Cached Iterative Weakening for Optimal Multi-Way Number Partitioning

Authors: Ethan Schreiber, Richard Korf

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In section 4, we experimentally compare CIW to the previous state of the art. We ran experiments for n from 40 to 60 and k from 3 to 12. For each combination of n and k, we generated 100 problem instances.
Researcher Affiliation Academia Ethan L. Schreiber and Richard E. Korf Department of Computer Science University of California, Los Angeles Los Angeles, CA 90095 USA {ethan,korf}@cs.ucla.edu
Pseudocode No The paper describes the algorithms in prose and through diagrams (Figure 1 and 2), but does not include structured pseudocode or an algorithm block.
Open Source Code Yes Benchmarks at https://sites.google.com/site/elsbenchmarks/. (This page further states: The source code for our current best NP solvers is located at the github repository)
Open Datasets Yes For each combination of n and k, we generated 100 problem instances. Each instance consists of n integers sampled uniformly at random in the range [1, 248 1] in order to generate hard instances without perfect partitions (Korf 2011). (Footnote 1 also points to a website that hosts these generated instances.)
Dataset Splits No The paper describes generating 100 problem instances for each combination of n and k, which are used for evaluation, but it does not specify explicit train/validation/test dataset splits for reproducibility.
Hardware Specification Yes All experiments were run on an Intel Xeon X5680 CPU at 3.33GHz.
Software Dependencies No The paper mentions using the BCP bin-packing solver from (Belov and Scheithauer 2006) but does not provide specific version numbers for any software dependencies, programming languages, or libraries.
Experiment Setup Yes For CIW, we need to choose a value for m, the number of subsets to initially generate during the precomputing phase (section 3.2). Ideally, we want m to be exactly the number of subsets in the range [perfect, C ], but we do not know this number in advance. If m is set too small, CIW will have to run ESS multiple times. If m is set too large, CIW will waste time generating subsets in the precomputing step that are never used in the iterative weakening step. For each combination of n and k, we initially set m to 10,000. Call mi the number of sets in the range [perfect, C ] for problem instance i. After instance 1 is complete, m is set to m1. After instance i is complete, m is set to the max of m1 through mi. The values of m used ranged from 24 for the second instance of (n = 56; k = 3) to 180,085 for the last eight instances of (n = 59; k = 12).