Learning Chordal Markov Networks by Dynamic Programming

Authors: Kustaa Kangas, Mikko Koivisto, Teppo Niinimäki

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

Reproducibility Variable Result LLM Response
Research Type Experimental On an eight-vertex benchmark instance, our implementation turns out to be about ten million times faster than a recently proposed, constraint satisfaction based algorithm (Corander et al., NIPS 2013). Within a few hours, it is able to solve instances up to 18 vertices, and beyond if we restrict the maximum clique size. We also study the performance of a recent integer linear programming algorithm (Bartlett and Cussens, UAI 2013). Our results suggest that, unless we bound the clique sizes, currently only the dynamic programming algorithm is guaranteed to solve instances with around 15 or more vertices.4 Experimental results
Researcher Affiliation Academia Kustaa Kangas Teppo Niinim aki Mikko Koivisto Helsinki Institute for Information Technology HIIT Department of Computer Science, University of Helsinki {jwkangas,tzniinim,mkhkoivi}@cs.helsinki.fi
Pseudocode No The paper describes the recurrence relations for its dynamic programming algorithm but does not present a formal pseudocode or algorithm block.
Open Source Code Yes Junctor is publicly available at www.cs.helsinki.fi/u/jwkangas/junctor/.
Open Datasets Yes We also evaluated both programs on several benchmark instances taken from the UCI repository [26]. The datasets are summarized in Table 1.
Dataset Splits No The paper discusses data generation and usage (e.g., '100, 1000, or 10,000 data samples' or 'synthetic data generated from Bayesian networks'), but it does not specify explicit training, validation, or test splits for its experiments.
Hardware Specification No The paper states resource limits ('4 CPU hours and 32 GB of memory', 'one week of CPU time and 128 GB of memory') but does not specify any particular hardware components like CPU or GPU models used for the experiments.
Software Dependencies No The paper mentions that 'Junctor' is implemented in C++ and compares it with 'GOBNILP', but it does not provide specific version numbers for these software components or any other dependencies.
Experiment Setup Yes For both programs, we varied the maximum width parameter w from 3 to 6 and, in addition, examined the case of unbounded width (w = ).The input for Junctor and GOBNILP was produced using the BDeu score with equivalent sample size 1.