Skip to content

CLAUDE.md

Latest commit

 

History

History
91 lines (64 loc) · 3.98 KB

CLAUDE.md

File metadata and controls

91 lines (64 loc) · 3.98 KB
Raw

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

What This Package Does

EstemPMM is an R package implementing the Polynomial Maximization Method (PMM) for parameter estimation in linear regression and time series models with non-Gaussian errors. Two estimator variants:

  • PMM2 (S=2): For asymmetric (skewed) errors. Efficiency factor g2 = 1 - gamma3^2 / (2 + gamma4).
  • PMM3 (S=3): For symmetric platykurtic errors (negative excess kurtosis). Efficiency factor g3 = 1 - gamma4^2 / (6 + 9*gamma4 + gamma6).

pmm_dispatch() auto-selects between OLS, PMM2, and PMM3 based on residual cumulants.

Build & Test Commands

# Check package (R CMD check equivalent)
R CMD check .

# Build tarball
R CMD build .

# Install from source
R CMD INSTALL .

# Run all tests
Rscript -e 'testthat::test_check("EstemPMM")'

# Run a single test file
Rscript -e 'testthat::test_file("tests/testthat/test-pmm2_linear.R")'

# Regenerate roxygen docs and NAMESPACE
Rscript -e 'roxygen2::roxygenise()'

# Load package for interactive dev (without installing)
Rscript -e 'devtools::load_all()'

Architecture

R source organization (R/)

The code splits into three layers: PMM2, PMM3, and shared infrastructure.

PMM2 layer:

  • pmm2_main.R — lm_pmm2() linear regression entry point
  • pmm2_ts_main.R — Time series entry points: ar_pmm2(), ma_pmm2(), arma_pmm2(), arima_pmm2(), ts_pmm2()
  • pmm2_common.R — Core PMM2 math: moment computation, variance factor g2, coefficient adjustment
  • pmm2_classes.R — S4 classes: PMM2fit, BasePMM2, ARPMM2, MAPMM2, ARMAPMM2, ARIMAPMM2
  • pmm2_ts_methods.R — S4 method implementations (summary, coef, residuals, plot, predict, AIC) for time series PMM2 classes
  • pmm2_inference.R — Bootstrap-based inference (pmm2_inference(), ts_pmm2_inference())
  • pmm2_unified.R — Unified interface for PMM2 nonlinear estimators
  • sarimax_wrapper.R — Seasonal models: sar_pmm2(), sma_pmm2(), sarma_pmm2(), sarima_pmm2()

PMM3 layer:

  • pmm3_main.R — lm_pmm3() linear regression entry point
  • pmm3_ts_main.R — Time series entry points: ar_pmm3(), ma_pmm3(), arma_pmm3(), arima_pmm3()
  • pmm3_solver.R — Core PMM3 math: 6th-order cumulant computation, variance factor g3
  • pmm3_utils.R — pmm3_variance_factor(), pmm_gamma6(), pmm_kurtosis()
  • pmm3_classes.R / pmm3_ts_classes.R — S4 classes: PMM3fit, TS3fit and subclasses
  • pmm3_ts_methods.R — S4 methods for PMM3 time series classes

Shared:

  • pmm3_dispatch.R — pmm_dispatch() auto-selector; computes cumulants, g2/g3 factors, recommends method
  • pmm2_utils.R — compute_moments(), pmm_skewness(), pmm_kurtosis(), test_symmetry()
  • pmm2_monte_carlo.R — pmm2_monte_carlo_compare() for simulation studies
  • data.R — Bundled dataset documentation (auto_mpg)

S4 class hierarchy

All model fits use S4 classes with standard generics (summary, coef, residuals, fitted, plot, predict, AIC):

  • Linear: PMM2fit, PMM3fit
  • Time series PMM2: BasePMM2 -> ARPMM2, MAPMM2, ARMAPMM2, ARIMAPMM2 (also SARPMM2, SMAPMM2, SARMAPMM2, SARIMAPMM2)
  • Time series PMM3: TS3fit -> ARPMM3, MAPMM3, ARMAPMM3, ARIMAPMM3

Key mathematical pattern

Both PMM2 and PMM3 follow the same workflow:

  1. Fit initial OLS/MLE model
  2. Compute residual cumulants (skewness for PMM2, kurtosis+6th-order for PMM3)
  3. Calculate efficiency factor (g2 or g3); if >= 1, fall back to OLS
  4. Adjust coefficients using the PMM correction formula
  5. Return fit object with both OLS and PMM estimates

Tests

Tests use testthat (edition 3). Key test files mirror the source structure: test-pmm2_linear.R, test-pmm2_ts.R, test-pmm3_linear.R, test-pmm3_ts.R, test-pmm3_dispatch.R, test-seasonal-models.R.

Dev scripts

dev_scripts/ contains Monte Carlo simulation runners and diagnostic tools (not part of the package).