Multik

Multidimensional array library for Kotlin.

Modules

• multik-core — contains ndarrays, methods called on them and [math], [stat] and [linalg] interfaces.
• multik-default — implementation including multik-kotlin and multik-openblas for performance.
• multik-kotlin — implementation of [math], [stat] and [linalg] interfaces on JVM.
• multik-openblas — implementation of [math], [stat] and [linalg] interfaces in native code using OpenBLAS.

Using in your projects

Gradle

In your Gradle build script:

1. Add the Maven Central Repository.
2. Add the org.jetbrains.kotlinx:multik-core:$multik_version api dependency.
3. Add an implementation dependency: org.jetbrains.kotlinx:multik-default:$multik_version, org.jetbrains.kotlinx:multik-kotlin:$multik_version or org.jetbrains.kotlinx:multik-openblas:$multik_version.

build.gradle:

repositories {
    mavenCentral()
}

dependencies {
    implementation "org.jetbrains.kotlinx:multik-core:0.2.3"
    implementation "org.jetbrains.kotlinx:multik-default:0.2.3"
}

build.gradle.kts:

repositories {
    mavenCentral()
}

dependencies {
    implementation("org.jetbrains.kotlinx:multik-core:0.2.3")
    implementation("org.jetbrains.kotlinx:multik-default:0.2.3")
}

For a multiplatform project, set the dependency in a common block:

kotlin {
    sourceSets {
        val commonMain by getting {
            dependencies {
                implementation("org.jetbrains.kotlinx:multik-core:0.2.3")
            }
        }
    }
}

or in a platform-specific block:

kotlin {
    sourceSets {
        val jvmName by getting {
            dependencies {
                implementation("org.jetbrains.kotlinx:multik-core-jvm:0.2.3")
            }
        }
    }
}

Jupyter Notebook

Install Kotlin kernel for Jupyter or just visit to Datalore.

Import stable multik version into notebook:

%use multik

Support platforms

Platforms	multik-core	multik-kotlin	multik-openblas	multik-default
JS	✅	✅	❌	✅
linuxX64	✅	✅	✅	✅
mingwX64	✅	✅	✅	✅
macosX64	✅	✅	✅	✅
macosArm64	✅	✅	✅	✅
iosArm64	✅	✅	❌	✅
iosX64	✅	✅	❌	✅
iosSimulatorArm64	✅	✅	❌	✅
JVM	✅	✅	linuxX64 - ✅ mingwX64 - ✅ macosX64 - ✅ macosArm64 - ✅ androidArm64 - ✅	androidArm32 - ❌ androidX86 - ❌ androidX64 - ❌

For Kotlin/JS, we use the new IR. We also use the new memory model in Kotlin/Native. Keep this in mind when using Multik in your multiplatform projects.

Note:

• on ubuntu 18.04 and older multik-openblas doesn't work due to older versions of glibc.
• multik-openblas for desktop targets (linuxX64, mingwX64, macosX64, macosArm64) is experimental and unstable. We will improve stability and perfomance as Kotlin/Native evolves.
• JVM target multik-openblas for Android only supports arm64-v8a processors.

Quickstart

Visit Multik documentation for a detailed feature overview.

Creating arrays

val a = mk.ndarray(mk[1, 2, 3])
/* [1, 2, 3] */

val b = mk.ndarray(mk[mk[1.5, 2.1, 3.0], mk[4.0, 5.0, 6.0]])
/*
[[1.5, 2.1, 3.0],
[4.0, 5.0, 6.0]]
*/

val c = mk.ndarray(mk[mk[mk[1.5f, 2f, 3f], mk[4f, 5f, 6f]], mk[mk[3f, 2f, 1f], mk[4f, 5f, 6f]]])
/*
[[[1.5, 2.0, 3.0],
[4.0, 5.0, 6.0]],

[[3.0, 2.0, 1.0],
[4.0, 5.0, 6.0]]]
*/


mk.zeros<Double>(3, 4) // create an array of zeros
/*
[[0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0]]
*/
mk.ndarray<Float, D2>(setOf(30f, 2f, 13f, 12f), intArrayOf(2, 2)) // create an array from a collection
/*
[[30.0, 2.0],
[13.0, 12.0]]
*/
val d = mk.ndarray(doubleArrayOf(1.0, 1.3, 3.0, 4.0, 9.5, 5.0), 2, 3) // create an array of shape(2, 3) from a primitive array
/*
[[1.0, 1.3, 3.0],
[4.0, 9.5, 5.0]]
*/
mk.d3array(2, 2, 3) { it * it } // create an array of 3 dimension
/*
[[[0, 1, 4],
[9, 16, 25]],

[[36, 49, 64],
[81, 100, 121]]]
*/

mk.d2arrayIndices(3, 3) { i, j -> ComplexFloat(i, j) }
/*
[[0.0+(0.0)i, 0.0+(1.0)i, 0.0+(2.0)i],
[1.0+(0.0)i, 1.0+(1.0)i, 1.0+(2.0)i],
[2.0+(0.0)i, 2.0+(1.0)i, 2.0+(2.0)i]]
 */

mk.arange<Long>(10, 25, 5) // creare an array with elements in the interval [10, 25) with step 5
/* [10, 15, 20] */

mk.linspace<Double>(0, 2, 9) // create an array of 9 elements in the interval [0, 2]
/* [0.0, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0] */

val e = mk.identity<Double>(3) // create an identity array of shape (3, 3)
/*
[[1.0, 0.0, 0.0],
[0.0, 1.0, 0.0],
[0.0, 0.0, 1.0]]
*/

val diag = mk.diagonal(mk[2, 4, 8]) // create a diagonal array
/*
[[2, 0, 0],
[0, 4, 0],
[0, 0, 8]]
 */

Array properties

a.shape // Array dimensions
a.size // Size of array
a.dim // object Dimension
a.dim.d // number of array dimensions
a.dtype // Data type of array elements

Arithmetic operations

val f = b - d // subtraction
/*
[[0.5, 0.8, 0.0],
[0.0, -4.5, 1.0]]
*/

d + f // addition
/*
[[1.5, 2.1, 3.0],
[4.0, 5.0, 6.0]]
*/

b / d // division
/*
[[1.5, 1.6153846153846154, 1.0],
[1.0, 0.5263157894736842, 1.2]]
*/

f * d // multiplication
/*
[[0.5, 1.04, 0.0],
[0.0, -42.75, 5.0]]
*/

Array mathematics

See documentation for other methods of mathematics, linear algebra, statistics.

a.sin() // element-wise sin, equivalent to mk.math.sin(a)
a.cos() // element-wise cos, equivalent to mk.math.cos(a)
b.log() // element-wise natural logarithm, equivalent to mk.math.log(b)
b.exp() // element-wise exp, equivalent to mk.math.exp(b)
d dot e // dot product, equivalent to mk.linalg.dot(d, e)

Aggregate functions

mk.math.sum(c) // array-wise sum
mk.math.min(c) // array-wise minimum elements
mk.math.maxD3(c, axis=0) // maximum value of an array along axis 0
mk.math.cumSum(b, axis=1) // cumulative sum of the elements
mk.stat.mean(a) // mean
mk.stat.median(b) // meadian

Copying arrays

val f = a.copy() // create a copy of the array and its data
val h = b.deepCopy() // create a copy of the array and copy the meaningful data

Operations of Iterable

c.filter { it < 3 } // select all elements less than 3
b.map { (it * it).toInt() } // return squares
c.groupNDArrayBy { it % 2 } // group elements by condition
c.sorted() // sort elements

Indexing/Slicing/Iterating

a[2] // select the element at the 2 index
b[1, 2] // select the element at row 1 column 2
b[1] // select row 1 
b[0..1, 1] // select elements at rows 0 to 1 in column 1
b[0, 0..2..1] // select elements at row 0 in columns 0 to 2 with step 1

for (el in b) {
    print("$el, ") // 1.5, 2.1, 3.0, 4.0, 5.0, 6.0, 
}

// for n-dimensional
val q = b.asDNArray()
for (index in q.multiIndices) {
    print("${q[index]}, ") // 1.5, 2.1, 3.0, 4.0, 5.0, 6.0, 
}

Inplace

val a = mk.linspace<Float>(0, 1, 10)
/*
a = [0.0, 0.1111111111111111, 0.2222222222222222, 0.3333333333333333, 0.4444444444444444, 0.5555555555555556, 
0.6666666666666666, 0.7777777777777777, 0.8888888888888888, 1.0]
*/
val b = mk.linspace<Float>(8, 9, 10)
/*
b = [8.0, 8.11111111111111, 8.222222222222221, 8.333333333333334, 8.444444444444445, 8.555555555555555,
8.666666666666666, 8.777777777777779, 8.88888888888889, 9.0]
*/

a.inplace { 
    math { 
        (this - b) * b
         abs()
    }
}
// a = [64.0, 64.88888, 65.77778, 66.66666, 67.55556, 68.44444, 69.333336, 70.22222, 71.111115, 72.0]

Building

To build the entire project, you need to set up an environment for building multik-openblas:

• JDK 1.8 or higher
• JAVA_HOME environment - to search for jni files
• Compilers gcc, g++, gfortran version 8 or higher. It is important that they are of the same version.

Run ./gradlew assemble to build all modules. If you don't need to build multik-openblas, just disable the cmake_build task and build the module you need.

Contributing

There is an opportunity to contribute to the project:

1. Implement math, linalg, stat interfaces.
2. Create your own engine successor from Engine, for example - JvmEngine.
3. Use mk.addEngine and mk.setEngine to use your implementation.