Emulating the Expert: Inverse Optimization through Online Learning

Authors: Andreas Bärmann, Sebastian Pokutta, Oskar Schneider

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Preliminary Computational Tests. While a full computational study is beyond the scope of this paper and left for future work, we implemented a first preliminary version of our algorithm, and we report computational results for a few select problems. and In Table 1, we show the computational results for the Integer Knapsack Problem with n = 1000 items.
Researcher Affiliation Academia 1Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany 2Georgia Institute of Technology, Atlanta, USA.
Pseudocode Yes Algorithm 1 Online Objective Function Learning
Open Source Code No The paper states: 'We have implemented our framework using python and Gurobi 7.0.1 (Gurobi Optimization, Inc., 2016).' but does not provide any link or explicit statement about making their own code open source.
Open Datasets No The paper states: 'We generated random instances for our computational results, considering T = 1000 observations for a varying number of goods n 2 {100, 500, 1000}.' It describes the generation process but does not provide access to a public dataset.
Dataset Splits No The paper describes an online learning setting where observations are revealed over time, and it does not specify explicit training, validation, and test dataset splits in a traditional sense.
Hardware Specification Yes Our preliminary computational experiments have been obtained on a Mac Book Pro (2016) with an Intel Core i5 CPU with two 2.00 GHz cores.
Software Dependencies Yes We have implemented our framework using python and Gurobi 7.0.1 (Gurobi Optimization, Inc., 2016).
Experiment Setup No The paper describes how problem instances were generated (e.g., 'The customer s unknown utility vector is chosen at random as (arbitrary) integer numbers from the interval [1, 1000] from a uniform distribution'), and refers to 'T = 1000 observations', but does not specify explicit hyperparameters or system-level training settings for their algorithm, such as a fixed learning rate or optimizer configurations.