Stratification for Constraint-Based Multi-Objective Combinatorial Optimization

Authors: Miguel Terra-Neves, Inês Lynce, Vasco Manquinho

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental An extensive experimental evaluation on publicly available MOCO instances shows that the new algorithm is competitive with stochastic methods and it is much more effective than existing constraint-based methods.
Researcher Affiliation Academia Miguel Terra-Neves, Inˆes Lynce, Vasco Manquinho INESC-ID / Instituto Superior T ecnico, Universidade de Lisboa, Portugal {neves,ines,vmm}@sat.inesc-id.pt
Pseudocode Yes Algorithm 1: Generic stratified MCS algorithm; Algorithm 2: Stratified MCS-based MOCO algorithm
Open Source Code No The paper does not provide a statement or link indicating that the source code for the methodology described is open-source or publicly available. It mentions using an existing library, Sat4j-PB.
Open Datasets Yes The evaluation is performed on publicly available VMC benchmarks2, based on subsets of workload traces randomly selected from the Google Cluster Data project3. Footnotes 2 and 3 provide URLs: '2http://sat.inesc-id.pt/dome/' and '3http://code.google.com/p/googleclusterdata/'.
Dataset Splits No The paper mentions using publicly available benchmarks but does not specify the training, validation, or test split percentages or sample counts for these datasets. It only mentions 'subsets of workload traces randomly selected'.
Hardware Specification No The paper states 'Each algorithm was ran with a memory limit of 4 GB and a time limit of 1800 seconds' but does not specify any particular CPU, GPU, or other hardware model used for the experiments.
Software Dependencies Yes All algorithms are implemented in Java and Sat4j-PB v.2.3.4 [Le Berre and Parrain, 2010] is used as satisfiability checker.
Experiment Setup Yes Moreover, we merge partitions whenever the satisfiability checker reaches a fixed number of conflicts (200000 in our experiments). Each algorithm was ran with a memory limit of 4 GB and a time limit of 1800 seconds. SCLD was configured with the LWR partitioning heuristic (β = 15) and the MERGE division reduction strategy. MOEAD was configured with crossover and mutation rates of 0.8 and 0.05 respectively, a population size of 100, a neighborhood size of 20, a 0.9 probability of crossover with individuals from the neighborhood and a maximum of 2 individuals that can be replaced by a single offspring.