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

// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright The Lance Authors

use arrow::{
    array::AsArray,
    datatypes::{Float16Type, Float32Type, Float64Type},
};
use arrow_array::Array;

use arrow_schema::DataType;
use lance_core::{Error, Result};
use lance_linalg::distance::MetricType;
use snafu::{location, Location};

use super::ScalarQuantizer;

#[derive(Debug, Clone)]
pub struct SQBuildParams {
    /// Number of bits of scaling range.
    pub num_bits: u16,

    /// Sample rate for training.
    pub sample_rate: usize,
}

impl Default for SQBuildParams {
    fn default() -> Self {
        Self {
            num_bits: 8,
            sample_rate: 256,
        }
    }
}

impl SQBuildParams {
    pub fn build(&self, data: &dyn Array, metric_type: MetricType) -> Result<ScalarQuantizer> {
        let fsl = data.as_fixed_size_list_opt().ok_or(Error::Index {
            message: format!(
                "SQ builder: input is not a FixedSizeList: {}",
                data.data_type()
            ),
            location: location!(),
        })?;

        let mut quantizer =
            ScalarQuantizer::new(self.num_bits, fsl.value_length() as usize, metric_type);

        match fsl.value_type() {
            DataType::Float16 => {
                quantizer.update_bounds::<Float16Type>(fsl)?;
            }
            DataType::Float32 => {
                quantizer.update_bounds::<Float32Type>(fsl)?;
            }
            DataType::Float64 => {
                quantizer.update_bounds::<Float64Type>(fsl)?;
            }
            _ => {
                return Err(Error::Index {
                    message: format!("SQ builder: unsupported data type: {}", fsl.value_type()),
                    location: location!(),
                })
            }
        }

        Ok(quantizer)
    }
}