tensor4s

What is it?

tensor4s aims to simplify numerical computations in scala by providing multi-dimensional arrays (tensors) and linear algebra operators using a simple and powerful syntax. Tensor4s is comparable but yet different to following existing linear algebra libraries:

• ScalaNLP's Breeze
• ND4J/ND4S

tensor4s aims not be a replacement but yet another linear algebra library with following features:

Implemented in Scala

Plain Java syntax is not well suited o support simple linear algebra expressions known python's numpy or matlab. The situation is different with Scala. Scala's rich syntax allows for a seamless integration of most know operators/expressions. Tensor4s is implemented from scratch in scala, bringing the powerful syntax of existing linear algebra library to the scala world.

Backed by plain old arrays

tensor4's tensors are backed by plain old java arrays. Tensors can be seen as multidimensional containers storing elements of arbitrary data types. Tensors are therefore not limited to primitive data types such as double or float, but may contain any data structure, even other tensors.

Note: Currently the status of this library is pre alpha. The code is likely to pass breaking changes in the future. Use at your own risk.

Build and install:

tenor4s uses SBT for building the library. Clone this git repository, open shell and publish artifacts locally:

sbt publishLocal

Include tensor4s to your library dependencies:

libraryDependencies ++= Seq(
  "com.github.arzt" %% "scala-tensor" % "0.0.0-SNAPSHOT"
)

Note: Adding tensor4s to Maven repository is planned for a later stable release.

Scaladoc

To Generate scaladoc documentation run the following command

sbt doc

Syntax

Creating tensors

//import linear algebra DSL:
//all following examples expect the presence of this imports
import com.github.arzt.tensor.TensorImplicits._
import com.github.arzt.tensor.Tensor

val shape = Vector(2, 3)
val data = Array[Int](
  0,1,2,
  3,4,5
)
//Each line creates the same 2 by 3 matrix
val tensor0 = Tensor(shape, data)
val tensor1 = data.asRows(2)
val tensor2 = data.asCols(3)
val tensor3 = data.asTensor(2, 3)
//Each tensor has the inferred type Tensor[Int]

//Row/Columns vectors:
val row: Tensor[Boolean] = Array(true, false, false).asRow
val col: Tensor[String] = Array("1", "2", "3").asCol

Reading/Updating single elements

import com.github.arzt.tensor.TensorImplicits._
import com.github.arzt.tensor.Tensor

val t = 
  Array(
    1,2,
    3,4,
    
    5,6,
    7,8
  )
  .asTensor(2, 2, 2)