Neur2SP: Neural Two-Stage Stochastic Programming

Authors: Rahul Mihir Patel, Justin Dumouchelle, Elias Khalil, Merve Bodur

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our extensive computational experiments on four benchmark 2SP problem classes with different structures (containing MIP and NLP second-stage problems) demonstrate the efficiency (time) and efficacy (solution quality) of Neur2SP.
Researcher Affiliation Academia Justin Dumouchelle Rahul Patel Elias B. Khalil Merve Bodur Department of Mechanical & Industrial Engineering, University of Toronto
Pseudocode No The paper describes the framework, neural network architectures, and MIP formulation mathematically and verbally, but does not include any labeled pseudocode or algorithm blocks.
Open Source Code Yes The code to reproduce all of the experiments is available at https://github.com/khalil-research/Neur2SP.
Open Datasets Yes We evaluate our approach on four 2SP problems that are commonly considered in the literature: a two-stage stochastic variant of the Capacitated Facility Location Problem (CFLP) [Cornuéjols et al., 1991], an Investment Problem (INVP) [Schultz et al., 1998], the Stochastic Server Location Problem (SSLP) [Ntaimo and Sen, 2005], and the Pooling Problem (PP) [Audet et al., 2004, Gupte et al., 2017, Gounaris et al., 2009, Haverly, 1978].
Dataset Splits No While the paper mentions the use of a 'validation set' for model selection, it does not specify the exact percentages or counts for the training, validation, and test dataset splits.
Hardware Specification Yes All experiments were run on a computing cluster with an Intel Xeon CPU E5-2683 and Nvidia Tesla P100 GPU with 64GB of RAM (for training).
Software Dependencies Yes Gurobi 9.1.2 [Gurobi Optimization, LLC, 2021] was used as the MIP solver. Scikit-learn 1.0.1 [Pedregosa et al., 2011] and Pytorch 1.10.0 [Paszke et al., 2019] were used for supervised learning.
Experiment Setup No The paper mentions using 'random search over 100 hyperparameter configurations for model selection' but does not explicitly provide specific hyperparameter values (e.g., learning rate, batch size, epochs, optimizer settings) or other detailed training configurations.