README.md

scalaopt

Scala Numerical Optimization library.

Currently this package is implementing classical methods used to solve unconstrained continuous optimization problems. All algorithms are implementing a minimize method that take as input an objective function f that we want to minimize. f is a scalar function of x a vector of continuous variables represented by any collection deriving from Seq[Double].

The following lines show how to minimize a quadratic function with the BFGS method:

scala> import com.github.bruneli.scalaopt.core._
scala> import gradient.BFGS._
scala> minimize((x: Variables) => x dot x, Vector(2.0, 4.0)) // Approximate derivatives
scala> minimize(((x: Variables) => x dot x, (x: Variables) => x * 2.0), Vector(2.0, 4.0)) // Exact derivatives

Variables is an alias for Seq[Double]. The two classes RichVector and RichMatrix provide some small DSL used internally to ease the understanding of vector and matrix algebra.

The core project has only one dependency on apache math common3 for the matrix algebra.

Below is a list of algorithms implemented:

• Quasi-Newton methods:
  • BFGS
• Conjugate gradient methods:
  • classical Conjugate Gradient
  • Newton Conjugate Gradient (line search method for large scale optimization)
  • Steihaug Conjugate Gradient (trust region method for large scale optimization)
• Least-squares problems:
  • Levenberg-Marquardt
• Derivative free methods:
  • Nelder-Mead
  • Powell

To get further details on the package you can look at the Reference Manual

scalaopt Usage

SBT

Add the following dependency to your build.sbt

libraryDependencies += "com.github.bruneli.scalaopt" % "scalaopt-core_2.11" % "0.2"

Maven

Add the following dependency to your pom file

<dependency>
    <groupId>com.github.bruneli.scalaopt</groupId>
    <artifactId>scalaopt-core_2.11</artifactId>
    <version>0.2</version>
</dependency>