Keigen

Keigen is a Kotlin (Android) library for fast matrix operations and linear algebra built on a C++ foundation.

Keigen is a Kotlin wrapper for Eigen, a linear algebra library written in C++.

Most common matrix-to-matrix and matrix-to-scalar operations are implemented. All arithmetic operators are covered +, -, *, get by row/col val x = matrix[row, col], set by row/col matrix[row, col] = 2, transpose, raw data array initialize/get/set FloatMatrix(rows, cols, aFloatArray).

Get started

Add jitpack.io repo to your root build.gradle:

allprojects {
    repositories {
        //...
        maven { url "https://jitpack.io" }
    }
}

Include in Android project:

implementation 'com.github.paramsen:keigen:1.6.1'

Simple and expressive with Kotlin operator overloads

DoubleMatrix, FloatMatrix, LongMatrix, IntMatrix, ShortMatrix, ByteMatrix. All Matrix types support the same functionality and has the same unit test coverage.

Matrix instantiation

2 by 3 matrix filled with 0s val a = LongMatrix(2, 3)
2 by 3 matrix filled with 1s val a = FloatMatrix(2, 3, 1f)

256 by 512 matrix from array data:

val data = ShortArray(256 * 512) { it.toShort() }
val a = ShortMatrix(256, 512, data)

IMPORTANT (!) Matrix destruction

All Matrix types has to be destructed when not in use anymore by calling matrix.dispose(). This is a side effect of building this library on top of the C++ library Eigen, all Matrix types has references to native allocations that need to be freed to not result in memory leaks.

Matrix to matrix operations

matrixC = matrixA + matrixB
matrixC = matrixA - matrixB
matrixC = matrixA * matrixB

matrixA += matrixB
matrixA -= matrixB
matrixA *= matrixB

Matrix to scalar operations

matrixC = matrixA + 2f
matrixC = matrixA - 2f
matrixC = matrixA * 2f
matrixC = matrixA / 2f

matrixA += 2f
matrixA -= 2f
matrixA *= 2f
matrixA /= 2f

Matrix get by index and array

val x = matrix[1, 3]
val floatArray = matrix.getData()

Matrix set by index and array

matrix[1, 3] = 5L
matrix.setData(aLongArray)

Other matrix operations

val matrixB = matrixA.transpose()

Matrix class code generation

All Kotlin Matrix classes are generated from the FloatMatrix base implementation. The unit tests are also generated from the FloatMatrixTest class, which means that each Matrix type has full test coverage.

The C++ interface for Eigen is implemented using generics (or templates in C++ lingo). It should be noted that the Kotlin interface isn't implemented using generics because the exact types need to be known at the JNI bridge (and due to type erasure in Java/Kotlin, the exact type of a generic variable cannot be inferred).

Development

Setup

Eigen is not directly included within this git repository, it's in a git module and is pulled from Eigens repository on demand.

To setup Eigen:

Run git submodule init; git submodule update in project root
Check that Eigen exists in keigen/src/main/native/Eigen

Release

There's a Gradle task that generates the README.md from template and git tags the current commit with the version number. JitPack builds on push of the tag.

Release steps are:

Bump version in noise/build.gradle
Run ./gradlew release in project root (generates readme)
Push generated readme changes to repo
Wait for JitPack to build

Licenses

Keigen is licensed under the permissive APL-2.0.

The included parts of Eigen (included as-is) are licensed under the permissive MPL-2.0, the EIGEN_MPL2_ONLY flag is used in the project as described in the link.

Artwork

The artwork combines the scholar owl in the Eigen artwork and the Kotlin logo colors. In Keigen, the owl has obviously found a new cool style, heavily inspired by The Matrix - not to be confused with a matrix.

The artwork is made by me.

paramsen/Keigen