Value-Directed Compression of Large-Scale Assignment Problems

Authors: Tyler Lu, Craig Boutilier

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on marketing contact optimization and political legislature problems validate the performance of our technique. The experiments in this section illustrate the performance and value of DCA. We begin with MMMOPs on customer data sets of various sizes (250K, 500K, 750K, 1M, 2M, 4M, 6M, and 10M) using a set of 20 campaigns and 5 channels.
Researcher Affiliation Academia Tyler Lu and Craig Boutilier Department of Computer Science University of Toronto {tl,cebly}@cs.toronto.edu
Pseudocode No The paper describes its methods through prose and mathematical formulations (e.g., LP models and equations) but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured, code-like procedural steps.
Open Source Code No The paper mentions that they 'have implemented DCA as a series of map-reduce operations using Apache Spark... on top of Hadoop s distributed filesystem', and that 'All LPs use Gurobi Optimizer 5.6'. However, it does not provide any explicit statement about releasing their own source code or a link to a repository containing it.
Open Datasets Yes We tested DCA on a political preference data set derived from a questionnaire completed by 1.2M voters prior to the 2013 national election in Australia... Administered by Vote Compass (http://votecompass.com/), a widely used interactive electoral literacy application. We thank... Clifton van der Linden and the Vote Compass team for making their dataset available.
Dataset Splits No The paper mentions using 'customer data sets of various sizes' and a 'political preference data set' for its experiments, but it does not specify any training, validation, or test dataset splits, percentages, or sample counts, nor does it refer to pre-defined splits.
Hardware Specification Yes All LPs use Gurobi Optimizer 5.6 on a single high-performance, large memory compute node (dual Intel 2.6GHz processors, 244 Gb RAM); sufficient statistic computation is distributed across a variable number of compute nodes.
Software Dependencies Yes All LPs use Gurobi Optimizer 5.6
Experiment Setup Yes All DCA splits are binary and are restricted to medians (quantile 0.5). To test scenario analysis, we applied DCA to the simultaneous solution of 1M-customer instances using six lead limits: a base level as above, with five additional LPs solved with different lead limits (0.5, 2, 3, 4 and 5 times the base level for all global and campaign-specific limits).