    // Verify normalized values
    let inner = output.values().as_any().downcast_ref::<Float32Array>().unwrap();
    let vals: Vec<f32> = inner.values().iter().copied().collect();

    // Row 0: [3/5, 4/5] = [0.6, 0.8]
    assert!((vals[0] - 0.6).abs() < 1e-6);
    assert!((vals[1] - 0.8).abs() < 1e-6);
    // Row 1: [0, 1]
    assert!((vals[2] - 0.0).abs() < 1e-6);
    assert!((vals[3] - 1.0).abs() < 1e-6);
    // Row 2: [1, 0]
    assert!((vals[4] - 1.0).abs() < 1e-6);
    assert!((vals[5] - 0.0).abs() < 1e-6);
}