Change Point Detection via Multivariate Singular Spectrum Analysis

Authors: Arwa Alanqary, Abdullah Alomar, Devavrat Shah

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct empirical experiments using benchmark and synthetic datasets. We find that the proposed method performs competitively or outperforms the state-of-the-art change point detection methods across datasets.
Researcher Affiliation Academia Arwa Alanqary Computational Science and Engineering MIT alanqary@mit.edu Abdull ah Alomar MIT aalomar@mit.edu Devavrat Shah MIT devavrat@mit.edu
Pseudocode Yes Algorithm 1 in Appendix C for a summary of the algorithm. ... Algorithm 2 in Appendix C.
Open Source Code No The paper does not provide an explicit statement or link to the source code for the authors' methodology.
Open Datasets Yes The evaluation is conducted on four benchmark datasets (see Table 2). Details of these datasets, their sources, and physical interpretations of change points are given in Appendix D.2. ... Beedance data [33], HASC dataset [18], Occupancy data [6], Yahoo! Webscope S5 [15].
Dataset Splits No The paper describes evaluation metrics and how true positives are determined for change point detection, but does not specify train/validation/test dataset splits (e.g., percentages or counts) for model training or tuning.
Hardware Specification No The paper does not mention any specific hardware (e.g., GPU/CPU models, memory, or cloud instance types) used for running its experiments.
Software Dependencies No The paper mentions external software packages (e.g., 'changepoint: An r package' [21], 'Microsoft. nimbusml package' [29]) but does not provide specific version numbers for these or any other software dependencies used in their implementation or experiments.
Experiment Setup Yes Our m SSA algorithm for CPD has five parameters: (1) the number of initial observations T0 used to estimate the subspace L0, (2) the lag parameter 1 < L T0, (3) the estimated order of the time series ˆk >0, (4) the CUSUM test threshold h>0, and (5) the shift-downwards constant c 0. ... For simplicity, we fix T0=200, L=20, k=10 across all experiments... The parameters c and h are selected through grid search.