1.1.1

In both CRAN and github versions:

• Fixed bug that caused names of IMLEGIT objects to be NULL (This was already fixed in 1.1.0 of github but not 1.1.0 of CRAN)
• Fixed bug that caused names of summary.IMLEGIT elements to be NULL
• coef_main was incorrect in example_2way, example_3way and example_3way_3latent, this was fixed
• Added best subsets of variables in output of bootstrap_select (and option to choose how much to show)
• Added a little more info on exclusion options in stepwise search as it was a bit unclear to their purposes and how to set them
• Fixed small errors in documentation
• Fixed bug in stepwise_search_IM where using cross-validation would result in an error
• Set max_iter=100 everywhere as default
• It makes no sense to do cross-validation within a bootstrap scheme, cross-validation options where thus removed from bootstrap_var_select
• Added AICc to LEGIT and IMLEGIT output
• Added AICc to stepwise_search, stepwisesearch_IM and boostrap_var_select
• removed cross-validation criterion from appearing in interactive stepwise search when not using cross-validation

In github version:

• added functions to account for gene-environment correlation (rGE) with LEGIT and IMLEGIT models [Experimental feature]

stepwise search for IMLEGIT + bootstrap var select + various fixes

• Added option search="both" to search for both G and E at the same time in stepwise search (This is the new default now)
• Added stepwise search for IMLEGIT
• Fixed a bug in backward search, p-values were always taken to be from the genes
• Added remove_miss option in stepwise search as by default it would not remove missing data except during each step where we add or remove a variable. This could be annoying for certains users who would like the program to automatically remove missing observations.
• A new function called boobstrap_var_select was added. With small datasets variable selection can be somewhat unstable so this create boostrapped versions of the dataset and do stepwise search on all of them, then it reports the percentage of times that each variable was selection. This bound to be very slow since this is bootstrapping, only use if you have a small sample size.
• Added warning when maxiter is reached
• Changed default from search="genes" to search="both" in stepwise search
• Increased max_iter and max_steps defaults to 100 just in cases
• Added snow, doSNOW, utils and foreach in import/depends. These are needed for bootstrap function
• updated vignette

Note : This is sightly different than CRAN version as I just noticed there was a small bug where the names of the IMLEGIT were NULL and this is fixed in this GitHub release.

1.0.4: Add files via upload

• Major feature : Added IMLEGIT function, an extension of the original LEGIT model in which you can set as many latent variable as desired instead of just G and E. Also added associated functions (only step-wise search is unavailable with IMLEGIT for now).
• Fixed a bug in the calculations of the residuals, it was only taking the last observation of the dataset when folds != NULL therefore giving the impression that there were no possible outliers.
• Added id option in LEGIT_cv so that the output potential outliers can show this ID rather than just the observation number
• Not looking at levels anymore for id in folds but only at the unique values, no need for factors anymore.
• Fixed multiple bugs in LEGIT_cv with folds option that arised in last update
• Fixed a major bug in LEGIT_cv and IMLEGIT_cv (Please redo your cross-validations just to be sure)
• Fixed minor things

v1.0.3.1: Add files via upload

• Missing observations in the unused variables of "data" won't affect the results anymore as unused variables are now dropped at the beginning.
• Added error message when datasets contain factors, everything must be dummy coded because otherwise matrix multiplications won't work.
• Updated documentation based on changes

v1.0.3: Add files via upload

1.0.3

• Fixed documentation to mention bidirectional-forward and bidirectional-backward search as they were already implemented!
• Fixed bug that appeared when doing backward search and cross-validation
• Added Huber and L1-norm cross-validation errors to the LEGIT_cv output and choice of search_criterion in the stepwise search. These are robust criterion that handle outliers better than the usual L2-norm (which the R^2 is based on).
• Added argument to change the parameter of the Huber cross-validation error
• Added reference for Huber and L1-norm cross-validation errors
• Added newline between the AUC plot and the last line in vignette

v1.0.1

• Added optional argument in LEGIT_cv to let the user provide their own folds (could be important for certain data types like time series or longitudinal data)
• Added longitudinal cv folds function that can be used with LEGIT_cv
• Updated description and references with the ArXiv link to the alternating optimization algorithm.

v1.0.0

Update DESCRIPTION