dupehell 0.5.5

Synthetic multi-domain dataset generator for record linkage benchmarking
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

// DupeHell -- MIT License
//
// Synthetic multi-domain dataset generator for record linkage benchmarking.
// No liability for misuse.

use arrow::array::{ArrayRef, UInt64Array};
use arrow::compute::take;
use arrow::record_batch::RecordBatch;

use crate::rng::Rng;

/// Remap a single column in a batch by randomly sampling from a FK pool.
///
/// # Arguments
/// * `batch` - The entity batch whose column will be remapped
/// * `pool_rb` - A RecordBatch containing FK identifier values (column 0) and,
///   when built for graph output, the corresponding `record_id` (column 1)
/// * `src_col` - Name of the column to remap in the batch
/// * `rng` - RNG for random index generation into the pool
/// * `track_src` - When true, also return the target `record_id`s (as the
///   raw Arrow array, no `Vec<String>` conversion) so callers can emit FK
///   edges. The RNG consumption is identical either way.
pub fn fk_remap_batch(
    batch: &RecordBatch,
    pool_rb: &RecordBatch,
    src_col: &str,
    rng: &mut Rng,
    track_src: bool,
) -> Result<(RecordBatch, Option<ArrayRef>), String> {
    let n = batch.num_rows();
    let pool_col = pool_rb.column(0);
    let pool_n = pool_col.len();

    if pool_n == 0 {
        return Err("FK pool is empty".into());
    }

    // Generate random indices into the FK pool, directly as a `UInt64Array`
    // (same RNG sequence, one pass instead of a transient `Vec<usize>`
    // followed by a conversion pass).
    let idx_arr = UInt64Array::from_iter_values((0..n).map(|_| rng.next_usize(pool_n) as u64));

    // Sample values from pool using Arrow take kernel
    let remapped: ArrayRef = take(pool_col, &idx_arr, None)
        .map_err(|e| format!("take error while remapping '{src_col}': {e}"))?;

    // Find column index in batch schema
    let schema = batch.schema();
    let col_idx = schema
        .index_of(src_col)
        .map_err(|_| format!("Column '{src_col}' not found in batch"))?;

    // Rebuild column array with remapped column replaced
    let new_columns: Vec<ArrayRef> = (0..batch.num_columns())
        .map(|i| {
            if i == col_idx {
                remapped.clone()
            } else {
                batch.column(i).clone()
            }
        })
        .collect();

    let batch =
        RecordBatch::try_new(schema, new_columns).map_err(|e| format!("RecordBatch error: {e}"))?;

    // Optionally recover the target record_ids (column 1 of the pool, built
    // only when graph output is enabled) using the same indices. RNG
    // consumption above is unchanged.
    let target_rids = if track_src && pool_rb.num_columns() >= 2 {
        Some(
            take(pool_rb.column(1), &idx_arr, None)
                .map_err(|e| format!("take error on target rids: {e}"))?,
        )
    } else {
        None
    };

    Ok((batch, target_rids))
}

#[cfg(test)]
mod tests {
    use super::*;
    use arrow::array::{AsArray, Int64Array, StringArray};
    use arrow::datatypes::{DataType, Field, Schema};
    use std::sync::Arc;

    fn make_batch() -> RecordBatch {
        let schema = Arc::new(Schema::new(vec![
            Field::new("order_id", DataType::Utf8, true),
            Field::new("customer_id", DataType::Utf8, true),
            Field::new("amount", DataType::Int64, true),
        ]));
        let order_id = StringArray::from(vec!["O001", "O002", "O003", "O004"]);
        let customer_id = StringArray::from(vec!["X001", "X002", "X003", "X004"]);
        let amount = Int64Array::from(vec![100, 200, 300, 400]);
        RecordBatch::try_new(
            schema,
            vec![Arc::new(order_id), Arc::new(customer_id), Arc::new(amount)],
        )
        .unwrap()
    }

    fn make_fk_pool() -> RecordBatch {
        let schema = Arc::new(Schema::new(vec![Field::new(
            "customer_id",
            DataType::Utf8,
            true,
        )]));
        let vals = StringArray::from(vec![
            "C001", "C002", "C003", "C004", "C005", "C006", "C007", "C008", "C009", "C010",
        ]);
        RecordBatch::try_new(schema, vec![Arc::new(vals)]).unwrap()
    }

    #[test]
    fn test_fk_remap_basic() {
        let batch = make_batch();
        let pool = make_fk_pool();
        let mut rng = Rng::new(42);

        let (remapped, target_rids) =
            fk_remap_batch(&batch, &pool, "customer_id", &mut rng, false).unwrap();
        assert_eq!(remapped.num_rows(), 4);
        assert_eq!(remapped.num_columns(), 3);
        assert!(target_rids.is_none());

        // Check that customer_id values now come from the pool
        let col = remapped
            .column_by_name("customer_id")
            .unwrap()
            .as_string::<i32>();
        for i in 0..4 {
            let val = col.value(i);
            assert!(val.starts_with("C"), "expected pool value, got {val}");
            assert_ne!(val, "X001", "should not retain original synthetic value");
        }

        // order_id and amount should be unchanged
        let oid = remapped
            .column_by_name("order_id")
            .unwrap()
            .as_string::<i32>();
        assert_eq!(oid.value(0), "O001");
        assert_eq!(oid.value(3), "O004");

        let amt = remapped
            .column_by_name("amount")
            .unwrap()
            .as_any()
            .downcast_ref::<Int64Array>()
            .unwrap();
        assert_eq!(amt.value(0), 100);
    }

    #[test]
    fn test_fk_remap_deterministic() {
        let batch = make_batch();
        let pool = make_fk_pool();
        let mut rng_a = Rng::new(42);
        let mut rng_b = Rng::new(42);

        let a = fk_remap_batch(&batch, &pool, "customer_id", &mut rng_a, false).unwrap();
        let b = fk_remap_batch(&batch, &pool, "customer_id", &mut rng_b, false).unwrap();

        let ca =
            a.0.column_by_name("customer_id")
                .unwrap()
                .as_string::<i32>();
        let cb =
            b.0.column_by_name("customer_id")
                .unwrap()
                .as_string::<i32>();
        for i in 0..4 {
            assert_eq!(ca.value(i), cb.value(i), "mismatch at {i}");
        }
    }

    #[test]
    fn test_fk_remap_empty_pool() {
        let batch = make_batch();
        let schema = Arc::new(Schema::new(vec![Field::new(
            "customer_id",
            DataType::Utf8,
            true,
        )]));
        let empty = RecordBatch::try_new(
            schema,
            vec![Arc::new(StringArray::from(Vec::<&str>::new()))],
        )
        .unwrap();
        let mut rng = Rng::new(0);
        let result = fk_remap_batch(&batch, &empty, "customer_id", &mut rng, false);
        assert!(result.is_err());
    }
}