Rust Numpy library - iterate by rows: unable to build a NpySingleIterBuilder::readwrite that returns rows instead of single values

Question

I am trying to iterate by rows over a Numpy Array. The array is accessed thru PyO3 and I think the library acess to the underlying C object just fine but I can't seem to find the reference for a more complex SingleIteratorBuilder that helps me access to the array with by rows.

This is the documentation page: https://docs.rs/numpy/0.12.1/numpy/npyiter/struct.NpySingleIterBuilder.html#method.readwrite (I see the project is still on its infancy)

This is my code in rust that is compiled into a python module

#[macro_use]
extern crate std;
extern crate ndarray;
use pyo3::prelude::*;
use pyo3::wrap_pyfunction;
use numpy::*;

#[pyfunction]
fn array_print(_py: Python<'_>,  matrix1: &PyArray2<i32> ){
     let matrix1_iter = NpySingleIterBuilder::readonly(matrix1.readonly()).build().unwrap();
 
    for i in matrix1_iter{
        println!("{}", i);
    };
}

/// A Python module implemented in Rust.
#[pymodule]
fn algo_match(_py: Python<'_>, m: &PyModule) -> PyResult<()> { 
    m.add_function(wrap_pyfunction!(array_print, m)?)?;
    Ok(())
}

And the python code for this case...

#matrixes is the name of the compiled Rust module
import matrixes as am 

arr1 = np.array([[1, 2, 3], [4, 5, 6], [1, 2, 3], [4, 5, 6]], ndmin=2)
am.array_print(arr1)

that returns what I was just saying

~project>python use_algorithm.py
1
2
3
4
5
6
1
2
3
4
5
6

Of course I have tried to use ndarray and have the entire pyArray copied into a new object but that is the worse case scenario because my matrixes are supposed to be large both on rows and columns. Duplicating data is perhaps the worse option.

A minor correction: do you want to iterate over rows? It's not called iterating "by" rows. — Peter Blackson, Dec 08 '20 at 12:36
Yes I want to iterate by rows. Where shall I change? In the title? — Matteo Montanari, Dec 08 '20 at 14:58

score 1 · Accepted Answer · answered Dec 08 '20 at 13:05

You can access .as_slice() if the array is contiguous. With the matrix viewed as a simple slice, you can iterate over rows with .chunks(n).

This will only be easy for iterating over rows. For columns, you may need itertools.

#[macro_use]
extern crate std;
extern crate ndarray;
use numpy::*;
use pyo3::prelude::*;
use pyo3::wrap_pyfunction;

#[pyfunction]
fn array_print(_py: Python<'_>, matrix1: &PyArray2<i64>) {
    let row_length = matrix1.shape()[1];
    for row in matrix1.readonly().as_slice().unwrap().chunks(row_length) {
        println!("{:?}", row);
    }
}

/// A Python module implemented in Rust.
#[pymodule]
fn matrixes(_py: Python<'_>, m: &PyModule) -> PyResult<()> {
    m.add_function(wrap_pyfunction!(array_print, m)?)?;
    Ok(())
}

I have tested it and it working like I wanted – Matteo Montanari Dec 08 '20 at 15:14 — Matteo Montanari, Dec 08 '20 at 15:14

Rust Numpy library - iterate by rows: unable to build a NpySingleIterBuilder::readwrite that returns rows instead of single values

1 Answers1