Dual Lagrangian Learning

This repository contains code to reproduce the results of the NeurIPS 2024 paper Dual Lagrangian Learning for Conic Optimization.

• OpenReview: https://openreview.net/forum?id=gN1iKwxlL5
• NeurIPS poster: https://neurips.cc/virtual/2024/poster/94146

Installation instructions

⚠️ This code requires the CUDA.jl package. It has been tested only on an NVIDIA V100 GPU, with the provided julia environment.

1. Install julia
2. Install commercial optimization solvers
   • Gurobi: see Gurobi.jl instructions. Gurobi provides free academic licenses.
   • Mosek: see Mosek.jl instructions. Mosek provides free academic licenses.
3. Install julia packages

   julia --project=. -e 'using Pkg; Pkg.instantiate(); Pkg.build()'

Running numerical experiments

Multi-dimensional knapsack

Slurm files are located in exp/multiknapsack/multiknapsack.sbatch.

Production planning

Slurm files are located in exp/rcprod/rcprod.sbatch.