Bayesian online change point detection with Hilbert space approximate Student-t process

Authors: Jeremy Sellier, Petros Dellaportas

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Improvements in prediction and training time are demonstrated with real-world data sets.
Researcher Affiliation Academia 1Department of Statistical Science, University College London, UK 2Department of Statistics, Univ. of Econ. and Business, Athens, Greece 3The Alan Turing Institute, UK.
Pseudocode Yes Algorithm 1 BOCPD run length estimation; Algorithm 2 HSSPAR-CP UPM implementation
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes The Nile data set records the lowest annual water levels of the Nile river during the period 622-1284. The data has been used for change point detection in Garnett et al. (2009) and Saatc i et al. (2010). The Well Log data set contains 4050 measurements of radioactivity taken during the drilling of a well. These data have been studied in the context of change point detection by Ruanaidh & Fitzgerald (2012) and by Fearnhead & Clifford (2003).
Dataset Splits No The paper specifies training and test sets for the datasets (e.g., '200 training points, 463 Test points' for Nile Data), but does not explicitly mention a separate validation dataset split.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, cloud instances) used for running the experiments.
Software Dependencies No The paper mentions 'scipy method linalg.blas.dger for Python' but does not provide specific version numbers for these software components.
Experiment Setup Yes We use a hazard function with a trainable constant hazard rate h initialized at 100... Our implementations of HSSPAR and HSGPAR use the Hilbert space reduced-rank kernel derived from Gaussian kernels with the number of basis functions m ranging from 5 to 15. For auto-regressive UPM (GPAR and HSSPAR variants), we use lag parameter p = 1, 2, 3.