Modelling and Solving Online Optimisation Problems
Authors: Alexander Ek, Maria Garcia de la Banda, Andreas Schutt, Peter J. Stuckey, Guido Tack1477-1485
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show we can automatically create efficient online and sliding-window solutions to optimisation problems. |
| Researcher Affiliation | Collaboration | Alexander Ek,1,2 Maria Garcia de la Banda,1 Andreas Schutt,2,3 Peter J. Stuckey,1,2 Guido Tack1,2 1Monash University, Australia 2CSIRO Data61, Australia 3The University of Melbourne, Australia |
| Pseudocode | Yes | Figure 2: Solving online problems iteratively. online-solve(M): while (new data) D := get current data() θ := get current soln() D := constrain(D, θ) θ := solve(instantiate(M, D )) output θ |
| Open Source Code | Yes | Code available at https://gitlab.com/minizinc/online-minizinc |
| Open Datasets | Yes | The online data used is constructed from the offline data file abz5 of the Mini Zinc benchmarks3 by repeatedly adding copies of the jobs from abz5 into an endless queue. ... 3https://github.com/Mini Zinc/minizinc-benchmarks/tree/master/jobshop (21-Nov-2019) |
| Dataset Splits | No | The paper describes how data is generated and consumed for iterative solving (e.g., 'The first i jobs in the queue form the initial iteration'), but it does not specify explicit train/validation/test dataset splits with percentages, counts, or references to predefined splits for model evaluation in a machine learning context. |
| Hardware Specification | Yes | All experiments were run on a 2.2 GHz Intel Core i7 processor with 16 GB RAM |
| Software Dependencies | Yes | using the lazy clause generation solver Chuffed (version bundled with the Mini Zinc IDE Version 2.2.3). The Mini Zinc interface for Python was used to simulate data changes between iterations |
| Experiment Setup | Yes | Each task duration is given as an expected value of an unknown distribution, which means that its realisation may differ from its given value. As a result, the realisation of a task s start time is also uncertain, since it may have to be pushed forward during execution, due to a predecessor task running slower than expected. We simulate the execution step in this experiment as follows. ... generated from a normal distribution with the mean parameter as the original task duration ... and the standard deviation as one tenth of the average task duration in the instance file (ft20). |