Warmstarting of Model-Based Algorithm Configuration

Authors: Marius Lindauer, Frank Hutter

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments for optimizing a flexible modern SAT solver on twelve different instance sets show that our methods often yield substantial speedups over existing AC methods (up to 165-fold) and can also find substantially better configurations given the same compute budget.
Researcher Affiliation Academia Marius Lindauer, Frank Hutter University of Freiburg {lindauer,fh}@cs.uni-freiburg.de
Pseudocode Yes Algorithm 1: Model-based Algorithm Configuration
Open Source Code Yes 3Code and data is publicly available at: http://www.ml4aad.org/smac/.
Open Datasets Yes Further details on these instances are given in the description of the configurable SAT solver challenge (Hutter et al. 2017), which are publicly available in the algorithm configuration library (Hutter et al. 2014a).
Dataset Splits Yes To avoid overfitting of the weights, we randomly split the current H into a training and validation set (2 : 1), use the training set to fit ˆc, and then compute predictions of ˆc and each ˆci on the validation set, which are used to fit the weights w.
Hardware Specification Yes All runs were run on a compute cluster with nodes equipped with two Intel Xeon E5-2630v4 and 128GB memory running Cent OS 7.
Software Dependencies Yes we ran SMAC (0.5.0) and our warmstarting variants
Experiment Setup Yes On each AC task, we ran 10 independent SMAC runs with a configuration budget of 2 days each. As a cost metric, we chose the commonly-used penalized average runtime metric (PAR10, i.e., counting each timeout as 10 times the runtime cutoff) with a cutoff of 300 CPU seconds.