Merge pull request #613 from russellb23/dev

QR factorizatio nwith column pivoting
This commit is contained in:
Sébastien Crozet 2021-02-25 15:45:58 +01:00 committed by GitHub
commit 06f92ad1e3
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 516 additions and 6 deletions

333
src/linalg/col_piv_qr.rs Normal file
View File

@ -0,0 +1,333 @@
use num::Zero;
#[cfg(feature = "serde-serialize")]
use serde::{Deserialize, Serialize};
use crate::allocator::{Allocator, Reallocator};
use crate::base::{DefaultAllocator, Matrix, MatrixMN, MatrixN, Unit, VectorN};
use crate::constraint::{SameNumberOfRows, ShapeConstraint};
use crate::dimension::{Dim, DimMin, DimMinimum, U1};
use crate::storage::{Storage, StorageMut};
use crate::ComplexField;
use crate::geometry::Reflection;
use crate::linalg::{householder, PermutationSequence};
/// The QR decomposition (with column pivoting) of a general matrix.
#[cfg_attr(feature = "serde-serialize", derive(Serialize, Deserialize))]
#[cfg_attr(
feature = "serde-serialize",
serde(bound(serialize = "DefaultAllocator: Allocator<N, R, C> +
Allocator<N, DimMinimum<R, C>>,
MatrixMN<N, R, C>: Serialize,
PermutationSequence<DimMinimum<R, C>>: Serialize,
VectorN<N, DimMinimum<R, C>>: Serialize"))
)]
#[cfg_attr(
feature = "serde-serialize",
serde(bound(deserialize = "DefaultAllocator: Allocator<N, R, C> +
Allocator<N, DimMinimum<R, C>>,
MatrixMN<N, R, C>: Deserialize<'de>,
PermutationSequence<DimMinimum<R, C>>: Deserialize<'de>,
VectorN<N, DimMinimum<R, C>>: Deserialize<'de>"))
)]
#[derive(Clone, Debug)]
pub struct ColPivQR<N: ComplexField, R: DimMin<C>, C: Dim>
where
DefaultAllocator: Allocator<N, R, C>
+ Allocator<N, DimMinimum<R, C>>
+ Allocator<(usize, usize), DimMinimum<R, C>>,
{
col_piv_qr: MatrixMN<N, R, C>,
p: PermutationSequence<DimMinimum<R, C>>,
diag: VectorN<N, DimMinimum<R, C>>,
}
impl<N: ComplexField, R: DimMin<C>, C: Dim> Copy for ColPivQR<N, R, C>
where
DefaultAllocator: Allocator<N, R, C>
+ Allocator<N, DimMinimum<R, C>>
+ Allocator<(usize, usize), DimMinimum<R, C>>,
MatrixMN<N, R, C>: Copy,
PermutationSequence<DimMinimum<R, C>>: Copy,
VectorN<N, DimMinimum<R, C>>: Copy,
{
}
impl<N: ComplexField, R: DimMin<C>, C: Dim> ColPivQR<N, R, C>
where
DefaultAllocator: Allocator<N, R, C>
+ Allocator<N, R>
+ Allocator<N, DimMinimum<R, C>>
+ Allocator<(usize, usize), DimMinimum<R, C>>,
{
/// Computes the ColPivQR decomposition using householder reflections.
pub fn new(mut matrix: MatrixMN<N, R, C>) -> Self {
let (nrows, ncols) = matrix.data.shape();
let min_nrows_ncols = nrows.min(ncols);
let mut p = PermutationSequence::identity_generic(min_nrows_ncols);
let mut diag = unsafe { MatrixMN::new_uninitialized_generic(min_nrows_ncols, U1) };
if min_nrows_ncols.value() == 0 {
return ColPivQR {
col_piv_qr: matrix,
p,
diag,
};
}
for i in 0..min_nrows_ncols.value() {
let piv = matrix.slice_range(i.., i..).icamax_full();
let col_piv = piv.1 + i;
matrix.swap_columns(i, col_piv);
p.append_permutation(i, col_piv);
householder::clear_column_unchecked(&mut matrix, &mut diag[i], i, 0, None);
}
ColPivQR {
col_piv_qr: matrix,
p,
diag,
}
}
/// Retrieves the upper trapezoidal submatrix `R` of this decomposition.
#[inline]
pub fn r(&self) -> MatrixMN<N, DimMinimum<R, C>, C>
where
DefaultAllocator: Allocator<N, DimMinimum<R, C>, C>,
{
let (nrows, ncols) = self.col_piv_qr.data.shape();
let mut res = self
.col_piv_qr
.rows_generic(0, nrows.min(ncols))
.upper_triangle();
res.set_partial_diagonal(self.diag.iter().map(|e| N::from_real(e.modulus())));
res
}
/// Retrieves the upper trapezoidal submatrix `R` of this decomposition.
///
/// This is usually faster than `r` but consumes `self`.
#[inline]
pub fn unpack_r(self) -> MatrixMN<N, DimMinimum<R, C>, C>
where
DefaultAllocator: Reallocator<N, R, C, DimMinimum<R, C>, C>,
{
let (nrows, ncols) = self.col_piv_qr.data.shape();
let mut res = self
.col_piv_qr
.resize_generic(nrows.min(ncols), ncols, N::zero());
res.fill_lower_triangle(N::zero(), 1);
res.set_partial_diagonal(self.diag.iter().map(|e| N::from_real(e.modulus())));
res
}
/// Computes the orthogonal matrix `Q` of this decomposition.
pub fn q(&self) -> MatrixMN<N, R, DimMinimum<R, C>>
where
DefaultAllocator: Allocator<N, R, DimMinimum<R, C>>,
{
let (nrows, ncols) = self.col_piv_qr.data.shape();
// NOTE: we could build the identity matrix and call q_mul on it.
// Instead we don't so that we take in account the matrix sparseness.
let mut res = Matrix::identity_generic(nrows, nrows.min(ncols));
let dim = self.diag.len();
for i in (0..dim).rev() {
let axis = self.col_piv_qr.slice_range(i.., i);
// TODO: sometimes, the axis might have a zero magnitude.
let refl = Reflection::new(Unit::new_unchecked(axis), N::zero());
let mut res_rows = res.slice_range_mut(i.., i..);
refl.reflect_with_sign(&mut res_rows, self.diag[i].signum());
}
res
}
/// Retrieves the column permutation of this decomposition.
#[inline]
pub fn p(&self) -> &PermutationSequence<DimMinimum<R, C>> {
&self.p
}
/// Unpacks this decomposition into its two matrix factors.
pub fn unpack(
self,
) -> (
MatrixMN<N, R, DimMinimum<R, C>>,
MatrixMN<N, DimMinimum<R, C>, C>,
PermutationSequence<DimMinimum<R, C>>,
)
where
DimMinimum<R, C>: DimMin<C, Output = DimMinimum<R, C>>,
DefaultAllocator: Allocator<N, R, DimMinimum<R, C>>
+ Reallocator<N, R, C, DimMinimum<R, C>, C>
+ Allocator<(usize, usize), DimMinimum<R, C>>,
{
(self.q(), self.r(), self.p)
}
#[doc(hidden)]
pub fn col_piv_qr_internal(&self) -> &MatrixMN<N, R, C> {
&self.col_piv_qr
}
/// Multiplies the provided matrix by the transpose of the `Q` matrix of this decomposition.
pub fn q_tr_mul<R2: Dim, C2: Dim, S2>(&self, rhs: &mut Matrix<N, R2, C2, S2>)
where
S2: StorageMut<N, R2, C2>,
{
let dim = self.diag.len();
for i in 0..dim {
let axis = self.col_piv_qr.slice_range(i.., i);
let refl = Reflection::new(Unit::new_unchecked(axis), N::zero());
let mut rhs_rows = rhs.rows_range_mut(i..);
refl.reflect_with_sign(&mut rhs_rows, self.diag[i].signum().conjugate());
}
}
}
impl<N: ComplexField, D: DimMin<D, Output = D>> ColPivQR<N, D, D>
where
DefaultAllocator:
Allocator<N, D, D> + Allocator<N, D> + Allocator<(usize, usize), DimMinimum<D, D>>,
{
/// Solves the linear system `self * x = b`, where `x` is the unknown to be determined.
///
/// Returns `None` if `self` is not invertible.
pub fn solve<R2: Dim, C2: Dim, S2>(
&self,
b: &Matrix<N, R2, C2, S2>,
) -> Option<MatrixMN<N, R2, C2>>
where
S2: StorageMut<N, R2, C2>,
ShapeConstraint: SameNumberOfRows<R2, D>,
DefaultAllocator: Allocator<N, R2, C2>,
{
let mut res = b.clone_owned();
if self.solve_mut(&mut res) {
Some(res)
} else {
None
}
}
/// Solves the linear system `self * x = b`, where `x` is the unknown to be determined.
///
/// If the decomposed matrix is not invertible, this returns `false` and its input `b` is
/// overwritten with garbage.
pub fn solve_mut<R2: Dim, C2: Dim, S2>(&self, b: &mut Matrix<N, R2, C2, S2>) -> bool
where
S2: StorageMut<N, R2, C2>,
ShapeConstraint: SameNumberOfRows<R2, D>,
{
assert_eq!(
self.col_piv_qr.nrows(),
b.nrows(),
"ColPivQR solve matrix dimension mismatch."
);
assert!(
self.col_piv_qr.is_square(),
"ColPivQR solve: unable to solve a non-square system."
);
self.q_tr_mul(b);
let solved = self.solve_upper_triangular_mut(b);
self.p.inv_permute_rows(b);
solved
}
// TODO: duplicate code from the `solve` module.
fn solve_upper_triangular_mut<R2: Dim, C2: Dim, S2>(
&self,
b: &mut Matrix<N, R2, C2, S2>,
) -> bool
where
S2: StorageMut<N, R2, C2>,
ShapeConstraint: SameNumberOfRows<R2, D>,
{
let dim = self.col_piv_qr.nrows();
for k in 0..b.ncols() {
let mut b = b.column_mut(k);
for i in (0..dim).rev() {
let coeff;
unsafe {
let diag = self.diag.vget_unchecked(i).modulus();
if diag.is_zero() {
return false;
}
coeff = b.vget_unchecked(i).unscale(diag);
*b.vget_unchecked_mut(i) = coeff;
}
b.rows_range_mut(..i)
.axpy(-coeff, &self.col_piv_qr.slice_range(..i, i), N::one());
}
}
true
}
/// Computes the inverse of the decomposed matrix.
///
/// Returns `None` if the decomposed matrix is not invertible.
pub fn try_inverse(&self) -> Option<MatrixN<N, D>> {
assert!(
self.col_piv_qr.is_square(),
"ColPivQR inverse: unable to compute the inverse of a non-square matrix."
);
// TODO: is there a less naive method ?
let (nrows, ncols) = self.col_piv_qr.data.shape();
let mut res = MatrixN::identity_generic(nrows, ncols);
if self.solve_mut(&mut res) {
Some(res)
} else {
None
}
}
/// Indicates if the decomposed matrix is invertible.
pub fn is_invertible(&self) -> bool {
assert!(
self.col_piv_qr.is_square(),
"ColPivQR: unable to test the invertibility of a non-square matrix."
);
for i in 0..self.diag.len() {
if self.diag[i].is_zero() {
return false;
}
}
true
}
/// Computes the determinant of the decomposed matrix.
pub fn determinant(&self) -> N {
let dim = self.col_piv_qr.nrows();
assert!(
self.col_piv_qr.is_square(),
"ColPivQR determinant: unable to compute the determinant of a non-square matrix."
);
let mut res = N::one();
for i in 0..dim {
res *= unsafe { *self.diag.vget_unchecked(i) };
}
res * self.p.determinant()
}
}

View File

@ -1,8 +1,8 @@
use crate::storage::Storage; use crate::storage::Storage;
use crate::{ use crate::{
Allocator, Bidiagonal, Cholesky, ComplexField, DefaultAllocator, Dim, DimDiff, DimMin, Allocator, Bidiagonal, Cholesky, ColPivQR, ComplexField, DefaultAllocator, Dim, DimDiff,
DimMinimum, DimSub, FullPivLU, Hessenberg, Matrix, Schur, SymmetricEigen, SymmetricTridiagonal, DimMin, DimMinimum, DimSub, FullPivLU, Hessenberg, Matrix, Schur, SymmetricEigen,
LU, QR, SVD, U1, SymmetricTridiagonal, LU, QR, SVD, U1,
}; };
/// # Rectangular matrix decomposition /// # Rectangular matrix decomposition
@ -13,8 +13,9 @@ use crate::{
/// | Decomposition | Factors | Details | /// | Decomposition | Factors | Details |
/// | -------------------------|---------------------|--------------| /// | -------------------------|---------------------|--------------|
/// | QR | `Q * R` | `Q` is an unitary matrix, and `R` is upper-triangular. | /// | QR | `Q * R` | `Q` is an unitary matrix, and `R` is upper-triangular. |
/// | QR with column pivoting | `Q * R * P⁻¹` | `Q` is an unitary matrix, and `R` is upper-triangular. `P` is a permutation matrix. |
/// | LU with partial pivoting | `P⁻¹ * L * U` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` is a permutation matrix. | /// | LU with partial pivoting | `P⁻¹ * L * U` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` is a permutation matrix. |
/// | LU with full pivoting | `P⁻¹ * L * U ~ Q⁻¹` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` and `Q` are permutation matrices. | /// | LU with full pivoting | `P⁻¹ * L * U * Q⁻¹` | `L` is lower-triangular with a diagonal filled with `1` and `U` is upper-triangular. `P` and `Q` are permutation matrices. |
/// | SVD | `U * Σ * Vᵀ` | `U` and `V` are two orthogonal matrices and `Σ` is a diagonal matrix containing the singular values. | /// | SVD | `U * Σ * Vᵀ` | `U` and `V` are two orthogonal matrices and `Σ` is a diagonal matrix containing the singular values. |
impl<N: ComplexField, R: Dim, C: Dim, S: Storage<N, R, C>> Matrix<N, R, C, S> { impl<N: ComplexField, R: Dim, C: Dim, S: Storage<N, R, C>> Matrix<N, R, C, S> {
/// Computes the bidiagonalization using householder reflections. /// Computes the bidiagonalization using householder reflections.
@ -60,6 +61,18 @@ impl<N: ComplexField, R: Dim, C: Dim, S: Storage<N, R, C>> Matrix<N, R, C, S> {
QR::new(self.into_owned()) QR::new(self.into_owned())
} }
/// Computes the QR decomposition (with column pivoting) of this matrix.
pub fn col_piv_qr(self) -> ColPivQR<N, R, C>
where
R: DimMin<C>,
DefaultAllocator: Allocator<N, R, C>
+ Allocator<N, R>
+ Allocator<N, DimMinimum<R, C>>
+ Allocator<(usize, usize), DimMinimum<R, C>>,
{
ColPivQR::new(self.into_owned())
}
/// Computes the Singular Value Decomposition using implicit shift. /// Computes the Singular Value Decomposition using implicit shift.
pub fn svd(self, compute_u: bool, compute_v: bool) -> SVD<N, R, C> pub fn svd(self, compute_u: bool, compute_v: bool) -> SVD<N, R, C>
where where

View File

@ -8,6 +8,7 @@ mod determinant;
// TODO: this should not be needed. However, the exp uses // TODO: this should not be needed. However, the exp uses
// explicit float operations on `f32` and `f64`. We need to // explicit float operations on `f32` and `f64`. We need to
// get rid of these to allow exp to be used on a no-std context. // get rid of these to allow exp to be used on a no-std context.
mod col_piv_qr;
mod decomposition; mod decomposition;
#[cfg(feature = "std")] #[cfg(feature = "std")]
mod exp; mod exp;
@ -31,6 +32,7 @@ mod symmetric_tridiagonal;
pub use self::bidiagonal::*; pub use self::bidiagonal::*;
pub use self::cholesky::*; pub use self::cholesky::*;
pub use self::col_piv_qr::*;
pub use self::convolution::*; pub use self::convolution::*;
#[cfg(feature = "std")] #[cfg(feature = "std")]
pub use self::exp::*; pub use self::exp::*;

View File

@ -60,8 +60,8 @@ where
return QR { qr: matrix, diag }; return QR { qr: matrix, diag };
} }
for ite in 0..min_nrows_ncols.value() { for i in 0..min_nrows_ncols.value() {
householder::clear_column_unchecked(&mut matrix, &mut diag[ite], ite, 0, None); householder::clear_column_unchecked(&mut matrix, &mut diag[i], i, 0, None);
} }
QR { qr: matrix, diag } QR { qr: matrix, diag }

161
tests/linalg/col_piv_qr.rs Normal file
View File

@ -0,0 +1,161 @@
#[cfg_attr(rustfmt, rustfmt_skip)]
use na::Matrix4;
#[test]
fn col_piv_qr() {
let m = Matrix4::new(
1.0, -1.0, 2.0, 1.0, -1.0, 3.0, -1.0, -1.0, 3.0, -5.0, 5.0, 3.0, 1.0, 2.0, 1.0, -2.0,
);
let col_piv_qr = m.col_piv_qr();
assert!(relative_eq!(
col_piv_qr.determinant(),
0.0,
epsilon = 1.0e-7
));
let (q, r, p) = col_piv_qr.unpack();
let mut qr = q * r;
p.inv_permute_columns(&mut qr);
assert!(relative_eq!(m, qr, epsilon = 1.0e-7));
}
#[cfg(feature = "arbitrary")]
mod quickcheck_tests {
macro_rules! gen_tests(
($module: ident, $scalar: ty) => {
mod $module {
use na::{DMatrix, DVector, Matrix3x5, Matrix4, Matrix4x3, Matrix5x3, Vector4};
use std::cmp;
#[allow(unused_imports)]
use crate::core::helper::{RandScalar, RandComplex};
quickcheck! {
fn col_piv_qr(m: DMatrix<$scalar>) -> bool {
let m = m.map(|e| e.0);
let col_piv_qr = m.clone().col_piv_qr();
let (q, r, p) = col_piv_qr.unpack();
let mut qr = &q * &r;
p.inv_permute_columns(&mut qr);
println!("m: {}", m);
println!("col_piv_qr: {}", &q * &r);
relative_eq!(m, &qr, epsilon = 1.0e-7) &&
q.is_orthogonal(1.0e-7)
}
fn col_piv_qr_static_5_3(m: Matrix5x3<$scalar>) -> bool {
let m = m.map(|e| e.0);
let col_piv_qr = m.col_piv_qr();
let (q, r, p) = col_piv_qr.unpack();
let mut qr = q * r;
p.inv_permute_columns(&mut qr);
relative_eq!(m, qr, epsilon = 1.0e-7) &&
q.is_orthogonal(1.0e-7)
}
fn col_piv_qr_static_3_5(m: Matrix3x5<$scalar>) -> bool {
let m = m.map(|e| e.0);
let col_piv_qr = m.col_piv_qr();
let (q, r, p) = col_piv_qr.unpack();
let mut qr = q * r;
p.inv_permute_columns(&mut qr);
relative_eq!(m, qr, epsilon = 1.0e-7) &&
q.is_orthogonal(1.0e-7)
}
fn col_piv_qr_static_square(m: Matrix4<$scalar>) -> bool {
let m = m.map(|e| e.0);
let col_piv_qr = m.col_piv_qr();
let (q, r, p) = col_piv_qr.unpack();
let mut qr = q * r;
p.inv_permute_columns(&mut qr);
println!("{}{}{}{}", q, r, qr, m);
relative_eq!(m, qr, epsilon = 1.0e-7) &&
q.is_orthogonal(1.0e-7)
}
fn col_piv_qr_solve(n: usize, nb: usize) -> bool {
if n != 0 && nb != 0 {
let n = cmp::min(n, 50); // To avoid slowing down the test too much.
let nb = cmp::min(nb, 50); // To avoid slowing down the test too much.
let m = DMatrix::<$scalar>::new_random(n, n).map(|e| e.0);
let col_piv_qr = m.clone().col_piv_qr();
let b1 = DVector::<$scalar>::new_random(n).map(|e| e.0);
let b2 = DMatrix::<$scalar>::new_random(n, nb).map(|e| e.0);
if col_piv_qr.is_invertible() {
let sol1 = col_piv_qr.solve(&b1).unwrap();
let sol2 = col_piv_qr.solve(&b2).unwrap();
return relative_eq!(&m * sol1, b1, epsilon = 1.0e-6) &&
relative_eq!(&m * sol2, b2, epsilon = 1.0e-6)
}
}
return true;
}
fn col_piv_qr_solve_static(m: Matrix4<$scalar>) -> bool {
let m = m.map(|e| e.0);
let col_piv_qr = m.col_piv_qr();
let b1 = Vector4::<$scalar>::new_random().map(|e| e.0);
let b2 = Matrix4x3::<$scalar>::new_random().map(|e| e.0);
if col_piv_qr.is_invertible() {
let sol1 = col_piv_qr.solve(&b1).unwrap();
let sol2 = col_piv_qr.solve(&b2).unwrap();
relative_eq!(m * sol1, b1, epsilon = 1.0e-6) &&
relative_eq!(m * sol2, b2, epsilon = 1.0e-6)
}
else {
false
}
}
fn col_piv_qr_inverse(n: usize) -> bool {
let n = cmp::max(1, cmp::min(n, 15)); // To avoid slowing down the test too much.
let m = DMatrix::<$scalar>::new_random(n, n).map(|e| e.0);
if let Some(m1) = m.clone().col_piv_qr().try_inverse() {
let id1 = &m * &m1;
let id2 = &m1 * &m;
id1.is_identity(1.0e-5) && id2.is_identity(1.0e-5)
}
else {
true
}
}
fn col_piv_qr_inverse_static(m: Matrix4<$scalar>) -> bool {
let m = m.map(|e| e.0);
let col_piv_qr = m.col_piv_qr();
if let Some(m1) = col_piv_qr.try_inverse() {
let id1 = &m * &m1;
let id2 = &m1 * &m;
id1.is_identity(1.0e-5) && id2.is_identity(1.0e-5)
}
else {
true
}
}
}
}
}
);
gen_tests!(complex, RandComplex<f64>);
gen_tests!(f64, RandScalar<f64>);
}

View File

@ -1,6 +1,7 @@
mod balancing; mod balancing;
mod bidiagonal; mod bidiagonal;
mod cholesky; mod cholesky;
mod col_piv_qr;
mod convolution; mod convolution;
mod eigen; mod eigen;
mod exp; mod exp;