CausalMixGPD

CausalMixGPD provides Dirichlet process mixture modeling (CRP, stick-breaking, and spliced variants) with optional generalized Pareto tails and a unified prediction/causal inference API.

Installation

# install.packages("remotes")
remotes::install_github(
  "arnabaich96/CausalMixGPD",
  build_vignettes = TRUE,
  INSTALL_opts = c("--html")
)

Recent Changes (2026-04)

• Added support for lognormal threshold link-distributions in GPD workflows, including spliced backend behavior.
• Improved initialization for stability with covariate-aware threshold/link seeding and stronger latent label starts.
• Added CRP retry logic for rare all--Inf initialization failures during MCMC startup.
• Updated wrappers/tests/manuscript examples to align with the new threshold-link and initialization behavior.

Documentation

• Function reference

Additional Docs

• Package roadmap

Validation Notes

• Performance acceptance tests and benchmark scripts are under tests/perf/.
• Main acceptance test entrypoint: tests/testthat/test-performance-acceptance.R.

Use DPMIXGPD_TEST_LEVEL=ci for CI-level acceptance checks and DPMIXGPD_TEST_LEVEL=full for full seeded-equivalence runs.