PANOC-lite experiments

This repository contains the source code to reproduce the experiments in the following manuscript:

A. Bodard, P. Pas, A. Themelis, P. Patrinos,

PANOC-lite: A simpler and more efficient algorithm for composite minimization.

Instructions

# Ensure submodules are present
git submodule update --init --recursive

# Generate the OCP benchmark problems
uv sync
uv run --package alpaqa-mpc-benchmarks --extra test pytest
. .venv/bin/activate
export CFLAGS=-march=native; export CXXFLAGS=-march=native
cmake --fresh -B build -D CMAKE_BUILD_TYPE=Release -G Ninja
cmake --build build -j
cmake --install build --prefix problems/staging

# Install the alpaqa Julia interface
julia -e "using Pkg; Pkg.add(\"BinaryBuilder\")"
julia alpaqa/interfaces/julia/build_tarballs.jl $(gcc -print-multiarch) --deploy=local
julia --project=alpaqa/interfaces/julia/Alpaqa -e "
using Pkg
Pkg.develop(path=joinpath(DEPOT_PATH[1], \"dev\", \"Alpaqa_jll\"))
Pkg.instantiate()
"
julia --project=. -e "
using Pkg
Pkg.develop(path=\"./alpaqa/interfaces/julia/Alpaqa\")
Pkg.instantiate()
"
# Sanity check for loading problems and running alpaqa
julia --project=. benchmarks/alpaqa.jl

Results

MPC experiments

To reproduce the MPC experiments, involving a quadcopter and a bicycle model, run

julia --project=. paper-experiments/mpc_plots.jl

Sparse logistic regression

To reproduce the sparse logistic regression results, run

julia --project=. paper-experiments/sparse_logistic_regression_tables.jl

The box plot and performance profile from the paper can then be produced by respectively running

julia --project=. paper-experiments/sparse_logistic_regression_box_plot.jl

and

julia --project=. paper-experiments/sparse_logistic_regression_performance_profile.jl

CUTEst experiments

To reproduce the CUTEst experiments, run

julia --project=. paper-experiments/cutest.jl

The performance profile from the paper can then be plotted by running

julia --project=. paper-experiments/cutest_performance_profiles.jl