Hawkes Process Inference With Missing Data

Authors: Christian Shelton, Zhen Qin, Chandini Shetty

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the efficacy of our approach, and its utility in improving predictive power and identifying latent structure in real-world data. ... For evaluation, we used an exponential and a power-law kernel. We generated two different sizes of problems, each in an easy and hard version. We then tested the timeaccuracy trade-offs of the base likelihood weighting and our MCMC method on each of these eight combinations. ... To demonstrate the utility of unobserved variables, we used homicide data provided by the Chicago Police Department from 1965 through 1995 (Block, Block, and Illinois Criminal Justice Information Authority 2005) filtered to only those events reported to be gang-related.
Researcher Affiliation Academia Christian R. Shelton University of California, Riverside cshelton@cs.ucr.edu Zhen Qin University of California, Riverside tzqin001@cs.ucr.edu Chandini Shetty University of California, Riverside cshet001@cs.ucr.edu
Pseudocode No The paper describes the MCMC sampling moves (e.g., 'Move 1: Virtual Children', 'Move 2: Virtualness', 'Move 3: Parent') in narrative text but does not present them as structured pseudocode or algorithm blocks.
Open Source Code Yes The code for general inference in Hawkes processes with optional parallelization, as well as the wrapper code to run the exact experiments done here and gather the results are available at https://github.com/cshelton/hawkesinf.
Open Datasets Yes We used homicide data provided by the Chicago Police Department from 1965 through 1995 (Block, Block, and Illinois Criminal Justice Information Authority 2005) filtered to only those events reported to be gang-related.
Dataset Splits No The paper states: 'training only on crimes from 1993 and 1994. We then tested on the last year of data (1995)'. It does not explicitly mention the use of a validation set or specific split percentages for training, validation, and testing.
Hardware Specification No The paper mentions running experiments on a 'single core' for comparison but provides no specific details about the hardware, such as CPU or GPU models, or memory specifications.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as programming languages, libraries, or frameworks used for implementation.
Experiment Setup Yes The single tunable parameter is κ. We set it to 1... For the MCMC method, this computational budget included the burn-in time, which we set to be 1000 iterations for the easy problems and 5000 iterations for hard problems. ... We included an L1 regularizer on the elements of M with strength set via a search over powers of 10. ... We use a 1-dimensional line search to find β.