use super::eigen::symmetric_eigen_via_ql;
use crate::domain::real::RealScalar;
use crate::domain::scalar::Scalar;
use crate::{
bidiagonalize, bunch_kaufman, cholesky_decompose, col_piv_qr, eigenvalues, full_piv_lu,
hessenberg, lu_decompose, qr_decompose, schur, svd_decompose, symmetric_eigenvalues_jacobi,
udu_decompose, BidiagonalDecomposition, BunchKaufmanDecomposition, CholeskyDecomposition,
ColPivQrDecomposition, FullPivLuDecomposition, HessenbergDecomposition, LuDecomposition,
QrDecomposition, RealSchur, SvdDecomposition, SymmetricEigenDecomposition, UduDecomposition,
};
use crate::{
det as det_kernel, inv as inv_kernel, kron as kron_kernel, matexp as matexp_kernel,
matmul as matmul_kernel, matpow as matpow_kernel, matrix_rank as matrix_rank_kernel,
matrix_rank_with_tolerance as matrix_rank_tol_kernel, norm_l1 as norm_l1_kernel,
norm_l2 as norm_l2_kernel, norm_max as norm_max_kernel, pinv as pinv_kernel,
singular_values as singular_values_kernel, solve as solve_kernel,
solve_least_squares as solve_least_squares_kernel, trace as trace_kernel,
};
use leto::Complex;
use leto::{Array, Array1, Array2, ArrayView, ArrayView1, ArrayView2, Result, Storage};
pub trait AsMatrixView<T> {
fn as_matrix_view(&self) -> ArrayView2<'_, T>;
}
impl<T> AsMatrixView<T> for ArrayView2<'_, T> {
#[inline]
fn as_matrix_view(&self) -> ArrayView2<'_, T> {
ArrayView::new(self.layout(), self.data())
}
}
impl<T, S: Storage<T>> AsMatrixView<T> for Array<T, S, 2> {
#[inline]
fn as_matrix_view(&self) -> ArrayView2<'_, T> {
self.view()
}
}
pub trait MatrixProduct<T: Scalar> {
fn matmul<R: AsMatrixView<T>>(&self, rhs: &R) -> Result<Array2<T>>;
fn kron<R: AsMatrixView<T>>(&self, rhs: &R) -> Result<Array2<T>>;
}
impl<T: Scalar, M: AsMatrixView<T>> MatrixProduct<T> for M {
#[inline]
fn matmul<R: AsMatrixView<T>>(&self, rhs: &R) -> Result<Array2<T>> {
let lhs = self.as_matrix_view();
let rhs = rhs.as_matrix_view();
let [rows, _] = lhs.shape();
let [_, cols] = rhs.shape();
let mut out = Array2::from_shape_vec([rows, cols], vec![T::ZERO; rows * cols])?;
matmul_kernel(&lhs, &rhs, &mut out.view_mut())?;
Ok(out)
}
#[inline]
fn kron<R: AsMatrixView<T>>(&self, rhs: &R) -> Result<Array2<T>> {
kron_kernel(&self.as_matrix_view(), &rhs.as_matrix_view())
}
}
pub trait MatrixNorm<T: RealScalar> {
fn norm_l1(&self) -> Result<T>;
fn norm_l2(&self) -> Result<T>;
fn norm_max(&self) -> Result<T>;
}
impl<T: RealScalar, M: AsMatrixView<T>> MatrixNorm<T> for M {
#[inline]
fn norm_l1(&self) -> Result<T> {
norm_l1_kernel(&self.as_matrix_view())
}
#[inline]
fn norm_l2(&self) -> Result<T> {
norm_l2_kernel(&self.as_matrix_view())
}
#[inline]
fn norm_max(&self) -> Result<T> {
norm_max_kernel(&self.as_matrix_view())
}
}
pub trait MatrixDecompose<T: RealScalar> {
fn lu(&self) -> Result<LuDecomposition<T>>;
fn full_piv_lu(&self) -> Result<FullPivLuDecomposition<T>>;
fn qr(&self) -> Result<QrDecomposition<T>>;
fn col_piv_qr(&self) -> Result<ColPivQrDecomposition<T>>;
fn cholesky(&self) -> Result<CholeskyDecomposition<T>>;
fn udu(&self) -> Result<UduDecomposition<T>>;
fn bunch_kaufman(&self) -> Result<BunchKaufmanDecomposition<T>>;
fn hessenberg(&self) -> Result<HessenbergDecomposition<T>>;
fn bidiagonalize(&self) -> Result<BidiagonalDecomposition<T>>;
fn svd(&self) -> Result<SvdDecomposition<T>>;
fn singular_values(&self) -> Result<Vec<T>>;
fn symmetric_eigen(&self) -> Result<SymmetricEigenDecomposition<T>>;
fn symmetric_eigenvalues(&self) -> Result<Vec<T>>;
fn eigenvalues(&self) -> Result<Vec<Complex<T>>>;
fn schur(&self) -> Result<RealSchur<T>>;
}
impl<T: RealScalar, M: AsMatrixView<T>> MatrixDecompose<T> for M {
#[inline]
fn lu(&self) -> Result<LuDecomposition<T>> {
lu_decompose(&self.as_matrix_view())
}
#[inline]
fn full_piv_lu(&self) -> Result<FullPivLuDecomposition<T>> {
full_piv_lu(&self.as_matrix_view())
}
#[inline]
fn qr(&self) -> Result<QrDecomposition<T>> {
qr_decompose(&self.as_matrix_view())
}
#[inline]
fn col_piv_qr(&self) -> Result<ColPivQrDecomposition<T>> {
col_piv_qr(&self.as_matrix_view())
}
#[inline]
fn cholesky(&self) -> Result<CholeskyDecomposition<T>> {
cholesky_decompose(&self.as_matrix_view())
}
#[inline]
fn udu(&self) -> Result<UduDecomposition<T>> {
udu_decompose(&self.as_matrix_view())
}
#[inline]
fn bunch_kaufman(&self) -> Result<BunchKaufmanDecomposition<T>> {
bunch_kaufman(&self.as_matrix_view())
}
#[inline]
fn hessenberg(&self) -> Result<HessenbergDecomposition<T>> {
hessenberg(&self.as_matrix_view())
}
#[inline]
fn bidiagonalize(&self) -> Result<BidiagonalDecomposition<T>> {
bidiagonalize(&self.as_matrix_view())
}
#[inline]
fn svd(&self) -> Result<SvdDecomposition<T>> {
svd_decompose(&self.as_matrix_view())
}
#[inline]
fn singular_values(&self) -> Result<Vec<T>> {
singular_values_kernel(&self.as_matrix_view())
}
#[inline]
fn symmetric_eigen(&self) -> Result<SymmetricEigenDecomposition<T>> {
symmetric_eigen_via_ql(&self.as_matrix_view())
}
#[inline]
fn symmetric_eigenvalues(&self) -> Result<Vec<T>> {
symmetric_eigenvalues_jacobi(&self.as_matrix_view())
}
#[inline]
fn eigenvalues(&self) -> Result<Vec<Complex<T>>> {
eigenvalues(&self.as_matrix_view())
}
#[inline]
fn schur(&self) -> Result<RealSchur<T>> {
schur(&self.as_matrix_view())
}
}
pub trait MatrixSolve<T: RealScalar> {
fn solve(&self, rhs: &ArrayView1<'_, T>) -> Result<Array1<T>>;
fn solve_least_squares(&self, rhs: &ArrayView1<'_, T>) -> Result<Array1<T>>;
fn inv(&self) -> Result<Array2<T>>;
fn det(&self) -> Result<T>;
fn pinv(&self) -> Result<Array2<T>>;
}
impl<T: RealScalar, M: AsMatrixView<T>> MatrixSolve<T> for M {
#[inline]
fn solve(&self, rhs: &ArrayView1<'_, T>) -> Result<Array1<T>> {
solve_kernel(&self.as_matrix_view(), rhs)
}
#[inline]
fn solve_least_squares(&self, rhs: &ArrayView1<'_, T>) -> Result<Array1<T>> {
solve_least_squares_kernel(&self.as_matrix_view(), rhs)
}
#[inline]
fn inv(&self) -> Result<Array2<T>> {
inv_kernel(&self.as_matrix_view())
}
#[inline]
fn det(&self) -> Result<T> {
det_kernel(&self.as_matrix_view())
}
#[inline]
fn pinv(&self) -> Result<Array2<T>> {
pinv_kernel(&self.as_matrix_view())
}
}
pub trait MatrixProperties<T: RealScalar> {
fn trace(&self) -> Result<T>;
fn rank(&self) -> Result<usize>;
fn rank_with_tolerance(&self, relative_tolerance: T) -> Result<usize>;
}
impl<T: RealScalar, M: AsMatrixView<T>> MatrixProperties<T> for M {
#[inline]
fn trace(&self) -> Result<T> {
trace_kernel(&self.as_matrix_view())
}
#[inline]
fn rank(&self) -> Result<usize> {
matrix_rank_kernel(&self.as_matrix_view())
}
#[inline]
fn rank_with_tolerance(&self, relative_tolerance: T) -> Result<usize> {
matrix_rank_tol_kernel(&self.as_matrix_view(), relative_tolerance)
}
}
pub trait MatrixFunction<T: RealScalar> {
fn matpow(&self, exponent: u32) -> Result<Array2<T>>;
fn matexp(&self) -> Result<Array2<T>>;
}
impl<T: RealScalar, M: AsMatrixView<T>> MatrixFunction<T> for M {
#[inline]
fn matpow(&self, exponent: u32) -> Result<Array2<T>> {
matpow_kernel(&self.as_matrix_view(), exponent)
}
#[inline]
fn matexp(&self) -> Result<Array2<T>> {
matexp_kernel(&self.as_matrix_view())
}
}