singleRcapture version 0.2.1

GitHub releases should now be more frequent.

Changes from each particular update from last GitHub release:

singleRcapture 0.2.1

• Fixed bugs in IRLS fitting when providing weights argument when calling
  estimatePopsize
• The weightsAsCounts option in controlModel now works properly,
  dfbeta and dfpopsize decrease weight of selected row in a model matrix
  instead of deleting it when this is set to TRUE
• simulate method now works for both family object (like ztpoisson()) and
  for objects returned by estimatePopsize
• Introduced singleRStaticCountData sub class for singleRclass and made
  estimatePopsize a method so that a new package singleRcaptureExtra
  (under development) can make all necessary calculations for pop size estimation
  when providing object fitted by countreg::zerotrunc or VGAM::vglm/VGAM::vgam
• Some bugfixes for multicore bootstrap
• Code was re-factored to make further development/maintenance for the package
  much easier
• Update will be uploaded to CRAN
• semiparametric bootstrap now has a much faster sampling algorithm (that does the same job)

Unit tests:

• Reduced computational burden of unit tests
• Multicore tests will only be performed after TEST_SINGLERCAPTURE_MULTICORE_DEVELOPER
  is set to "true" via Sys.setenv and _R_CHECK_LIMIT_CORES_ to false

singleRcapture 0.2.0.1

• Added offset argument to estimatePopsize
• Added options for parallel computing in bootstrap and in dfbeta
• Added deviance for all negative binomial based models.
  (NOTE: They are very slow for now and I believe it may change after I verify
  one theoretical results that will lead to significant speed increase for
  these computations)
• Overhaul of starting points (new methods and added linear predictors start in IRLS)
• Code for weights in IRLS fitting was speed up
• Minor bugfixes

singleRcapture 0.2.0

The package is now at CRAN

• features and improvements:

  • Added final Hurdleztnegbin model
  • Vastly improved redoPopSize which now handles bootstrap on a fitted model
    non standard covariance matrixes newdata argument user supplied coef etc.
  • Added predict.singleR method which offers standard error for both link,
    response as well as mean predictions
  • No unexpected warnings should occur now in main function when using
    the package correctly
  • All control arguments are now verified before being passed
  • Fitting is now more reliable
  • Added information about stats::optim error codes
  • Added warnings for functions computing deviance

• bugfixes:

  • fixed bugs occurring when using mathematical functions as part of formulas
    i.e. when setting formula to something like: y ~ log(x) + I(x ^ t) + I(t ^ 2)

singleRcapture 0.1.4

• features

  • Added ztoinegbin, oiztnegbin and ztHurdlenegbin models
  • Added an optional arguments to all family-functions to specify a link
    function for distribution parameters
  • Updated and standardized documentation
  • Added more warnings
  • Added some more methods for singleR class in some commonly used glm
    functions, in particular texreg::screenreg should work well now

• changes

  • Changed some default arguments
  • Added option to save logs from IRLS fitting

• bugfixes

  • Fixed some issues with intercept only models
  • Fixed some slight miscalculations in information matrixes for one inflated
    models making fitting them much more reliable

• github repository

  • More and better Rcmd tests

singleRcapture 0.1.3.2 -- NTTS

• features:

  • Added function that implements population size estimates for stratas
  • More warnings in fitting
  • More options in control functions
  • Corrected/implemented deviance residuals for more models

• changes:

  • Now the whole package uses cammelCase
  • Performance upgrades
  • Corrected some miss calculated moments
  • Change exported data so that factors are actually factors not just characters
  • Removed unused dependency

• github repository

  • Added automated R-cmd check

singleRcapture 0.1.3.1

• features:
  • Basically all of documentation was redone and now features most of important
    theory on SSCR methods and some information on (v)glms
  • Added checks on positivity of working weights matrixes to stabilise "IRLS" algorithm
  • Added most of sandwich capabilities to the package, in particular:
    • S3 method for vcovHC was implemented
    • vcovCL should work on singleR class objects
      should work with "HC0" and "HC1" type argument values
  • Basic version of function redoPopEstimation for updating the
    population size estimation after post-hoc procedures was implemented
  • popSizeEst function for extracting population size estimation
    results was implemented
  • Minor improvements to memory usage were made and
    computation was speed up a little
  • Changed names of mle and robust fitting methods to optim and IRLS
    respectively
  • Some bugfixes
  • More warnings messages in estimate_popsize.fit

singleRcapture 0.1.3

• features:
  • Multiple new models
  • IRLS generalised for distributions with multiple parameters
  • bugfixes
  • QOL improvements
  • extended bootstrap and most other methods for new models

singleRcapture 0.1.2

• features:
  • control parameters for model
  • control parameters for regression in bootstrap sampling
  • leave one out diagnostics for popsize and regression parameters (dfbetas were corrected)
  • fixes for Goodness of fit tests in zero one truncated models
  • computational improvements in IRLS
  • other small bugfixes

singleRcapture 0.1.1

• bug fixes and some of the promised features for 0.2.0 in particular
  • More tiny tests
  • Some fixes for marginal frequencies
  • Deviance implemented
  • dfbetas and levarage matrix
  • Parametric bootstraps work correctly for the most part there is just some polishing left to do

Full Changelog: v0.1.0...0.2.1

singleRcapture 0.1.0

The first version of the singleRcapture package that is somewhat suitable for use.

Features:

• Zero truncated count data regression with Poisson, Geometric and Negative Binomial models and their application to population size estimate, analytic, nonparametric and parametric bootstrap variance estimations are available. Two methods of constructing analytic confidence intervals are supported studentised and by logarithmic transformation.

• Zero truncated count data regression with Poisson, Geometric and Negative Binomial models and their application to population size estimate, analytic and non parametric bootstrap variance estimations are available.

• Two mixture models, chao and zelterman, utilising logit regression for population size estimation, analytic and nonparametric variance estimations are available.

• The above models are applied with their respective family class functions (model parameter) that are passed to main function estimate_popsize.

• Two fitting algorithms, IRLS and simple MLE by optim.

• summary method for the model. It is handled differently in singleRcapture than in glm's.

• marginalFreq function for creating a table of marginal frequencies.

• summary method for marginal frequencies that applies chisq and G tests for goodness of fit

• vcov function for extracting the covariance matrix of regression parameters.

• basic control.pop.var parameters for variance estimation.