Best way to emulate NumPy index arrays

[Tudyx]

As stated in the doc NumPy index arrays is not supported yet.
I have a multidimensional array and i want to emulate the result of NumPy index array. For instance, if my data is multi dimensional i want to acces a part of it only by providing a single index, for instance with data[0]. It's ok for me to not have the syntactic sugar [] of the index trait.
Could you please give me some advise about functions in the ndarray crate i could use to try to emulate that?

[CarlKCarlK]

Not sure this is exactly on point, but I'm working on supporting fancy indexing within a new Rust project. I'm porting Bed-Reader, our genomics program, from Python to Rust. The program reads 2-D data from a format called PLINK Bed files. The data can be bigger than 1 million individuals (rows) by 1 million DNA locations (columns). When the file is large, users want read just a slice at a time.

The solution I'm creating supports indexing by nothing ("read all"), a number, a negative number (read from end), a vector or array of indexes (including negatives), any Rust range, any ndarray slice, Boolean vectors or arrays. I've put some examples below.

The approach uses a builder pattern. Behind the scenes, it automatically converts all the supported indexing methods into an enum called Index. That gets converted to a vector of indexes when applied to an ndarray of known size.

The code is open source. Let me know if there is interest making this its own crate.

• Carl

Read individual (samples) from 20 to 30 and every second SNP (variant).

use bed_reader::ReadOptions;
use ndarray::s;

let file_name = "bed_reader/tests/data/some_missing.bed";
let mut bed = Bed::new(file_name)?;
let val = ReadOptions::builder()
    .iid_index(s![..;2])
    .sid_index(20..30)
    .f64()
    .read(&mut bed)?;

assert!(val.dim() == (50, 10));

Read every value in chromosome 5.

let mut bed = Bed::new(file_name)?;
let val = ReadOptions::builder()
    .sid_index(bed.chromosome()?.map(|elem| elem == "5"))
    .f64()
    .read(&mut bed)?;

assert!(val.dim() == (100, 6));

use ndarray as nd;
use bed_reader::Bed;
use bed_reader::assert_eq_nan;
use bed_reader::ReadOptions;
use ndarray::s;

let file_name = "bed_reader/tests/data/some_missing.bed";
let mut bed = Bed::new(file_name)?;

// Read the individual at index position 3

let val = ReadOptions::builder()
    .iid_index(3)
    .f64()
    .read(&mut bed)?;
assert!(val.dim() == (1, 100));

// Read the individuals at index positions 0, 5, and 1st-from-last.

let val = ReadOptions::builder()
    .iid_index([0, 5, -1])
    .f64()
    .read(&mut bed)?;

assert!(val.dim() == (3, 100));

// Read the individuals at index positions 20 (inclusive) to 30 (exclusive).

let val = ReadOptions::builder()
    .iid_index(20..30)
    .f64()
    .read(&mut bed)?;

assert!(val.dim() == (10, 100));

// Read the individuals at every 2nd index position.

let val = ReadOptions::builder()
    .iid_index(s![..;2])
    .f64()
    .read(&mut bed)?;

assert!(val.dim() == (50, 100));

// Read chromosome 5 of the female individuals.

let female = bed.sex()?.map(|elem| *elem == 2);
let chrom_5 = bed.chromosome()?.map(|elem| elem == "5");
let val = ReadOptions::builder()
    .iid_index(female)
    .sid_index(chrom_5)
    .f64()
    .read(&mut bed)?;

assert!(val.dim() == (50, 6));

[Tudyx]

Thanks a lot for your proposal! The Bed-Reader seems to be a very interesting project.
For my problem, is solve it using the index_axis method. It's seems to work find for now.

use ndarray::Axis;
// Emulate the NumPy data[index]
data.index_axis(Axis(0), index)