Statistically and Computationally Efficient Change Point Localization in Regression Settings
Authors: Daren Wang, Zifeng Zhao, Kevin Z. Lin, Rebecca Willett
JMLR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive numerical experiments are conducted to demonstrate the robust and favorable performance of VPWBS over two state-of-the-art algorithms, especially when the size of change in the regression coefficients {β t }n t=1 is small. ... In this section, we conduct extensive numerical experiments to examine the performance of VPWBS under various simulation settings and further compare it with two other stateof-the-art methods in the literature, specifically, EBSA in Leonardi and B uhlmann (2016) and SGL in Zhang et al. (2015b). Implementations of the numerical experiments can be found at the Git Hub link here. We discuss the implementation details such as settings for each algorithm and estimation accuracy metrics in Section 4.1 and present the simulation results in Section 4.2. |
| Researcher Affiliation | Academia | Daren Wang EMAIL Department of ACMS University of Notre Dame Indiana, USA Zifeng Zhao EMAIL Mendoza College of Business University of Notre Dame Indiana, USA Kevin Z. Lin EMAIL Wharton Department of Statistics and Data Science University of Pennsylvania Pennsylvania, USA Rebecca Willett EMAIL Department of Statistics University of Chicago Illinois, USA |
| Pseudocode | Yes | Algorithm 1 Local group Lasso based Screening. LGS ({xt, yt}n t=1, (s, e], λ). Algorithm 2 Variance-Projected Wild Binary Segmentation. VPWBS ({(am, bm]}M m=1, λ, τ, ζ) Algorithm 3 Wild Binary Segmentation via SGL. WBSSGL({(am, bm]}M m=1, λ, γ, τ, ζ) |
| Open Source Code | Yes | Implementations of the numerical experiments can be found at the Git Hub link here. |
| Open Datasets | No | The paper describes generating synthetic data for simulations rather than using a pre-existing public dataset. For example: "Specifically, we generate the data {xt, yt}n t=1 according to simulation setting (i) in Section 4.2..." |
| Dataset Splits | Yes | Specifically, given the original sample {xt, yt}n t=1, we set the training data to be the oddly-indexed observations {x2t 1, y2t 1}n/2 t=1 and the test data to be the evenly-indexed observations {x2t, y2t}n/2 t=1, where we assume, without loss of generality, n is even. |
| Hardware Specification | No | The paper does not mention any specific hardware used for running the experiments, such as CPU or GPU models, or cloud computing resources. |
| Software Dependencies | No | For each combination of (λ, γ), we solve the original SGL in (7) via the R package SGL. This mentions an R package but does not provide its version number or any other software dependencies with version information. |
| Experiment Setup | Yes | As discussed in Section 3, there are four tuning parameters (M, λ, τ, ζ) in VPWBS. Throughout the simulation section, we set M = 40 and set ζ = 5, which roughly corresponds to M = (log(n))2 and ζ = log(n) across all simulation settings in Section 4.2... For all simulation experiments in Section 4.2, we set Λ = {0.5, 1, 1.5, 2} and T = {1, 4, 7, 10, , 49}. |