Setup

git clone https://bitbucket.org/djhshih/dppmix
git submodule update --init

library(devtools)
document()
install()

Code architecture

The package is organized into two main layers:

• R/: user-facing R API, model orchestration, and thin wrappers around compiled code
• src/: performance-critical C++ implementations exposed through Rcpp

R layer

Main files in R/:

• R/dppmix_mcmc_mvnorm.R: top-level MCMC fitting function and MCMC update logic for the multivariate normal DPP mixture model
• R/dppmix_em_mvnorm.R: EM-based fitting logic and objective/gradient orchestration
• R/dppmix_em_mvnorm_original.R: original pure-R implementations retained for reference and equivalence testing
• R/dppmix_em_mvnorm_cpp.R: thin R wrappers for compiled EM/covariance helpers
• R/estimate.R, R/simulate.R, R/math.R: estimation helpers, simulation helpers, and shared math utilities

There is also a submodule-like shared core under R/dppcore/, currently used for reusable covariance-related code.

C++ layer

Main files in src/:

• src/covariance.cpp: compiled covariance-kernel helpers
• src/ldet.cpp: compiled log-determinant helper
• src/dppmix_em_mvnorm.cpp: compiled EM update/objective/gradient helpers

The compiled interface follows the naming convention already used by the EM code:

• original R implementation: suffix _original
• R wrapper for compiled code: suffix _cpp
• C++ implementation exported through Rcpp: suffix _cpp_impl

Example:

• dppmix_em_mvnorm_update_gamma_original in R
• dppmix_em_mvnorm_update_gamma_cpp in R
• dppmix_em_mvnorm_update_gamma_cpp_impl in C++

Build/bindings

Rcpp bindings are generated into:

• R/RcppExports.R
• src/RcppExports.cpp

The package registers native routines through NAMESPACE with useDynLib(dppmix, .registration=TRUE).

Current porting status

• EM and covariance hot paths already have compiled implementations in src/
• the multivariate-normal MCMC path now follows the same split architecture:
  • R/dppmix_mcmc_mvnorm.R: public API and orchestration
  • R/dppmix_mcmc_mvnorm_original.R: original R helper implementations retained for reference and regression testing
  • R/dppmix_mcmc_mvnorm_cpp.R: thin wrappers around compiled MCMC helpers
  • src/dppmix_mcmc_mvnorm.cpp: compiled MCMC implementations

Compiled MCMC helpers currently cover:

• update_z
• update_Sigma
• compute_Sk
• llikelihood
• mu_is_valid
• update_lambda
• update_mu
• split_move
• combine_move
• update_K

Tests in tests/testthat/test-mcmc-cpp-equivalence.R exercise deterministic equivalence, stochastic-kernel invariants, RJMCMC move validity, and short smoke-style regression checks.