Merge branch 'rust-ml:main' into main

.github/workflows/checking.yml

@@ -10,7 +10,7 @@ jobs:
       fail-fast: false
       matrix:
         toolchain: 
-          - 1.54.0
+          - 1.65.0
           - stable
           - nightly
         os:

.github/workflows/codequality.yml

@@ -10,7 +10,7 @@ jobs:
     strategy:
       matrix:
         toolchain:
-          - 1.54.0
+          - 1.65.0
           - stable
 
     steps:

.github/workflows/testing.yml

@@ -11,11 +11,11 @@ jobs:
       fail-fast: false
       matrix:
         toolchain:
-          - 1.54.0
+          - 1.65.0
           - stable
         os:
-          - ubuntu-18.04
-          - windows-2019
+          - ubuntu-latest
+          - windows-latest
 
     steps:
       - name: Checkout sources

CHANGELOG.md

@@ -0,0 +1,7 @@
+# Changelog
+
+## [0.2.0] - 2024-08-28
+
+### Changed
+
+- Update ndarray to 0.16 along with associated dependencies

Cargo.toml

@@ -1,27 +1,28 @@
 [package]
 name = "linfa-linalg"
-version = "0.1.0"
+version = "0.2.0"
 edition = "2018"
 authors = ["Yuhan Lin <yuhanliin@protonmail.com>"]
 license = "MIT/Apache-2.0"
 readme = "README.md"
 description = "Pure-Rust implementation of linear algebra routines for ndarray"
 repository = "https://github.com/rust-ml/linfa-linalg"
+rust-version = "1.65"
 
 keywords = ["ndarray", "matrix", "linalg"]
 categories = ["algorithms", "mathematics", "science"]
 
 [dependencies]
-ndarray = { version = "0.15", features = ["approx"] }
+ndarray = { version = "0.16", features = ["approx"] }
 num-traits = "0.2.0"
 thiserror = "1"
 rand = { version = "0.8", optional=true }
 
 [dev-dependencies]
-approx = "0.4"
+approx = "0.5"
 proptest = "1.0"
-proptest-derive = "0.2.0"
-ndarray-rand = "0.14"
+proptest-derive = "0.5.0"
+ndarray-rand = "0.15"
 rand_xoshiro = { version = "0.6" }
 
 [features]

src/lib.rs

@@ -10,6 +10,7 @@
 //! linker errors if no BLAS backend is specified.
 
 #![allow(clippy::many_single_char_names)]
+#![allow(clippy::result_large_err)]
 
 pub mod bidiagonal;
 pub mod cholesky;

src/lobpcg/algorithm.rs

@@ -1,10 +1,10 @@
+//! Locally Optimal Block Preconditioned Conjugated
+//!
+//! This module implements the Locally Optimal Block Preconditioned Conjugated (LOBPCG) algorithm,
+//!  which can be used as a solver for large symmetric eigenproblems.
 use ndarray::concatenate;
 use ndarray::prelude::*;
 use num_traits::NumCast;
-///! Locally Optimal Block Preconditioned Conjugated
-///!
-///! This module implements the Locally Optimal Block Preconditioned Conjugated (LOBPCG) algorithm,
-///which can be used as a solver for large symmetric eigenproblems.
 use std::iter::Sum;
 
 use crate::{cholesky::*, eigh::*, norm::*, triangular::*};
@@ -100,15 +100,15 @@ fn orthonormalize<T: NdFloat>(v: Array2<T>) -> Result<(Array2<T>, Array2<T>)> {
 ///
 /// # Arguments
 /// * `a` - An operator defining the problem, usually a sparse (sometimes also dense) matrix
-/// multiplication. Also called the "stiffness matrix".
+///   multiplication. Also called the "stiffness matrix".
 /// * `x` - Initial approximation of the k eigenvectors. If `a` has shape=(n,n), then `x` should
-/// have shape=(n,k).
+///   have shape=(n,k).
 /// * `m` - Preconditioner to `a`, by default the identity matrix. Should approximate the inverse
-/// of `a`.
+///   of `a`.
 /// * `y` - Constraints of (n,size_y), iterations are performed in the orthogonal complement of the
-/// column-space of `y`. It must be full rank.
+///   column-space of `y`. It must be full rank.
 /// * `tol` - The tolerance values defines at which point the solver stops the optimization. The approximation
-/// of a eigenvalue stops when then l2-norm of the residual is below this threshold.
+///   of a eigenvalue stops when then l2-norm of the residual is below this threshold.
 /// * `maxiter` - The maximal number of iterations
 /// * `order` - Whether to solve for the largest or lowest eigenvalues
 ///

src/lobpcg/eig.rs

@@ -274,7 +274,7 @@ impl<A: NdFloat + Sum, R: Rng> Iterator for TruncatedEigIterator<A, R> {
                     let eigvecs_arr: Vec<_> = constraints
                         .columns()
                         .into_iter()
-                        .chain(eigvecs.columns().into_iter())
+                        .chain(eigvecs.columns())
                         .collect();
 
                     stack(Axis(1), &eigvecs_arr).unwrap()

src/lobpcg/mod.rs

@@ -7,7 +7,7 @@
 //! ```
 //! where A is symmetric and (x, lambda) the solution. It has the following advantages:
 //! * matrix free: does not require storing the coefficient matrix explicitely and only evaluates
-//! matrix-vector products.
+//!   matrix-vector products.
 //! * factorization-free: does not require any matrix decomposition
 //! * linear-convergence: theoretically guaranteed and practically observed
 //!

src/lobpcg/svd.rs

@@ -1,6 +1,6 @@
-///! Truncated singular value decomposition
-///!
-///! This module computes the k largest/smallest singular values/vectors for a dense matrix.
+//! Truncated singular value decomposition
+//!
+//! This module computes the k largest/smallest singular values/vectors for a dense matrix.
 use crate::{
     lobpcg::{lobpcg, random, Lobpcg},
     Order, Result,