lassoinf

lassoinf provides selective inference tools leveraging a high-performance C++ backend. It is a hybrid codebase written in Python and C++, utilizing pybind11 for bindings and the Eigen library for fast linear algebra.

Features

• Affine Constraints: Core numerical tools for inference tasks.
• Dual Implementation: Logic provided in pure Python (lassoinf.affine_constraints) and fast C++ counterparts (lassoinf_cpp).
• High Performance: The C++ extensions target C++14 and integrate seamlessly with numpy and scipy.

Requirements

Python

• Python >= 3.10
• numpy, scipy, matplotlib
• meson-python, ninja, pybind11 (for building the extension)

R

• R >= 4.0
• Rcpp, RcppEigen

General

• A C++14 compliant compiler

Installation

Python

pip install lassoinf

To build from source, you can install lassoinf and build its C++ extension directly using pip:

pip install .

For an editable install during development (using the meson-python backend):

pip install --no-build-isolation -Csetup-args=-Dbuildtype=debug -e .

Editable install with uv

uv venv
source .venv/bin/activate
uv pip install meson-python meson ninja pybind11 setuptools_scm
uv pip install --no-build-isolation -e .

--no-build-isolation is required: meson-python rebuilds the C++ extension on import after C++ sources change, and that rebuild uses the build tools installed in the venv, so keep them installed. Add ".[test]" instead of . to also install the test dependencies.

If the build fails with stale paths (e.g. pybind11/pybind11.h not found after a previous non-editable install), remove the old build directory and reinstall:

rm -rf build/cp*

R

The R package lassoinf requires Rcpp, RcppEigen, and R6. You can install the package from the terminal or directly from the source directory.

The C++ library is shared with Python and must be synced first. The C++ sources live in cpp/ (used by the Python extension). R CMD build and CRAN require the R package to be self-contained, so R_pkg/src/lassoinf/ holds a copy of them that is committed to the repository. After editing anything under cpp/, refresh the copy before installing, building or testing the R package:

sh R_pkg/tools/sync_cpp.sh

Commit the updated R_pkg/src/lassoinf/ together with the change to cpp/. Otherwise, the R package builds against stale C++ code.

From the terminal:

# Build and install the package
R CMD INSTALL R_pkg

From within an R session:

# Install required dependencies if not already present
install.packages(c("Rcpp", "RcppEigen", "R6", "mvtnorm"))

# Install the package from the source directory
install.packages("R_pkg", repos = NULL, type = "source")

Or using devtools:

devtools::install("R_pkg")

Testing the R Package

To run the R test suite, you can use the testthat package from the command line:

cd R_pkg
Rscript -e 'testthat::test_dir("tests/testthat")'

Documentation

For more information, refer to the documentation in the docs/ folder. For the R package, you can view the vignette:

vignette("gaussian_lasso_boot", package = "lassoinf")

Alternatively, you can browse the pkgdown documentation site in R_pkg/docs/index.html.

License

This project is licensed under the BSD 3-Clause License. See the LICENSE file for details.