Optimal Torpedo Scheduling
Authors: Adrian Goldwaser, Andreas Schutt
JAIR 2018 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We propose an exact solution method based on Logic-based Benders Decomposition using Mixed-Integer and Constraint Programming, which optimally solves and proves, for the first time, the optimality of all instances from the ACP Challenge 2016 within 10 minutes. In addition, we adapted our method to handle large-scale instances and instances with a more general rail network. This adaptation optimally solved all challenge instances within one minute and was able to solve instances of up to 100,000 hot metal pickups. [...] 5. Experiments We conducted experiments on a variety of instances, either taken from the ACP 2016 Challenge, created using the instance generator provided at the ACP Challenge website or, in the case of the extended network, our own instance generator. [...] Table 3 shows the results on the set comp for each instance when using opportunistic partitioning (Algorithm 6). |
| Researcher Affiliation | Academia | Adrian Goldwaser EMAIL Data61, CSIRO, & The University of New South Wales Sydney, NSW, Australia Andreas Schutt EMAIL Data61, CSIRO, & The University of Melbourne Melbourne, VIC, Australia |
| Pseudocode | Yes | Algorithm 1: Assignment of torpedo runs to torpedoes. [...] Algorithm 2: Computation of departure times from the oxygen converter. [...] Algorithm 3: Resource overload check. [...] Algorithm 4: Computation of arrival times at the blast furnace. [...] Algorithm 5: Computation of a backward matching. [...] Algorithm 6: Computation of the sub-problems via opportunistic partitioning. [...] Algorithm 7: get Profile(start, end) : Computation of the resource profiles. [...] Algorithm 8: Computation of the sub-problems via overload partitioning. |
| Open Source Code | No | No explicit statement about the release of source code for the methodology described in the paper was found. The provided link 'https://github.com/AdGold/TorpedoSchedulingInstances' refers to the instances used in experiments, not the source code itself. |
| Open Datasets | Yes | We conducted experiments on a variety of instances, either taken from the ACP 2016 Challenge, created using the instance generator provided at the ACP Challenge website or, in the case of the extended network, our own instance generator. We group all instances into two benchmark sets called ACPNet and Ext Net. All instances are available at https://github.com/AdGold/TorpedoSchedulingInstances. |
| Dataset Splits | No | The paper describes solving instances from benchmark sets (ACPNet, Ext Net) and mentions using an instance generator. However, it does not specify explicit training/test/validation splits for the data in the context of machine learning model evaluation. The problem is an optimization problem where instances are solved, not split for training/testing. |
| Hardware Specification | Yes | We ran all our experiments on Dell Power Edge M630 machines having Intel Xeon E5-2660 V3 processors running at 2.6 Ghz with a 25 MB cache and 8GB RAM, unless otherwise stated. [...] Note that for the large size instances in large, the runtime was more than 30 hours and it had to be run on a machine with extra RAM in order to be able to solve the MIP. This machine had the same specifications as the others except that it used 128GB of RAM instead. |
| Software Dependencies | Yes | Our solution method was implemented in Python 3.5.1 interfacing Gurobi 7.0.2 using the Python library gurobipy and Chuffed using system calls. |
| Experiment Setup | Yes | Gurobi was used for solving the MIP problem and executed with the default settings and on 2 cores. Chuffed (Chu, 2011) was used for solving the CP problems. The time-tabling propagator with explanation generation as described in (Schutt, 2011; Schutt, Feydy, Stuckey, & Wallace, 2011) was used for all cumulative constraints in the CP model. We impose a runtime limit of 5 minutes for Chuffed for solving any CP problem. No runtime limit was imposed for solving the MIP problem. |