CapsAndRuns: An Improved Method for Approximately Optimal Algorithm Configuration

Authors: Gellert Weisz, Andras Gyorgy, Csaba Szepesvari

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments verify that our method can significantly outperform its competitors.
Researcher Affiliation Collaboration 1Deep Mind, London, UK. 2On leave from Imperial College London, London, UK. 3On leave from University of Alberta, Edmonton, AB, Canada.
Pseudocode Yes Algorithm 1 CAPSANDRUNS, Algorithm 2 QUANTILEEST, Algorithm 3 RUNTIMEEST
Open Source Code No The paper mentions using the "open-source minisat solver" but does not state that the code for CAPSANDRUNS or their experiments is open-source or provide a link to it.
Open Datasets Yes In the experiments we used the same benchmark dataset as in our previous paper (Weisz et al., 2018). The dataset contains runtimes of 972 different configurations of minisat on a set of 20118 SAT problems generated using CNFuzz DD,7 with a timeout of 15 CPU minutes. 7http://fmv.jku.at/cnfuzzdd/
Dataset Splits No The paper mentions using a set of 20118 SAT problems but does not provide specific details on how this dataset was split into training, validation, or test sets.
Hardware Specification No The paper does not provide any specific hardware details (e.g., CPU, GPU models, memory) used for running the experiments or simulations.
Software Dependencies No The paper mentions various tools and solvers like "minisat solver (Sorensson & Een, 2005)", "SMAC (Hutter et al., 2011; 2013)", "Param ILS (Hutter, 2007; Hutter et al., 2009)", "GGA (Ans otegui et al., 2009; 2015)", and "irace (Birattari et al., 2002; L opez-Ib anez et al., 2011)", but it does not specify version numbers for these or other software dependencies.
Experiment Setup Yes We simulated runs of STRUCTURED PROCRASTINATION (Kleinberg et al., 2017), LEAPSANDBOUNDS (Weisz et al., 2018), and CAPSANDRUNS (this work), with parameters ε = 0.05, δ = 0.2, and error probability 0.1 (ζ = 1/60)