lance-nd 0.1.0

N-dimensional datasets stored in Lance, one dataset or chunk per row
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

//! `WriteOptions::casting`: new rows with another type than the table.

use std::sync::Arc;

use arrow::array::{
    ArrayRef, AsArray, Float32Array, Float64Array, Int32Array, Int64Array, StringArray, UInt16Array,
};
use arrow::datatypes::{DataType, Field, Float32Type, Float64Type, Int32Type, Schema};
use futures::stream;
use lance::Dataset;
use lance_nd::{Casting, Error, NdWriter, WriteMode, WriteOptions, scan, write};
use nd_arrow_array::{Attributes, Dimensions, NdArrowArray, NdRecordBatch};

/// One row with the 1-D variable `v` and the attributes `attrs` on it.
fn row(values: ArrayRef, attrs: Attributes) -> NdRecordBatch {
    let dims = Dimensions::try_new([("n", values.len())]).unwrap();
    let field = Field::new("v", values.data_type().clone(), true);
    let array = NdArrowArray::try_new(values, dims)
        .unwrap()
        .with_attributes(attrs);
    NdRecordBatch::try_new_inferred(Arc::new(Schema::new(vec![field])), vec![array]).unwrap()
}

fn plain(values: ArrayRef) -> NdRecordBatch {
    row(values, Attributes::new())
}

fn options(casting: Casting) -> WriteOptions {
    WriteOptions {
        stats: true,
        casting,
        ..Default::default()
    }
}

/// A writer for a new table with the first row, then the other rows, each
/// in its own write. Returns the result of each later write and the table.
async fn write_rows(
    dir: &tempfile::TempDir,
    casting: Casting,
    first: NdRecordBatch,
    later: Vec<NdRecordBatch>,
) -> (Vec<lance_nd::Result<()>>, Dataset) {
    let uri = dir.path().join("table.lance").to_str().unwrap().to_string();
    let mut writer = NdWriter::create(&uri, &options(casting)).await.unwrap();
    writer.write(stream::iter([Ok(first)])).await.unwrap();
    let mut results = Vec::new();
    for row in later {
        results.push(writer.write(stream::iter([Ok(row)])).await);
    }
    (results, writer.finish().await.unwrap())
}

async fn values(dataset: &Dataset) -> Vec<ArrayRef> {
    let rows = scan(dataset).try_collect().await.unwrap();
    rows.iter()
        .map(|row| row.column_by_name("v").unwrap().values().clone())
        .collect()
}

fn stats_type(dataset: &Dataset) -> DataType {
    let schema = Schema::from(dataset.schema());
    let DataType::Struct(bounds) = schema.field_with_name("__stats__:v").unwrap().data_type()
    else {
        panic!("a stats column is a struct");
    };
    bounds.find("min").unwrap().1.data_type().clone()
}

#[tokio::test]
async fn no_casting_rejects_another_type() {
    let dir = tempfile::tempdir().unwrap();
    let first = plain(Arc::new(Float64Array::from(vec![1.5])));
    let later = vec![plain(Arc::new(Float32Array::from(vec![2.5f32])))];
    let (results, dataset) = write_rows(&dir, Casting::No, first, later).await;
    assert!(matches!(results[0], Err(Error::SchemaMismatch(_))));
    assert_eq!(dataset.count_rows(None).await.unwrap(), 1);
}

#[tokio::test]
async fn safe_widens_new_values() {
    let dir = tempfile::tempdir().unwrap();
    let first = plain(Arc::new(Float64Array::from(vec![1.5])));
    let later = vec![plain(Arc::new(Float32Array::from(vec![2.5f32, 3.25])))];
    let (results, dataset) = write_rows(&dir, Casting::Safe, first, later).await;
    results[0].as_ref().unwrap();
    let values = values(&dataset).await;
    let second = values[1].as_primitive::<Float64Type>();
    assert_eq!(second.values().to_vec(), [2.5, 3.25]);
    assert_eq!(stats_type(&dataset), DataType::Float64);
}

#[tokio::test]
async fn safe_rejects_a_narrowing() {
    let dir = tempfile::tempdir().unwrap();
    let first = plain(Arc::new(Float32Array::from(vec![1.5f32])));
    let later = vec![
        plain(Arc::new(Float64Array::from(vec![2.5]))),
        plain(Arc::new(Int64Array::from(vec![2]))),
    ];
    let (results, dataset) = write_rows(&dir, Casting::Safe, first, later).await;
    assert!(results.iter().all(|r| r.is_err()), "{results:?}");
    assert_eq!(dataset.count_rows(None).await.unwrap(), 1);
}

#[tokio::test]
async fn safe_follows_numpy() {
    // uint16 fits int32, and int32 fits float64.
    let dir = tempfile::tempdir().unwrap();
    let first = plain(Arc::new(Int32Array::from(vec![1])));
    let later = vec![plain(Arc::new(UInt16Array::from(vec![65_535u16])))];
    let (results, dataset) = write_rows(&dir, Casting::Safe, first, later).await;
    results[0].as_ref().unwrap();
    assert_eq!(
        values(&dataset).await[1]
            .as_primitive::<Int32Type>()
            .value(0),
        65_535
    );

    let dir = tempfile::tempdir().unwrap();
    let first = plain(Arc::new(Float64Array::from(vec![0.5])));
    let later = vec![plain(Arc::new(Int32Array::from(vec![7])))];
    let (results, _) = write_rows(&dir, Casting::Safe, first, later).await;
    results[0].as_ref().unwrap();
}

#[tokio::test]
async fn same_kind_narrows_new_values() {
    let dir = tempfile::tempdir().unwrap();
    let first = plain(Arc::new(Float32Array::from(vec![1.5f32])));
    let later = vec![plain(Arc::new(Float64Array::from(vec![0.1])))];
    let (results, dataset) = write_rows(&dir, Casting::SameKind, first, later).await;
    results[0].as_ref().unwrap();
    let values = values(&dataset).await;
    assert_eq!(values[1].as_primitive::<Float32Type>().value(0), 0.1f32);
    assert_eq!(stats_type(&dataset), DataType::Float32);
}

#[tokio::test]
async fn same_kind_rejects_overflow_and_other_kinds() {
    let dir = tempfile::tempdir().unwrap();
    let first = plain(Arc::new(Int32Array::from(vec![1])));
    let later = vec![
        plain(Arc::new(Int64Array::from(vec![5_000_000_000]))),
        plain(Arc::new(Int64Array::from(vec![42]))),
        plain(Arc::new(Float64Array::from(vec![1.0]))),
        plain(Arc::new(StringArray::from(vec!["7"]))),
    ];
    let (results, dataset) = write_rows(&dir, Casting::SameKind, first, later).await;
    assert!(results[0].is_err(), "overflow");
    results[1].as_ref().unwrap();
    assert!(results[2].is_err(), "float into int");
    assert!(results[3].is_err(), "text into int");
    let values = values(&dataset).await;
    assert_eq!(values.len(), 2);
    assert_eq!(values[1].as_primitive::<Int32Type>().value(0), 42);
}

#[tokio::test]
async fn attributes_follow_the_casting() {
    let dir = tempfile::tempdir().unwrap();
    let with = |fill: Attributes| row(Arc::new(Float64Array::from(vec![1.0])), fill);
    let first = with(Attributes::new().with("fill", -1.0f64));
    let later = vec![
        with(Attributes::new().with("fill", -2.0f32)),
        with(Attributes::new().with("fill", vec![1.0f64, 2.0])),
    ];
    let (results, dataset) = write_rows(&dir, Casting::Safe, first, later).await;
    results[0].as_ref().unwrap();
    assert!(results[1].is_err(), "a list is not a number");
    assert_eq!(dataset.count_rows(None).await.unwrap(), 2);
}

#[tokio::test]
async fn write_fn_casts_to_the_table() {
    let dir = tempfile::tempdir().unwrap();
    let uri = dir.path().join("table.lance").to_str().unwrap().to_string();
    let rows = |rows: Vec<NdRecordBatch>| stream::iter(rows.into_iter().map(Ok));
    let first = plain(Arc::new(Float64Array::from(vec![1.5])));
    write(&uri, rows(vec![first]), &options(Casting::No))
        .await
        .unwrap();
    let narrow = plain(Arc::new(Float32Array::from(vec![2.5f32])));
    let append = WriteOptions {
        mode: WriteMode::Append,
        ..options(Casting::Safe)
    };
    let dataset = write(&uri, rows(vec![narrow]), &append).await.unwrap();
    let values = values(&dataset).await;
    assert_eq!(values[1].as_primitive::<Float64Type>().value(0), 2.5);
}