Multiview Clustering Simulation

This repository contains a C++ implementation of a Multiview Gibbs Sampler for clustering, along with R scripts to interface with and run the simulations.

Project Structure

The project has recently been refactored into the following structure:

• src/: Contains all the raw C++ .cpp and .h source files.
• scripts/: Contains the R scripts (e.g., CPP_Simulation.R, real.R) used to prepare data, run the C++ sampler, and analyze the results.
• dataset/: Contains the datasets used by the simulations.

Prerequisites

To run the simulations, you will need:

• R installed on your system.
• The Rcpp package installed in R: install.packages("Rcpp")
• Various other R libraries for clustering and visualization, such as dplyr, mcclust, ggplot2, etc. (Check individual simulation scripts for full library requirements).

How to Run Your First Simulation

1. Set your Working Directory: Everything runs from the Multiview/ folder. Ensure your R session is working out of this directory:

   setwd("path/to/Multiview-Clustering/Multiview")

2. Run an R Script: You can run a simulation script directly using Rscript from the command line, or by executing the file in RStudio.

   From the command line (while in the Multiview/ folder):

   Rscript scripts/CPP_Simulation.R

   The R scripts are programmed to compile the C++ source files dynamically on the fly via Rcpp::sourceCpp("src/multiview_gibbs.cpp").

Adjusting Hyperparameters

The run_gibbs_cpp function inside the simulation scripts accepts initial hyperparameters that configure the behavior of the sampler from the R interface:

res_gibbs <- run_gibbs_cpp(
    data_views = data_views,
    M          = 10000,
    burn_in    = 1000,
    thin       = 5,
    alpha_global_init = 1.0,
    sigma_global_init = 0.6,
    alpha_v_init = c(1.0, 1.5), # Vector for View configuration
    sigma_v_init = c(0.5, 0.4), # Vector for View configuration
    a_tau_prior = 2.0,          # Tau prior control
    b_tau_prior = 1.0
)

Available Hyperparameters:

• alpha_global_init: Initial value for the global concentration parameter. Defaults to 1.0.
• sigma_global_init: Initial value for the global discount parameter. Defaults to 0.6.
• alpha_v_init: Vector of initial values for view-specific concentration. Defaults to 1.0.
• sigma_v_init: Vector of initial values for view-specific discount. Defaults to 0.5.
• tau_v_init: Vector of initial values for view-specific precision. Defaults to an empirically driven estimate.
• a_tau_prior: Shape parameter for the Inverse-Gamma prior on tau. Defaults to 2.0.
• b_tau_prior: Scale parameter for the Inverse-Gamma prior on tau. Defaults to 1.0.
• K_init_tables: Initial number of global tables. Defaults to 4.
• K_init_dishes: Initial number of dishes per view. Defaults to 2.