This document will be used to keep track of changes made between release versions. I'll do my best to note any breaking changes!
Matrix::lup_decomp has been deprecated and will be removed soon.PermutationMatrix. This type provides more efficent operations with permutation matrices.Matrix. Under io feature flag.BaseMatrix::col_iter and BaseMatrixMut::col_iter_mut functions.PartialPivLU struct which contains the result of an LUP decomposition. This struct will replace the Matrix::lup_decomp function in a future release.SliceIter.min/max functions.vector! macro for Vector construction.min and max functions to the BaseMatrix trait.Row/Column types to Vector.BaseMatrix::sum_rows function. also gives passive improvement to mean and variance.vector module.matrix! macro use square brackets consistently.This release includes mostly quality of life changes for users of rulinalg. We do some work to conform more to community standards with naming, for example iter_rows becoming row_iter. Additionally several Matrix functions now consume self where before they took a reference and immediately cloned self.
Another noticable change is the addition of new Row and Column types. These types are returned by functions which access single rows or columns in a matrix. With these new types we aim to make it easy for users to do matrix operations on single rows and columns while maintaining performance where necessary.
This release also welcomes an overhaul of the Metric trait. This trait didn't really make sense and only allowed computation of the euclidean norm. We have created new Norm and Metric traits for both Vector and matrix types (we would like a single pair of traits but this is not possible without specialization). These new traits allow users to write code which is generic over norms and specify their own norms. We also provide Metric implementations for all Norm implementors by computing the norm of the difference between the objects.
The full release notes are below.
reslice function for MatrixSlice and MatrixSliceMut has been depreciated.*_iter(_mut). Affected functions are: iter_diag, iter_diag_mut, iter_rows, iter_rows_mut.BaseMatrix diag function now returns an iterator.Metric trait and all implementations.self instead of cloning internally: eigenvalues, eigendecomp, lup_decomp, solve, inverse and det.get_row no longer returns a &[T]. Instead it returns the new Row type.&[T] item. Instead if uses the new Row type.BaseMatrix and BaseMatrixMut traits to a new matrix/base module.FromIterator for Vector.from_fn for Vector.get_unchecked for Vector.try_into function using num's NumCast for Matrix.Metric; MatrixNorm and VectorNorm. These come with MatrixMetric and VectorMetric traits too.Euclidean and Lp norms.get_row functions now return the new Row type.get_col function which returns the new Column type.row_iter function uses the new Row type as the iterator Item.swap_rows and swap_cols functions are now no-ops if given two identical indices.slice module for developer QOL.assert_matrix_eq! and assert_vector_eq! macros for easier equality checks. Provides multiple equality comparisons: ulp, abs, float, exact.lu module.iter_diag and iter_diag_mut functions to BaseMatrix and BaseMatrixMut respectively.matrix! macro now works on empty matrices.decomposition module.assert! that a matrix is triangular.matrix! macro and other tidier constructors.MachineEpsilon trait. The same functionality now exists in num.From/Into for traits for Vec and Vector.det() now returns 0 instead of panicking if Matrix is singular.matrix! macro for creating (small) matrices.from_fn constructor for Matrix.IndexMut for Vector.iter and iter_mut for Vector.IntoIter for Vector.Matrix::variance now returns a Result.swap_rows and swap_cols function to BaseMatrixMut.Display for Vector.sum_rows and sum_cols functions.select_rows and select_cols. These functions now take an ExactSizeIterator instead of a slice.This is a large release which refactors most of the matrix module. We modify the BaseSlice trait to encompass Matrix functionality too - hence renaming it BaseMatrix. The motivation behind this is to allow us to be generic over Matrix/MatrixSlice/MatrixSliceMut.
BaseSlice trait as BaseMatrix. Implement this trait for Matrix too.Matrix functionality is now implemented behind the BaseMatrix trait. It will need to be used to access this functionality.BaseMatrixMut trait for Matrix and MatrixSliceMut.Matrix only or for MatrixSlice(Mut) only now work with both!sub_slice method.-None
Eq trait.Hash trait.get_row_* methods for all Matrix types. Includes mutable and unchecked get functions.Matrix (instead of cloning at the start).get_unchecked_mut returned &T instead of &mut T.This release marks the separation of rulinalg from rusty-machine.
Rulinalg is now its own crate!