Description:

ml linalg

SIMD-based linear algebra and statistics for data science with Dart

TABLE OF CONTENTS

What is linear algebra
What is SIMD
Vectors

A couple of words about the underlying architecture

Vector benchmarks
Vector operations

Vector summation
Vector and List summation
Summation of Vectors of different dtype
Vector subtraction
Vector and List subtraction
Subtraction of Vectors of different dtype
Element-wise Vector by Vector multiplication
Element-wise Vector and List multiplication
Element-wise multiplication of Vectors of different dtype
Element-wise Vector by Vector division
Element-wise Vector and List division
Element-wise division of Vectors of different dtype
Euclidean norm
Manhattan norm
Mean value
Median value
Sum of all vector elements
Product of all vector elements
Element-wise power
Element-wise exp
Dot product
Sum of a vector and a scalar
Subtraction of a scalar from a vector
Multiplication of a vector by a scalar
Division of a vector by a scalar
Euclidean distance between two vectors
Manhattan distance between two vectors
Cosine distance between two vectors
Vector normalization (using Euclidean norm)
Vector normalization (using Manhattan norm)
Vector rescaling (min-max normalization)
Vector serialization
Vector mapping






Matrices

Matrix operations

Creation of diagonal matrix
Creation of scalar matrix
Creation of identity matrix
Creation of column matrix
Creation of row matrix
Sum of a matrix and another matrix
Sum of a matrix and a scalar
Multiplication of a matrix and a vector
Multiplication of a matrix and another matrix
Multiplication of a matrix and a scalar
Hadamard product (element-wise matrices multiplication)
Element-wise matrices subtraction
Matrix transposition
Matrix LU decomposition
Matrix Cholesky decomposition
Matrix LU inversion
Matrix Cholesky inversion
Lower triangular matrix inversion
Upper triangular matrix inversion
Solving a system of linear equations
Obtaining Matrix eigenvectors and eigenvalues, Power Iteration method
Matrix row-wise reduce
Matrix column-wise reduce
Matrix row-wise mapping
Matrix column-wise mapping
Matrix element-wise mapping
Matrix' columns filtering (by column index)
Matrix' columns filtering (by column)
Getting max value of the matrix
Getting min value of the matrix
Matrix element-wise power
Matrix element-wise exp
Sum of all matrix elements
Product of all matrix elements
Matrix indexing and sampling
Add new columns to a matrix
Matrix serialization/deserialization




vector_math and ml_linalg differences
Contacts


Linear algebra #
In a few words, linear algebra is a branch of mathematics that works with vectors and matrices.
Vectors and matrices are practical tools in real-life applications, such as machine learning algorithms. These significant
mathematical entities are implemented in plenty of programming languages.
As Dart offers developers good instrumentation, e.g. highly optimized virtual machine, specific data types and rich out-of-the-box library, Dart-based implementation of vectors and matrices has to be quite performant.
Among numerous standard Dart tools, there are SIMD data types, and support of SIMD computational
architecture served as inspiration for creating this library.
What is SIMD? #
SIMD stands for Single instruction, multiple data - it's a computer architecture that allows
to perform uniform mathematical operations in parallel on a list-like data structure. For instance, one has two arrays:
final a = [10, 20, 30, 40];
final b = [50, 60, 70, 80];
copied to clipboard
and one needs to add these arrays element-wise. Using the regular architecture this operation could be done in the following
manner:
final c = List(4);

c[0] = a[0] + b[0]; // operation 1
c[1] = a[1] + b[1]; // operation 2
c[2] = a[2] + b[2]; // operation 3
c[3] = a[3] + b[3]; // operation 4
copied to clipboard
As you may have noticed, we need to do 4 operations one by one in a row using regular
computational approach. But with help of SIMD architecture we may do one arithmetic operation on several operands in
parallel, thus element-wise sum of two arrays can be done for just one step:



Vectors #
A couple of words about the underlying architecture #
The library contains two high performant vector classes based on Float32x4
and Float64x2 data types -
Float32x4Vector and Float64x2Vector (the second one is generated from the source code of the first vector's implementation)
Most of element-wise operations in the first one are performed in four "threads" and in the second one - in two "threads".
Implementation of both classes is hidden from the library's users. You can create a
Float32x4Vector or a Float64x2Vector instance via Vector factory (see examples below).
One can create Float32x4-based vectors the following way:
import 'package:ml_linalg/linalg.dart';

void main() {
final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0], dtype: DType.float32);
}
copied to clipboard
or simply
import 'package:ml_linalg/linalg.dart';

void main() {
final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
}
copied to clipboard
since dtype is set to DType.float32 by default.
One can create Float64x2-based vectors the following way:
import 'package:ml_linalg/linalg.dart';

void main() {
final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0], dtype: DType.float64);
}
copied to clipboard
Float32x4-based vectors are much faster than Float64x2-based ones, but Float64x2-based vectors are more precise since
they use 64 bits to represent numbers in the memory versus 32 bits for Float32x4-based vectors.
Nevertheless, Float32x4 representation uses by default since usually 32 bits is enough for number precision, and along
with that, this representation is more performant.
The vectors are immutable: once created, the vector cannot be changed. All the vector operations
lead to creation of a new vector instance (of course, if the operation is supposed to return a Vector).
Both classes implement Iterable<double> interface, so it's possible to use them as regular
iterable collections.
It's possible to use vector instances as keys for HashMap and similar data structures
and to look up a value by the vector-key, since the hash code for equal vectors is the same:
import 'package:ml_linalg/vector.dart';

final map = HashMap<Vector, bool>();

map[Vector.fromList([1, 2, 3, 4, 5])] = true;

print(map[Vector.fromList([1, 2, 3, 4, 5])]); // true
print(Vector.fromList([1, 2, 3, 4, 5]).hashCode == Vector.fromList([1, 2, 3, 4, 5]).hashCode); // true
copied to clipboard
Vector benchmarks #
To see the performance benefits provided by the library's vector classes, one may visit benchmark directory: one may find
there a baseline benchmark -
element-wise summation of two regular List instances and a benchmark
of a similar operation, but performed on two Float32x4Vector instances on the same amount of elements and compare
the timings:

Baseline benchmark (executed on Macbook Air mid 2017), 2 regular lists each with 10,000,000 elements:





Actual benchmark (executed on Macbook Air mid 2017), 2 vectors each with 10,000,000 elements:




It took 15 seconds to create a new regular list by summing the elements of two lists, and 0.7 second to sum two vectors -
the difference is significant.
Vector operations examples #
Vector summation
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector1 + vector2;

print(result.toList()); // [3.0, 5.0, 7.0, 9.0, 11.0]
copied to clipboard
Vector and List summation
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final result = vector + [2.0, 3.0, 4.0, 5.0, 6.0];

print(result.toList()); // [3.0, 5.0, 7.0, 9.0, 11.0]
copied to clipboard
Summation of Vectors of different dtype
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0], dtype: DType.float32);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0], dtype: DType.float64);
final result = vector1 + vector2;

print(result.toList()); // [3.0, 5.0, 7.0, 9.0, 11.0]
copied to clipboard
Vector subtraction
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([4.0, 5.0, 6.0, 7.0, 8.0]);
final vector2 = Vector.fromList([2.0, 3.0, 2.0, 3.0, 2.0]);
final result = vector1 - vector2;

print(result.toList()); // [2.0, 2.0, 4.0, 4.0, 6.0]
copied to clipboard
Vector and List subtraction
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([4.0, 5.0, 6.0, 7.0, 8.0]);
final result = vector - [2.0, 3.0, 2.0, 3.0, 2.0];

print(result.toList()); // [2.0, 2.0, 4.0, 4.0, 6.0]
copied to clipboard
Subtraction of vectors of different dtype
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([4.0, 5.0, 6.0, 7.0, 8.0], dtype: DType.float32);
final vector2 = Vector.fromList([2.0, 3.0, 2.0, 3.0, 2.0], dtype: DType.float64);
final result = vector1 - vector2;

print(result.toList()); // [2.0, 2.0, 4.0, 4.0, 6.0]
copied to clipboard
Element wise Vector by Vector multiplication
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector1 * vector2;

print(result.toList()); // [2.0, 6.0, 12.0, 20.0, 30.0]
copied to clipboard
Element wise Vector and List multiplication
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final result = vector * [2.0, 3.0, 4.0, 5.0, 6.0];

print(result.toList()); // [2.0, 6.0, 12.0, 20.0, 30.0]
copied to clipboard
Element wise multiplication of Vectors of different dtype
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0], dtype: DType.float32);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0], dtype: DType.float64);
final result = vector1 * vector2;

print(result.toList()); // [2.0, 6.0, 12.0, 20.0, 30.0]
copied to clipboard
Element wise Vector by Vector division
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([6.0, 12.0, 24.0, 48.0, 96.0]);
final vector2 = Vector.fromList([3.0, 4.0, 6.0, 8.0, 12.0]);
final result = vector1 / vector2;

print(result.toList()); // [2.0, 3.0, 4.0, 6.0, 8.0]
copied to clipboard
Element-wise Vector and List division
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([6.0, 12.0, 24.0, 48.0, 96.0]);
final result = vector / [3.0, 4.0, 6.0, 8.0, 12.0];

print(result.toList()); // [2.0, 3.0, 4.0, 6.0, 8.0]
copied to clipboard
Element wise division of vectors of different dtype
This operation doesn't benefit from SIMD
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([6.0, 12.0, 24.0, 48.0, 96.0], dtype: DType.float32);
final vector2 = Vector.fromList([3.0, 4.0, 6.0, 8.0, 12.0], dtype: DType.float64);
final result = vector1 / vector2;

print(result.toList()); // [2.0, 3.0, 4.0, 6.0, 8.0]
copied to clipboard
Euclidean norm
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector.norm();

print(result); // sqrt(2^2 + 3^2 + 4^2 + 5^2 + 6^2) = sqrt(90) ~~ 9.48
copied to clipboard
Manhattan norm
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector.norm(Norm.manhattan);

print(result); // 2 + 3 + 4 + 5 + 6 = 20.0
copied to clipboard
Mean value
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector.mean();

print(result); // (2 + 3 + 4 + 5 + 6) / 5 = 4.0
copied to clipboard
Median value
Even length
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([10, 12, 4, 7, 9, 12]);
final result = vector.median();

print(result); // 9.5
copied to clipboard
Odd length
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([10, 12, 4, 7, 9, 12, 34]);
final result = vector.median();

print(result); // 10
copied to clipboard
Sum of all vector elements
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector.sum();

print(result); // 2 + 3 + 4 + 5 + 6 = 20.0
copied to clipboard
Product of all vector elements
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector.prod();

print(result); // 2 * 3 * 4 * 5 * 6 = 720
copied to clipboard
Element-wise power
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector.pow(3);

print(result); // [2 ^ 3 = 8.0, 3 ^ 3 = 27.0, 4 ^ 3 = 64.0, 5 ^3 = 125.0, 6 ^ 3 = 216.0]
copied to clipboard
Element-wise exp
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector.exp();

print(result); // [e ^ 2, e ^ 3, e ^ 4, e ^ 5, e ^ 6]
copied to clipboard
Dot product of two vectors
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector1.dot(vector2);

print(result); // 1.0 * 2.0 + 2.0 * 3.0 + 3.0 * 4.0 + 4.0 * 5.0 + 5.0 * 6.0 = 70.0
copied to clipboard
Sum of a vector and a scalar
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final scalar = 5.0;
final result = vector + scalar;

print(result.toList()); // [6.0, 7.0, 8.0, 9.0, 10.0]
copied to clipboard
Subtraction of a scalar from a vector
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final scalar = 5.0;
final result = vector - scalar;

print(result.toList()); // [-4.0, -3.0, -2.0, -1.0, 0.0]
copied to clipboard
Multiplication of a vector by a scalar
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final scalar = 5.0;
final result = vector * scalar;

print(result.toList()); // [5.0, 10.0, 15.0, 20.0, 25.0]
copied to clipboard
Division of a vector by a scalar
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([25.0, 50.0, 75.0, 100.0, 125.0]);
final scalar = 5.0;
final result = vector.scalarDiv(scalar);

print(result.toList()); // [5.0, 10.0, 15.0, 20.0, 25.0]
copied to clipboard
Euclidean distance between two vectors
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector1.distanceTo(vector2, distance: Distance.euclidean);

print(result); // ~~2.23
copied to clipboard
Manhattan distance between two vectors
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector1.distanceTo(vector2, distance: Distance.manhattan);

print(result); // 5.0
copied to clipboard
Cosine distance between two vectors
import 'package:ml_linalg/linalg.dart';

final vector1 = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final vector2 = Vector.fromList([2.0, 3.0, 4.0, 5.0, 6.0]);
final result = vector1.distanceTo(vector2, distance: Distance.cosine);

print(result); // 0.00506
copied to clipboard
Vector normalization using Euclidean norm
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, 2.0, 3.0, 4.0, 5.0]);
final result = vector.normalize(Norm.euclidean);

print(result); // [0.134, 0.269, 0.404, 0.539, 0.674]
copied to clipboard
Vector normalization using Manhattan norm
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, -2.0, 3.0, -4.0, 5.0]);
final result = vector.normalize(Norm.manhattan);

print(result); // [0.066, -0.133, 0.200, -0.266, 0.333]
copied to clipboard
Vector rescaling (min-max normalization)
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, -2.0, 3.0, -4.0, 5.0, 0.0]);
final result = vector.rescale();

print(result); // [0.555, 0.222, 0.777, 0.0, 1.0, 0.444]
copied to clipboard
Vector serialization
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, -2.0, 3.0, -4.0, 5.0, 0.0]);
final serialized = vector.toJson();
print(serialized); // it yields a serializable representation of the vector

final restoredVector = Vector.fromJson(serialized);
print(restoredVector); // [1.0, -2.0, 3.0, -4.0, 5.0, 0.0]
copied to clipboard
Vector mapping
import 'package:ml_linalg/linalg.dart';

final vector = Vector.fromList([1.0, -2.0, 3.0, -4.0, 5.0, 0.0]);
final mapped = vector.mapToVector((el) => el * 2);

print(mapped); // [2.0, -4.0, 6.0, -8.0, 10.0, 0.0]
print(mapped is Vector); // true
print(identical(vector, mapped)); // false
copied to clipboard
Matrices #
Along with SIMD vectors, the library contains SIMD-based Matrices. One can use the matrices via
Matrix factory. The matrices are immutable as well
as vectors and also they implement Iterable interface (to be more precise, Iterable<Iterable<double>>), thus it's
possible to use them as a regular iterable collection.
Matrices are serializable, and that means that one can easily convert a Matrix instance to a json-serializable map
via toJson method, see the examples below.
Matrix operations examples #
Creation of diagonal matrix
import 'package:ml_linalg/matrix.dart';

final matrix = Matrix.diagonal([1, 2, 3, 4, 5]);

print(matrix);
copied to clipboard
The output:
Matrix 5 x 5:
(1.0, 0.0, 0.0, 0.0, 0.0)
(0.0, 2.0, 0.0, 0.0, 0.0)
(0.0, 0.0, 3.0, 0.0, 0.0)
(0.0, 0.0, 0.0, 4.0, 0.0)
(0.0, 0.0, 0.0, 0.0, 5.0)
copied to clipboard
Creation of scalar matrix
import 'package:ml_linalg/matrix.dart';

final matrix = Matrix.scalar(3, 5);

print(matrix);
copied to clipboard
The output:
Matrix 5 x 5:
(3.0, 0.0, 0.0, 0.0, 0.0)
(0.0, 3.0, 0.0, 0.0, 0.0)
(0.0, 0.0, 3.0, 0.0, 0.0)
(0.0, 0.0, 0.0, 3.0, 0.0)
(0.0, 0.0, 0.0, 0.0, 3.0)
copied to clipboard
Creation of identity matrix
import 'package:ml_linalg/matrix.dart';

final matrix = Matrix.identity(5);

print(matrix);
copied to clipboard
The output:
Matrix 5 x 5:
(1.0, 0.0, 0.0, 0.0, 0.0)
(0.0, 1.0, 0.0, 0.0, 0.0)
(0.0, 0.0, 1.0, 0.0, 0.0)
(0.0, 0.0, 0.0, 1.0, 0.0)
(0.0, 0.0, 0.0, 0.0, 1.0)
copied to clipboard
Creation of column matrix
final matrix = Matrix.column([1, 2, 3, 4, 5]);

print(matrix);
copied to clipboard
The output:
Matrix 5 x 1:
(1.0)
(2.0)
(3.0)
(4.0)
(5.0)
copied to clipboard
Creation of row matrix
final matrix = Matrix.row([1, 2, 3, 4, 5]);

print(matrix);
copied to clipboard
The output:
Matrix 1 x 5:
(1.0, 2.0, 3.0, 4.0, 5.0)
copied to clipboard
Sum of a matrix and another matrix
import 'package:ml_linalg/linalg.dart';

final matrix1 = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, .0, -2.0, -3.0],
]);
final matrix2 = Matrix.fromList([
[10.0, 20.0, 30.0, 40.0],
[-5.0, 16.0, 2.0, 18.0],
[2.0, -1.0, -2.0, -7.0],
]);
print(matrix1 + matrix2);
// [
// [11.0, 22.0, 33.0, 44.0],
// [0.0, 22.0, 9.0, 26.0],
// [11.0, -1.0, -4.0, -10.0],
// ];
copied to clipboard
Sum of a matrix and a scalar
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, .0, -2.0, -3.0],
]);
print(matrix + 7);
// [
// [8.0, 9.0, 10.0, 11.0],
// [12.0, 13.0, 14.0, 15.0],
// [16.0, 7.0, 5.0, 4.0],
// ];
copied to clipboard
Multiplication of a matrix and a vector
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, .0, -2.0, -3.0],
]);
final vector = Vector.fromList([2.0, 3.0, 4.0, 5.0]);
final result = matrix * vector;
print(result);
// a vector-column [
// [40],
// [96],
// [-5],
//]
copied to clipboard
Multiplication of a matrix and another matrix
import 'package:ml_linalg/linalg.dart';

final matrix1 = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, .0, -2.0, -3.0],
]);
final matrix2 = Matrix.fromList([
[1.0, 2.0],
[5.0, 6.0],
[9.0, .0],
[-9.0, 1.0],
]);
final result = matrix1 * matrix2;
print(result);
//[
// [2.0, 18.0],
// [26.0, 54.0],
// [18.0, 15.0],
//]
copied to clipboard
Multiplication of a matrix and a scalar
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, .0, -2.0, -3.0],
]);
print(matrix * 3);
// [
// [3.0, 6.0, 9.0, 12.0],
// [15.0, 18.0, 21.0, 24.0],
// [27.0, .0, -6.0, -9.0],
// ];
copied to clipboard
Hadamard product (element-wise matrices multiplication)
import 'package:ml_linalg/linalg.dart';

final matrix1 = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, 0.0, -2.0, -3.0],
]);
final matrix2 = Matrix.fromList([
[7.0, 1.0, 9.0, 2.0],
[2.0, 4.0, 3.0, -8.0],
[0.0, -10.0, -2.0, -3.0],
]);
print(matrix1.multiply(matrix2));
// [
// [ 7.0, 2.0, 27.0, 8.0],
// [10.0, 24.0, 21.0, -64.0],
// [ 0.0, 0.0, 4.0, 9.0],
// ];
copied to clipboard
Element wise matrices subtraction
import 'package:ml_linalg/linalg.dart';

final matrix1 = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, .0, -2.0, -3.0],
]);
final matrix2 = Matrix.fromList([
[10.0, 20.0, 30.0, 40.0],
[-5.0, 16.0, 2.0, 18.0],
[2.0, -1.0, -2.0, -7.0],
]);
print(matrix1 - matrix2);
// [
// [-9.0, -18.0, -27.0, -36.0],
// [10.0, -10.0, 5.0, -10.0],
// [7.0, 1.0, .0, 4.0],
// ];
copied to clipboard
Matrix transposition
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
[9.0, .0, -2.0, -3.0],
]);
final result = matrix.transpose();
print(result);
//[
// [1.0, 5.0, 9.0],
// [2.0, 6.0, .0],
// [3.0, 7.0, -2.0],
// [4.0, 8.0, -3.0],
//]
copied to clipboard
Matrix LU decomposition
final matrix = Matrix.fromList([
[4, 12, -16],
[12, 37, -43],
[-16, -43, 98],
], dtype: dtype);
final decomposed = matrix.decompose(Decomposition.LU);

// yields approximately the same matrix as the original one:
print(decomposed.first * decomposed.last);
copied to clipboard
Matrix Cholesky decomposition
final matrix = Matrix.fromList([
[4, 12, -16],
[12, 37, -43],
[-16, -43, 98],
], dtype: dtype);
final decomposed = matrix.decompose(Decomposition.cholesky);

// yields approximately the same matrix as the original one:
print(decomposed.first * decomposed.last);
copied to clipboard
Keep in mind that Cholesky decomposition is applicable only for positive definite and symmetric matrices
Matrix LU inversion
final matrix = Matrix.fromList([
[-16, -43, 98],
[33, 12.4, 37],
[12, -88.3, 4],
], dtype: dtype);
final inverted = matrix.inverse(Inverse.LU);

print(inverted * matrix);
// The output (there can be some round-off errors):
// [1, 0, 0],
// [0, 1, 0],
// [0, 0, 1],
copied to clipboard
Matrix Cholesky inversion
final matrix = Matrix.fromList([
[4, 12, -16],
[12, 37, -43],
[-16, -43, 98],
], dtype: dtype);
final inverted = matrix.inverse(Inverse.cholesky);

print(inverted * matrix);
// The output (there can be some round-off errors):
// [1, 0, 0],
// [0, 1, 0],
// [0, 0, 1],
copied to clipboard
Keep in mind that since this kind of inversion is based on Cholesky decomposition, the inversion is applicable only for positive definite and symmetric matrices
Lower triangular matrix inversion
final matrix = Matrix.fromList([
[ 4, 0, 0],
[ 12, 37, 0],
[-16, -43, 98],
], dtype: dtype);
final inverted = matrix.inverse(Inverse.forwardSubstitution);

print(inverted * matrix);
// The output (there can be some round-off errors):
// [1, 0, 0],
// [0, 1, 0],
// [0, 0, 1],
copied to clipboard
Upper triangular matrix inversion
final matrix = Matrix.fromList([
[4, 12, -16],
[0, 37, -43],
[0, 0, -98],
], dtype: dtype);
final inverted = matrix.inverse(Inverse.backwardSubstitution);

print(inverted * matrix);
// The output (there can be some round-off errors):
// [1, 0, 0],
// [0, 1, 0],
// [0, 0, 1],
copied to clipboard
Solving a system of linear equations
A matrix notation for a system of linear equations:
AX=B
copied to clipboard
To solve the system and find X, one may use the solve method:
import 'package:ml_linalg/linalg.dart';

void main() {
final A = Matrix.fromList([
[1, 1, 1],
[0, 2, 5],
[2, 5, -1],
], dtype: dtype);
final B = Matrix.fromList([
[6],
[-4],
[27],
], dtype: dtype);
final result = A.solve(B);

print(result); // the output is close to [[5], [3], [-2]]
}
copied to clipboard
Obtaining Matrix eigenvectors and eigenvalues, Power Iteration method
The method returns a collection of pairs of an eigenvector and its corresponding eigenvalue.
By default Power iteration method is used.
final matrix = Matrix.fromList([
[1, 0],
[0, 2],
]);
final eigen = matrix.eigen();

print(eigen); // It prints the following: [Value: 1.999, Vector: (0.001, 0.999);]
copied to clipboard
Matrix row-wise reduce
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
]);
final reduced = matrix.reduceRows((combine, row) => combine + row);
print(reduced); // [6.0, 8.0, 10.0, 12.0]
copied to clipboard
Matrix column-wise reduce
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 17.0, 18.0],
[21.0, 22.0, 23.0, 24.0],
]);
final result = matrix.reduceColumns((combine, vector) => combine + vector);
print(result); // [50, 66, 90]
copied to clipboard
Matrix row-wise mapping
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
]);
final modifier = Vector.filled(4, 2.0);
final newMatrix = matrix.rowsMap((row) => row + modifier);
print(newMatrix);
// [
// [3.0, 4.0, 5.0, 6.0],
// [7.0, 8.0, 9.0, 10.0],
// ]
copied to clipboard
Matrix column-wise mapping
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0],
]);
final modifier = Vector.filled(2, 2.0);
final newMatrix = matrix.columnsMap((column) => column + modifier);
print(newMatrix);
// [
// [3.0, 4.0, 5.0, 6.0],
// [7.0, 8.0, 9.0, 10.0],
// ]
copied to clipboard
Matrix element-wise mapping
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 0.0, 18.0],
[21.0, 22.0, -23.0, 24.0],
], dtype: DType.float32);
final result = matrix.mapElements((element) => element * 2);

print(result);
// [
// [22.0, 24.0, 26.0, 28.0],
// [30.0, 32.0, 0.0, 36.0],
// [42.0, 44.0, -46.0, 48.0],
// ]
copied to clipboard
Matrix' columns filtering (by column index)
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 17.0, 18.0],
[21.0, 22.0, 23.0, 24.0],
], dtype: dtype);

final indicesToExclude = [0, 3];
final result = matrix.filterColumns((column, idx) => !indicesToExclude.contains(idx));

print(result);
// [
// [12.0, 13.0],
// [16.0, 17.0],
// [22.0, 23.0],
// ]
copied to clipboard
Matrix' columns filtering (by column)
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 33.0, 13.0, 14.0],
[15.0, 92.0, 17.0, 18.0],
[21.0, 22.0, 23.0, 24.0],
], dtype: dtype);

final result = matrix.filterColumns((column, _) => column.sum() > 100);

print(result);
// [
// [33.0],
// [92.0],
// [22.0],
// ];
copied to clipboard
Getting max value of the matrix
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 17.0, 18.0],
[21.0, 22.0, 23.0, 24.0],
[24.0, 32.0, 53.0, 74.0],
]);
final maxValue = matrix.max();
print(maxValue);
// 74.0
copied to clipboard
Getting min value of the matrix
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 0.0, 18.0],
[21.0, 22.0, -23.0, 24.0],
[24.0, 32.0, 53.0, 74.0],
]);
final minValue = matrix.min();
print(minValue);
// -23.0
copied to clipboard
Matrix element-wise power
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0],
[4.0, 5.0, 6.0],
[7.0, 8.0, 9.0],
]);
final result = matrix.pow(3.0);

print(result);
// [1 ^ 3 = 1, 2 ^ 3 = 8, 3 ^ 3 = 27 ]
// [4 ^ 3 = 64, 5 ^ 3 = 125, 6 ^ 3 = 216]
// [7 ^ 3 = 343, 8 ^ 3 = 512, 9 ^ 3 = 729]
copied to clipboard
Matrix element-wise exp
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0],
[4.0, 5.0, 6.0],
[7.0, 8.0, 9.0],
]);
final result = matrix.exp();

print(result);
// [e ^ 1, e ^ 2, e ^ 3]
// [e ^ 4, e ^ 5, e ^ 6]
// [e ^ 7, e ^ 8, e ^ 9]
copied to clipboard
Sum of all matrix elements
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0],
[4.0, 5.0, 6.0],
[7.0, 8.0, 9.0],
]);
final result = matrix.sum();

print(result); // 1.0 + 2.0 + 3.0 + 4.0 + 5.0 + 6.0 + 7.0 + 8.0 + 9.0
copied to clipboard
Product of all matrix elements
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[1.0, 2.0, 3.0],
[4.0, 5.0, 6.0],
[7.0, 8.0, 9.0],
]);
final result = matrix.product();

print(result); // 1.0 * 2.0 * 3.0 * 4.0 * 5.0 * 6.0 * 7.0 * 8.0 * 9.0
copied to clipboard
Matrix indexing and sampling
To access a certain row vector of the matrix one may use [] operator:
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 0.0, 18.0],
[21.0, 22.0, -23.0, 24.0],
[24.0, 32.0, 53.0, 74.0],
]);

final row = matrix[2];

print(row); // [21.0, 22.0, -23.0, 24.0]
copied to clipboard
The library's matrix interface offers sample method that is supposed to return a new matrix,
consisting of different segments of a source matrix. It's possible to build a new matrix from certain columns and
vectors and they should not be necessarily subsequent.
For example, one needs to create a matrix from rows 1, 3, 5 and columns 1 and 3. To do so,
it's needed to perform the following:
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
//| 1 | | 3 |
[4.0, 8.0, 12.0, 16.0, 34.0], // 1 Range(0, 1)
[20.0, 24.0, 28.0, 32.0, 23.0],
[36.0, .0, -8.0, -12.0, 12.0], // 3 Range(2, 3)
[16.0, 1.0, -18.0, 3.0, 11.0],
[112.0, 10.0, 34.0, 2.0, 10.0], // 5 Range(4, 5)
]);
final result = matrix.sample(
rowIndices: [0, 2, 4],
columnIndices: [0, 2],
);
print(result);
/*
[4.0, 12.0],
[36.0, -8.0],
[112.0, 34.0]
*/
copied to clipboard
Add new columns to a matrix
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 0.0, 18.0],
[21.0, 22.0, -23.0, 24.0],
[24.0, 32.0, 53.0, 74.0],
], dtype: DType.float32);

final updatedMatrix = matrix.insertColumns(0, [
Vector.fromList([1.0, 2.0, 3.0, 4.0]),
Vector.fromList([-1.0, -2.0, -3.0, -4.0]),
]);

print(updatedMatrix);
// [
// [1.0, -1.0, 11.0, 12.0, 13.0, 14.0],
// [2.0, -2.0, 15.0, 16.0, 0.0, 18.0],
// [3.0, -3.0, 21.0, 22.0, -23.0, 24.0],
// [4.0, -4.0, 24.0, 32.0, 53.0, 74.0],
// ]

print(updatedMatrix == matrix); // false
copied to clipboard
Matrix serialization/deserialization
To convert a matrix to a json-serializable map one may use toJson method:
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromList([
[11.0, 12.0, 13.0, 14.0],
[15.0, 16.0, 0.0, 18.0],
[21.0, 22.0, -23.0, 24.0],
[24.0, 32.0, 53.0, 74.0],
]);

final serialized = matrix.toJson();
copied to clipboard
To restore a serialized matrix one may use Matrix.fromJson constructor:
import 'package:ml_linalg/linalg.dart';

final matrix = Matrix.fromJson(serialized);
copied to clipboard
Differences between vector math and ml linalg #
There are similar solutions on the internet, the most famous of which is vector_math by the Google team. At first
glance, vector_math and ml_linalg look similar - both of them are based on SIMD, but in fact, these are two completely
different libraries:
vector_math supports only four dimensions for vectors and matrices at max; ml_linalg can handle vectors and matrices
of potentially infinite length, keeping SIMD nature.
Contacts #
If you have questions, feel free to write me on

Twitter
Facebook
Linkedin

License

For personal and professional use. You cannot resell or redistribute these repositories in their original state.

Share

• Share on Facebook
• Share on Twitter