Structured Bayesian Networks: From Inference to Learning with Routes

Authors: Yujia Shen, Anchal Goyanka, Adnan Darwiche, Arthur Choi7957-7965

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we demonstrate the utility of our inference algorithm, showing that it can be an order-of-magnitude more efficient than more traditional approaches to exact inference. We demonstrate the utility of our learning algorithm, showing that it can learn more accurate models and classifiers from GPS data. In Section 6, we provide an empirical analysis.
Researcher Affiliation Academia Yujia Shen, Anchal Goyanka, Adnan Darwiche, Arthur Choi Computer Science Department University of California, Los Angeles {yujias,anchal,darwiche,aychoi}@cs.ucla.edu
Pseudocode Yes Algorithm 1 Construct Decision Ctree(cluster DAG B, topological ordering π)
Open Source Code No The paper does not provide an explicit statement or link for the open-source code of the described methodology. It mentions using third-party packages like networkx, graphhopper, and GRAPHILLION.
Open Datasets Yes We obtained public map data of San Francisco (SF) from openstreetmap.org, and selected 7 increasingly larger regions of SF. We took the cabspotting dataset of GPS traces collected from taxicab routes in SF (Piorkowski, Sarafijanovoc-Djukic, and Grossglauser 2009).
Dataset Splits No The paper specifies the sizes for training (8,196 routes) and test (128 routes) sets, but does not mention the use or size of a separate validation set or a cross-validation setup.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments. It only generally refers to training and evaluation.
Software Dependencies No The paper mentions using 'the networkx python module', 'the graphhopper package', and 'the GRAPHILLION package', but it does not specify any version numbers for these software dependencies.
Experiment Setup Yes We used 8,196 routes from this dataset to learn the structure and parameters of our SBNs (using Laplace smoothing). We trained two sets of SBN parameters (using Laplace smoothing), one for each dataset, yielding a (structured) naive Bayes classifier.