getting no method named `svd` found for reference `&ArrayBase<OwnedRepr<T>, Dim<[usize; 2]>>`

[emiddleton]

Not sure if this is the write place to ask such questions but I couldn't find anywhere else more specific. I am trying to implement the (Moore-Penrose) pseudo-inverse of a matrix[1]. My code currently looks like

use ndarray::prelude::*;
use ndarray_linalg::*;
use std::cmp::min;
use thiserror::Error;

#[non_exhaustive]
#[derive(Error, Debug)]
pub enum Error {
    #[error("svd failed")]
    SvdFailed,
}

pub fn pinv(a: &Array2<f64>) -> Result<Array2<f64>, Error> {
    //a.conjugate()
    let (n, m) = a.dim();
    let result: (Option<Array1<f64>>, Array1<f64>, Option<Array1<f64>>) =
        a.svd(true, true).map_err(Error::SvdFailed)?;

I am running into issues getting

error[E0599]: no method named `svd` found for reference `&ArrayBase<OwnedRepr<T>, Dim<[usize; 2]>>` in the current scope
  --> numrs/src/linalg.rs:16:43
   |
16 |     let (u, s, vt): (_, Array1<_>, _) = a.svd(true, true).unwrap();
   |                                           ^^^ help: there is an associated function with a similar name: `std`

the dependencies part of my Cargo.toml file is the same versions as the ndarray-linalg crate on github[2]

[dependencies]
num-complex = "0.4.0"
num-traits  = "0.2.14"
rand = "0.8.3"
thiserror = "1.0.24"

[dependencies.ndarray]
version = "0.15.1"
features = ["blas", "approx", "std"]
default-features = false

#ndarray-linalg = { version = "0.13.1", features = ["openblas-static"] }
[dependencies.ndarray-linalg]
github = "https://github.com/rust-ndarray/ndarray-linalg/ndarray-linalg"
version = "0.13.1"
features = ["openblas-static"]

[dependencies.lax]
version = "0.1.0"
github = "https://github.com/rust-ndarray/ndarray-linalg/lax"
default-features = false

Any idea's what is going on here. I downloaded the ndarray-linalg source from github and the test run fine with

$ cargo test --features=openblas svd

1. https://github.com/numpy/numpy/blob/v1.20.0/numpy/linalg/linalg.py#L1915-L2011
2. https://github.com/rust-ndarray/ndarray-linalg/blob/master/ndarray-linalg/Cargo.toml

[jturner314]

The versions of ndarray and ndarray-linalg in the Cargo.toml are incompatible. The latest version of ndarray-linalg (0.13.1) provides functionality for ndarray 0.14. Hopefully we'll get a chance to release a new version of ndarray-linalg for ndarray 0.15 soon (#281). Try swiching to ndarray 0.14 for now. By the way, thanks for writing such a detailed, easy to understand issue. :)

[emiddleton]

Thanks, Is there any way to run locally with the unreleased version of ndarray-linalg. That was what I was trying to do as I am using some features in ndarray 1.15.

Also is there anything I can do to help get a 0.15 version released? Is there a list pre-0.15 release issues?

[jturner314]

I'm not sure about the status of the next version of ndarray-linalg. My impression is that it was blocked by #277, but I could be wrong. @termoshtt would know more. Something like this would work temporarily:

[dependencies]
ndarray = "0.15"
ndarray-linalg = { git = "https://github.com/rust-ndarray/ndarray-linalg.git", branch = "use-build-uninit", features = ["openblas"] }

This specifies the branch from #287 and the OpenBLAS backend, but you could choose a different version or backend. See the Cargo docs for more information.

[emiddleton]

Just as a follow up on this, I was not able to get this working without explicitly downloading the source from github and including it by path. i.e.

[dependencies.ndarray]
version = "=0.15.1"
features = ["blas", "approx", "std"]
default-features = false

[dependencies.ndarray-linalg]
path = "../ndarray-linalg/ndarray-linalg"
features = ["openblas-static"]

[dependencies.lax]
path = "../ndarray-linalg/lax"
default-features = false

[tbetcke]

Apologies to add to this issue. Is there an approximate timeline for the release of the next version of ndarray-linalg? For a project I have already written a couple of components using ndarray 0.15 and now notice that I should have used ndarray 0.14 for ndarray-linalg compatibility. The question is whether to downgrade 0.15 specific library calls to 0.14 compatible versions in my already written code or to simply wait until the next ndarray-linalg version is released.