Changelog

Source: NEWS.md

FamModel 1.0.0

R code changes and improvements

Contrast class

  • Changed to store estimate of Lθ − m rather than Lθ.
  • Changed chi-square df calculation to use rank of L rather than number of rows.
  • Improved output from print method.

FamData class

  • Added checks to constructor to make sure that individual order and identifiers are identical in data and phi members.
  • Added pedigree check to constructor.
  • Converted all integer storage type numeric columns in input data set to double storage type to avoid later comparison errors with identical().
  • Added identification of families with consanguinity and get_consang method.
  • Modified get_data method to return copy of data member to avoid inadvertent modification by reference.
  • Added print method.
  • Modified make_model_mats_lmm to track included observations from data member by fmid and id and include these identifiers as names or row names to increase robustness.
  • Modified lmm method to use only non-probands in lm() fit used to obtain initial parameter estimates for optimization.

FamLMMFit subclass

  • Disallowed construction outside of internal FamModel methods and functions.
  • Made sure that analytical, not numeric, Hessian was being used in covariance matrix calculation.
  • Added data mapping checks to constructor to protect against bugs due to future API changes and ensure accurate diagnostic output.
  • Added get_model_res method with additional diagnostic features.
  • Added scaled gradient convergence criterion and negative Hessian diagnostics to print and get_h2_a_lrts methods.
  • Modified get_h2_a_lrts to cache heritability LRT results after first call and use cached values on subsequent calls unless the user explicitly overrides.

Other

  • Moved common LMM optimization code to utility function and switched optimization to use optim() without relative function convergence criterion.
  • Switched from variance to SD paramaterization for linear mixed model to improve scaling properties of optimization problem.
  • Added rigorous argument checks to make_coef_mat() and print_ests().
  • Added title argument to print_ests().
  • Replaced all.equal() with identical() for all object comparisons where floating point imprecision should not be an issue.

C++ template infrastructure changes

  • Fixed bug in sing_asc_lmm model that would yield misspecified trait variance for inbred individuals.
  • Moved diagnostic calculations, which are not necessary for optimization, to FamLMMFit$get_model_res() method to speed up iterations.
  • Transitioned to shared library with same name as package to avoid issues with newer pkgload versions during development.
  • Refactored to allow easy incorporation of multiple models into a single shared library as per https://github.com/mlysy/TMBtools.

Documentation changes

  • Added vignette("linear_mixed_models").
    • Moved methodological details from individual class and method documentation to vignette.
    • Discussed conditions required for valid inferences when true values fall on boundaries of the parameter space.
    • Discussed properties of estimated residuals and diagnostics relevant to assessing goodness of fit.

FamModel 0.2.0

Major Changes

  • Renamed package to FamModel.
  • Added new classes for working with model results:
    • FamModelFit abstract class.
    • FamLMMFit subclass specialized for linear mixed model results.
    • Contrast class for results of general contrasts for a FamModelFit object.
  • Modified linear mixed modeling framework to allow:
    • Variance component heterogeneity across families from different populations.
    • Return of various types of residuals.
  • Added utility functions for printing coefficient matrices.
  • Made various other improvements for robustness.
  • Updated and expanded documentation with additional methodological details and references.

FamMix 0.1.3

Minor Changes

  • Modified FamData$lmm() method by adding code to transform likelihood maximization problem by scaling parameters by square root of their Hessian diagonal elements at the initial estimates, which should improve conditioning of the Hessian and therefore numerical convergence.

FamMix 0.1.2

Minor Changes

  • Changed FamData constructors to use character string column names and updated documentation.

FamMix 0.1.1

Bug Fixes

  • Fixed bug in FamData$new() using phi signature.

FamMix 0.1.0

  • Initial release.