Spherical Rotation Dimension Reduction with Geometric Loss Functions

Authors: Hengrui Luo, Jeremy E. Purvis, Didong Li

JMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental A comprehensive simulation study, along with a successful application to human cell cycle data, further highlights the advantages of SRCA compared to state-of-the-art alternatives, demonstrating its superior performance in approximating the manifold while preserving inherent geometric structures. [...] Section 4. Numerical Experiments
Researcher Affiliation Academia Hengrui Luo EMAIL Lawrence Berkeley National Laboratory Berkeley, CA, 94720, USA Department of Statistics, Rice University Houston, TX, 77005, USA; Jeremy E. Purvis EMAIL Department of Genetics University of North Carolina at Chapel Hill Chapel Hill, NC 27599, USA; Didong Li EMAIL Department of Biostatistics University of North Carolina at Chapel Hill Chapel Hill, NC 27599, USA
Pseudocode Yes Algorithm 1: SRCA dimension reduction algorithm; Algorithm 2: SRCA dimension reduction algorithm with l1 relaxation; Algorithm 3: SRCA dimension reduction algorithm with branch-and-bound
Open Source Code Yes Our code for SRCA implementations and experiments are publicly available at https://github.com/hrluo/Spherical Rotation Dimension Reduction.
Open Datasets Yes UCI repository (https://archive.ics.uci.edu/ml): Banknote, Climate, Concrete, Ecoli1, Leaf, Power Plant, User Knowledge. Microarray: Alon (Alon et al., 1999). GTEx (https://gtexportal.org/home/).
Dataset Splits No Although the paper mentions using "out-sample MSEs" and discusses various datasets, it does not provide specific details on the training, validation, or test split percentages or methodology (e.g., "80/10/10 split", "5-fold cross-validation").
Hardware Specification No The paper discusses computational time and complexity, but does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for the experiments.
Software Dependencies No The paper mentions using the "co Ranking R-package" but does not specify its version number or any other software dependencies with their respective versions.
Experiment Setup Yes The first parameter that dictates the behavior of most DR methods is the retained dimension d , which can be determined by subsequent purpose (e.g., the t SNE and UMAP usually take d = 2, 3 for visualizations). The second parameter is the choice of rotation methods, which is highly data dependent and affects the clustering and visualization most. [...] To account for the diverse units of the 40 selected features, we applied z-score normalization to the data. [...] In the situation where the tail behavior of the noise is close to Gaussian and the W is known (or, by default I), PCA is our default choice; but in the situation where the noise is non-Gaussian and we do not have much knowledge for W, then ICA (Hyv arinen and Oja, 2000) is a better alternative. [...] Here, we provide a new version of SRCA with sparse penalty, which only involves an additional penalty term in the loss function we designed. [...] with a tuning parameter ΞΎ > 0.