Changelog

Source: NEWS.md

degen 0.15.0

• Code elegance improvements:
  • Extracted format_eigenvector() helper to eliminate duplicate code in identifiability analysis
  • Refactored find_identified_functions() and describe_null_directions() to use shared helper
• Fixed bug where verbose = TRUE output was silently swallowed in equivalence_classes() when comparing fewer than 5 model pairs

degen 0.14.0

• Added sensitivity analysis functions:
  • sensitivity_analysis() for testing robustness to tolerance and grid size
  • bootstrap_equivalence() for bootstrap testing of equivalence claims
  • cross_validate_equivalence() for cross-validation of equivalence
  • Plot methods for all sensitivity results
• Added Bayesian model comparison:
  • compare_posteriors() for comparing models using MCMC samples
  • prior_sensitivity() for testing how priors affect identifiability
  • extract_samples() for extracting posterior samples from rstan, brms, rstanarm
• Added symbolic analysis:
  • symbolic_fisher() for analytical Fisher information (normal, exponential, gamma, etc.)
  • detect_linear_reparam() for detecting linear reparameterizations between models
  • check_exact_equivalence() for testing exact parameter equivalence

degen 0.13.0

• Added plot_likelihood_surface() for 2D contour visualizations
• Extended parallel processing support:
  • compare_surfaces() now accepts cl parameter for parallel grid evaluation
  • fisher_information() accepts cl for parallel expected info computation
• Added export utilities:
  • as.data.frame() methods for all result objects
  • export_results() for CSV/JSON export
  • to_latex() for LaTeX table formatting
• Added batch comparison utilities:
  • compare_all_pairs() for systematic pairwise comparison
  • equivalence_matrix() for visual comparison matrix
  • find_equivalent_to() to find models equivalent to a reference
• Added model fitting integration:
  • fit_model() for MLE estimation
  • model_spec_from_fit() to extract model_spec from lm/glm objects
  • coef() and vcov() methods for fitted models
  • model_vcov() for variance-covariance from Fisher information
• Added example datasets:
  • equivalent_models: Pairs of mathematically equivalent models
  • nonidentifiable_example: Non-identifiable sum model
  • ecological_models: Three competing population growth models
  • mixture_model: Gaussian mixture with label-switching
• Added progress reporting:
  • Progress bars with ETA for compare_surfaces() and equivalence_classes()
  • progress parameter to control progress bar display
  • debug_comparison() for detailed trace of equivalence testing
• Added new vignettes:
  • vignette("model-specification"): Writing log-likelihood functions
  • vignette("comparing-models"): Model comparison workflows
  • vignette("identifiability-diagnostics"): Identifiability analysis
  • vignette("case-studies"): Real-world examples

degen 0.12.0

• Added model specification helpers for 10 common distributions:
  • model_spec_normal(), model_spec_exponential(), model_spec_gamma()
  • model_spec_poisson(), model_spec_binomial(), model_spec_beta()
  • model_spec_lognormal(), model_spec_weibull(), model_spec_negbinom()
  • model_spec_uniform()
• Added data simulation support:
  • add_simulator() to attach simulation function to model_spec
  • simulate.model_spec() S3 method for generating data
  • can_simulate() predicate
• Added analytical Fisher information via fisher_analytical():
  • Closed-form formulas for normal, exponential, poisson, gamma, binomial, beta, lognormal
  • Auto-detection of distribution from model name
  • Fallback to numerical computation
• Added model validation functions:
  • validate_loglik() for comprehensive likelihood validation
  • check_gradient() to compare numerical vs analytical gradients
  • diagnose_model() for full model health check

degen 0.11.0

• Added visualization functions using base R graphics:
  • plot.fisher_info() for eigenvalue spectrum and eigenvector heatmap
  • plot.surface_comparison() for likelihood discrepancy scatter plots
  • plot.identifiability_result() for parameter status visualization
  • plot_profile_likelihood() for profile likelihood curves with CI bands
  • plot.equiv_classes() for network diagram of equivalence relationships
• Added parallel processing support:
  • setup_cluster() and stop_cluster() for cluster management
  • equivalence_classes() and profile_all() accept cl argument
• Added profile likelihood confidence intervals:
  • profile_ci() to extract CIs from profile likelihood
  • confint.model_spec() S3 method for profile-based CIs
  • Support for multiple confidence levels (90%, 95%, 99%)

degen 0.10.0

• Added Rcpp backend for sampling functions:
  • Latin hypercube sampling via lhs_sample_cpp()
  • Random sampling via random_sample_cpp()
  • sample_par_space_cpp() for bounded parameter space sampling
• Added expected Fisher information via outer product of gradients (OPG)
• Added optimization-based surface comparison method
• Performance optimizations:
  • Cached source log-likelihoods in compare_surfaces()
  • Path compression in union-find algorithm
  • Hoisted loop invariants in identifiability classification

degen 0.9.0

• Added comprehensive integration tests
• Added shared test fixtures in helper-fixtures.R
• Verified complete workflows end-to-end
• Tested edge cases and consistency properties

degen 0.8.0

• Added introduction vignette with examples
• Organized pkgdown reference by topic
• Added vignette dependencies (knitr, rmarkdown)

degen 0.7.0

• Added equivalence_classes() for grouping multiple models by equivalence
• Added n_classes(), class_members(), are_equivalent() utilities
• Union-find algorithm for efficient class detection
• Added is_equiv_classes() type predicate

degen 0.6.0

• Added identifiability_check() for comprehensive identifiability diagnostics
• Added profile_likelihood() for profile likelihood analysis
• Automatic parameter classification (identified/weakly/non-identifiable)
• Recommendations for addressing identifiability issues
• Added is_identifiability_result() type predicate

degen 0.5.0

• Added fisher_information() for computing observed Fisher information
• Added info_eigenvalues(), info_condition(), info_rank() accessors
• Added null_directions() to find non-identifiable parameter combinations
• Added is_fisher_info() type predicate

degen 0.4.0

• Added compare_surfaces() for detecting observational equivalence
• Grid-based comparison with multi-start optimization
• Latin hypercube sampling for parameter space exploration
• Added is_surface_comparison() type predicate

degen 0.3.0

• Added equivalence_pair() for pairing models for comparison
• Added spec_a(), spec_b() accessors
• Added par_dims(), same_par_count() utilities
• Added is_equivalence_pair() type predicate

degen 0.2.0

• Added model_spec() for defining models via log-likelihood functions
• Added generics: par_names(), par_bounds(), n_par(), loglik()
• Added is_model_spec() type predicate
• S3 methods: print, summary, format for model_spec

degen 0.1.0

• Initial package structure
• Added internal validation utilities (check_numeric(), check_function(), check_bounds())
• Set up GitHub Actions for R CMD check, test coverage, and pkgdown