clustRviz

clustRviz provides fast computation and dynamic visualization of convex clustering solution paths by applying Algorithmic Regularization to produce efficient regularization-path estimates for high-dimensional data.

Key Features:

• Algorithmic Regularization: Applies an iterative one-step approximation scheme to compute high-quality estimates of regularization paths with rapid convergence under non-data-dependent assumptions.
• CARP (Convex Clustering via Algorithmic Regularization Paths): Implements the CARP algorithm to accelerate computation of convex clustering solutions, achieving over a 100-fold speed-up compared to existing methods as reported.
• Convex clustering-based dendrograms: Constructs dendrograms derived from convex clustering solution paths for hierarchical representation of cluster merges.
• Dynamic path-wise visualizations: Produces path-wise visualizations of clustering solutions across regularization parameters to inspect cluster evolution.
• RcppEigen implementation: Utilizes RcppEigen for internal numerical linear algebra and performance-critical computations.

Scientific Applications:

• Genomics: Exploration of clustering structure in high-dimensional genomic datasets using convex clustering solution paths.
• Text analysis: Application to clustering and structure discovery in textual data represented in high-dimensional feature spaces.
• High-dimensional data analysis: Examination of cluster formation and regularization-path behavior in large, high-dimensional biological and other scientific datasets.

Methodology:

Implements the CARP algorithm based on Algorithmic Regularization using an iterative one-step approximation scheme with a theoretical framework ensuring global convergence of the approximate solution path to the exact path, and is implemented using RcppEigen.