Numo::Linalg : Linear Algebra library with BLAS/LAPACK binding to Numo::NArray

GitHub | RubyGems

Under development!

Introduction

This is a binding of BLAS/LAPACK for Numo::NArray using dynamic linking loader. This design allows you to change backend libraries without re-compiling.

Numo::Linalg API

• Matrix and vector products
  • dot, matmul
• Decomposition
  • lu, lu_fact, lu_inv, lu_solve, ldl, cholesky, cho_fact, cho_inv, cho_solve, qr, svd, svdvals, orth, null_space
• Matrix eigenvalues
  • eig, eigh, eigvals, eigvalsh
• Norms and other numbers
  • norm, cond, det, slogdet, matrix_rank, matrix_power
• Solving equations and inverting matrices
  • solve, lstsq, inv, pinv

Low-level modules

• Numo::Linalg::Blas - Low-level BLAS functions
• Numo::Linalg::Lapack - Low-level LAPACK functions

Installation

• Install Numo::NArray

• Install LAPACK or alternative package.

  • Numo::Linalg requires C-interface CBLAS and LAPACKE interface. These are included in LAPACK package.

  • Recommended: use one of following faster libraries:

    • ATLAS
    • OpenBLAS
    • Intel MKL

  • Note that the performance depends on the backend library as shown in benchmark.

• Install Numo::Linalg

$ gem install numo-linalg

or

$ git clone https://github.com/ruby-numo/numo-linalg.git
$ cd linalg
$ rake build
$ gem install pkg/numo-linalg-*.gem

Using

• Load Numo::Linalg module with default backend:

require "numo/linalg"

• Read also instruction for Selecting Backend Library.

Authors

• Masahiro Tanaka
• Makoto Kishimoto
• Atsushi Tatsuma

Acknowledgments

• This work is partly supported by 2016 Ruby Association Grant.

ToDo

• More functions
• write test
• Documentation