Landscape Surrogate: Learning Decision Losses for Mathematical Optimization Under Partial Information
We are introducing LANCER, a versatile framework designed to tackle challenging optimization problems, such as those found in nonlinear combinatorial problems, smart predict+optimize framework, etc.
This source code accompanies our paper 📜: https://arxiv.org/abs/2307.08964
Please 🌟star🌟 this repo and cite our paper 📜 if you like (and/or use) our work, thank you!
Arman Zharmagambetov (First Author), Brandon Amos, Aaron Ferber, Taoan Huang, Bistra Dilkina, Yuandong Tian (Principal Investigator)
This release contains our implementation of LANCER and its application to 1) learning surrogates for mixed-integer nonlinear programming (MINLP) and; 2) smart Predict+Optimize (a.k.a. decision-focused learning or DFL), but it can also be applied to a range of other large-scale optimization problems, such as hyper-parameter optimization, model-based reinforcement learning, etc.
Please follow the steps in installation.md to setup the environment.
MINLP/README.md contains instructions to validate LANCER for MINLP tasks. We evaluate on two benchmarks: Stochastic Shortest Path and Combinatorial Portfolio Optimization/Selection with 3rd order objective.
DFL/README.md contains instructions to validate LANCER for DFL tasks. We evaluate on three benchmarks: Shortest Path, Multidimensional Knapsack and Portfolio Optimization/Selection.
Our source code is under CC-BY-NC 4.0 license.