This repository contains a header-only C++ library implementing the Augmented Implicitly Restarted Lanczos Bidiagonalization Algorithm (IRLBA) from Baglama and Reichel (2005). IRLBA is a fast and memory-efficient method for truncated singular value decomposition, and is particularly useful for approximate principal components analysis of large matrices. The code here is derived from the C code in the irlba R package, refactored to use the Eigen library for matrix algebra.

Using this library is as simple as including the header file in your source code:

#include "irlba/irlba.hpp"

irlba::Options opt;
// optional; specify the workspace, etc.
opt.extra_work = 20;
opt.max_iterations = 50;

// Define an Eigen matrix for the input.
Eigen::MatrixXd mat;

// Get the first 5 singular triplets:
auto result = irlba::compute_simple(mat, 5, opt);
result.U; // left singular vectors
result.V; // right singular vectors
result.D; // singular values

To perform a PCA:

// Get the first 5 principal components without scaling:
auto pcres = irlba::pca(mat, true, false, 5, opt);
pcres.components;
pcres.rotation;
pcres.variances;

See the reference documentation for more details.

The irlba::Matrix interface allows us to use different matrix representations for irlba::compute(). For example, we can defer column-centering to efficiently perform PCA on a sparse matrix:

// Some sparse matrix.
Eigen::SparseMatrix<double> spmat;

// Wrap the Eigen matrix in a wrapper for compatibility with irlba::Matrix.
irlba::SimpleMatrix<Eigen::VectorXd, Eigen::MatrixXd, decltype(&spmat)> wrapped(&spmat);

// Define column centers.
Eigen::VectorXd center;

// Create a matrix where the column-centering is performed during multiplication.
// This avoids instantiating the actual centered matrix, which would lose sparsity.
irlba::CenteredMatrix<Eigen::VectorXd, Eigen::MatrixXd> centered(&wrapped, &center);

auto centered_res = irlba::compute(centered, 5, opt);

We provide several subclasses that implement the irlba::Matrix interface:

• SimpleMatrix, which wraps existing Eigen matrices.
• CenteredMatrix, for deferred centering of columns.
• ScaledMatrix, for deferred scaling of rows or columns.
• ParallelSparseMatrix, for parallelized multiplication of a sparse matrx.

Developers can easily create their own Matrix subclass by implementing methods for matrix-vector multiplication. For example, the scran_pca library performs PCA on a matrix of residuals without ever explicitly creating that matrix.

If you're using CMake, you just need to add something like this to your CMakeLists.txt:

include(FetchContent)

FetchContent_Declare(
  irlba 
  GIT_REPOSITORY https://github.com/LTLA/CppIrlba
  GIT_TAG master # or any version of interest
)

FetchContent_MakeAvailable(irlba)

Then you can link to irlba to make the headers available during compilation:

# For executables:
target_link_libraries(myexe ltla::irlba)

# For libaries
target_link_libraries(mylib INTERFACE ltla::irlba)

find_package(ltla_irlba CONFIG REQUIRED)
target_link_libraries(mylib INTERFACE ltla::irlba)

To install the library use:

mkdir build && cd build
cmake .. -DIRLBA_TESTS=OFF
cmake --build . --target install

By default, this will use FetchContent to fetch all external dependencies. If you want to install them manually, use -DPOWERIT_FETCH_EXTERN=OFF. See extern/CMakeLists.txt to find compatible versions of each dependency.

If you're not using CMake, the simple approach is to just copy the files - either directly or with Git submodules - and include their path during compilation with, e.g., GCC's -I. Note that this requires the dependencies listed in extern/CMakeLists.txt.

Baglama, James and Reichel, Lothar (2005). Augmented implicitly restarted Lanczos bidiagonalization methods. SIAM J. Sci. Comput., 27(1), 19-42.