NEWS

vcrpart 1.0-6 (2024-10-04)

Replace in C-code 'Calloc' memory allocations by 'R_Calloc' and 'Free' by 'R_Free' by as compiling with _R_USE_STRICT_R_HEADERS_=true will fail in the future.

Newest grid.rect() version seems seems to use white as default background (and not transparent as previously). Was a problem in panel_coef().
Solve new issue with character arguments at Fortran calls.
estfun() and gefp() are not anymore exported as methods for olmm(). estfun.olmm() and estfun.gefp() were renamed to olmm_estfun() and olmm_gefp().
Removed vignette because of warning messages on the size of the pdf.
Documentation of output values of various methods for 'fvcm', 'olmm' and 'tvcm'.

In R 4.1, length(fix) > 1 in 'olmm_control' caused an error. Fixed by Gilbert Ritschard.
Vignette as Rnw file.

Replace in C-code 'alloca' memory allocations by 'R_alloc' to avoid memory allocation problems, including using 'vmaxget' and 'vmaxset'.
Improved two bugs concerning memory overflow.

Internal modification for the 'singular.ok' argument of future R versions.
Replaced 'memcpy' calls in C-code with 'for' loops because they were used for small objects.
Improve R-Code for 'alloca' calls (adding 'R_CheckStack()' after calls).

Added vignette.
Update references.
Improve documentation for formulas.
The 'data' slot of 'tvcm' object newly includes several variables used for fitting. These data are cleaned with respect to missing values, but no transformations are applied.
In 'fvcolmm_control' the 'alpha' argument has mistakingly been ignored. Bug fix.
Added new warnings for 'tvcglm_control'.
Bug fix: for 'tvcglm_control' and 'tvcolmm_control' the '...' arguments have previously been ignored.
added default for 'minsize' to 'fvcm_control' and 'fvcglm_control'.
Add new argument 'fitargs' to 'tvcm'. Allows to pass arguments for the fitting function in 'tvcm' and 'fvcm'. This was possible before via the '...', but leads to argument name clashes in some cases. See the help of 'tvcm'.
Improve the help of 'vcpart-formula'.
fixed an unnecessary error case in 'tvcm_folds'.
add check for the response variable in 'olmm'.
bug fix for 'predict.fvcm'.
bug fix in for internal C function 'olmm_setPar'.

Update of citation info
modified 'olmm_control' for the 'optim' optimizier. Now 'BFGS' is set as the default and a warning is shown if a non-gradient based method is chosen while 'numGrad = FALSE'.
Enable option 'sparse' for function 'contr.wsum'.
Improved C-code for olmm. Does not have impacts for users. Please report if any changes appear.

Changed titles in 'panel_coef', 'tvcm.print' and 'tvcm.summary'
Add '...' to the 'VarCorr' method for 'olmm' objects to resolve the warning since 'nlme' version 3.1-123. Removed 'rdig' argument.
Because of an internal bug the print method for 'fvcm' objects printed always 'Bagging based varying coefficient model', even if 'mtry < Inf'. Bug fixed. Thanks Michel Ballings for the hint.
In cases of 'mtry = Inf' the algorithm returned an error. Now 'mtry = Inf' is allowed.
In cases of 'mtry < Inf' the 'print.fvcm' shows now the set 'mtry' parameter.
Improved the 'coef' method for models fitted with 'tvcm', 'tvcglm' or 'tvcolmm'. 'by' variables of 'vc' terms without splits are not anymore assigned to fixed i.e. main effects.
Corrected internal bug of 'predict.fvcm'. Except for 'type = "coef"', and models with 'family = gaussian()' or 'family = cumulative()', predictions gave wrong results. Sorry.
The default setting for 'na.action' on 'tvcolmm', 'tvcglm', 'tvcm' and 'olmm' is set to 'na.action = na.omit'. Thanks to the anonymous referee of JSS.
Added the argument 'fast' to 'tvcglm_control'

Corrected issues with moderators with name 'x' in 'vc' terms. Thanks Jonathan Williams for the hint.
Corrected issues in cases no 'vc' terms are specified in 'tvcm' or 'fvcm'. In these cases there will be a warning message and a linear model as specified in 'fit' will be returned. Correction carries over to 'tvcglm', 'tvcolmm', 'fvcolmm' and 'fvcglm'.
Replaced 'ptry', 'ntry' and 'vtry' parameters for random selections of the 'fvcm' function by 'mtry', see the 'fvcm_control' function. 'mtry' specifies the number of random selections of combinations of partition, node and variable.
Removed internal bugs of the 'tvcolmm' and 'tvglm' functions. The changes have no consequences for the users.
Removed internal bugs in 'fvcm'.
Improved help file of 'fvcm'.
Changed the defaults for 'maxwidth' and 'maxdepth' in 'tvcm_control' from '1e9' to 'Inf'.
Changed default parameters for the number of folds and verbose in 'fvcm_control', 'fvcolmm_control' and 'fvcglm_control'.

Changed a default definition for 'formula' of 'olmm'. Terms which are not specified by 'ce' or 'ge' are now always assigned to 'ge' terms. The old revision used a model specific assignement. It is generally best to use the 'ce' and 'ge' terms to get the desired model.
Corrected labels for random effects of 'olmm' objects.
Revision of the 'predict' method for 'olmm'. The new function implements the three prediction option of Skrondal and Rabe-Hesketh (2009).
Revised references.

Improvements for 'prunepath' method.
Moved documentation of 'prunepath' to 'tvcm-assessment'.
Added example for 'prune'.
Internal changes in 'estfun.olmm' and new argument 'center' for the function 'predecor_control'.
New argument 'fast' for 'tvcm_control' to control whether an approximative or the accurate model is used at exhaustive search.
Corrected bugs in 'tvcm_exsearch_nomToOrd' and 'tvcm_getNumSplits' which appeared in in special cases.
Improved help manual for 'tvcglm' and 'tvcolmm'.
Revised references.

Added seed option to 'tvcm_control'.
The new implementation clearly distinguishes between the two functions 'tvcolmm' and 'tvcolmm' with separate help files. The general function 'tvcm' is still available.
Added convenience function 'tvcolmm_control' and 'tvglm_control'.
Improvement for 'panel_coef': Points and lines surpassing the boxes are now suppressed.
Added variable centering as default for split selection.
Redefinition of tuning parameters for 'tvcm'. The main tuning parameter is now 'cp'. See the help of 'tvcm' and 'tvcm_control' for details.
Added 'nimpute' argument for 'tvcm_control'.
Added detail section to the help page of 'tvcm_control'.
Removed AIC table from 'print.tvcm' (AIC and BIC seem not relevant measures for models fitted by 'tvcm').
Added 'PL' data set.
Removed bug for numeric estimation of covariance of 'olmm' objects.
Added 'depth' and 'width' methods.

First CRAN release.
'tvcm' and 'fvcm' allow for multiple 'vc' terms, i.e. coefficient-specific partitions.
Complete revision of syntaxes, argument names and default parameters. R commands for the former version 0.1-14 are hardly compatibel. Please contact <[email protected]> in case of questions.
The default splitting criterion of 'tvcm' was changed from (old) coefficient constancy tests combined with greedy loss minimization to (new) exhaustive greedy loss minimization. Splitting based on the (old) requires to set the argument 'sctest = TRUE' in 'tvcm_control'.
The greedy loss reduction minimization stage of 'tvcm' does now fixate coefficients which are not splitted at the estimates of the current model. The technique intends acceleration.
Cross-validation is now integrated in the 'tvcm' function.
'cvloss' and 'prune' are parallelized using the 'parallel' package.