Current versions

npde 3.2

News

Changelog

Version 3.0, January 2021 to 3.2, November 2021

1. Change in default behaviour
   • diagnostic graphs show npd instead of npde by default (see Nguyen et al., CPT:PSP 2017)
2. New features (see details in the PDF documentation and the online help files):
   • all graphs have been recreated using the ggplot2 library
   • Important note: to avoid adding a dependency on Cairo the plots are saved using ggsave() from ggplot2; this means the postscript output lacks all the prediction intervals as transparency is not supported (warning messages) => to output to eps, use cairo_ps with the Cairo library to save the required plots
   • graphs of npde with a reference profile included to show the evolution of the process being modelled
3. Main bugfixes
   • autonpde has been modified to remove the deparse step (may improve the usage of npde inside a loop)

A full list of changes can be found in the CHANGES file.

Version 2.0, October 1st 2012

1. Manuals:
   • the user guide has been updated to include the new methods and datasets
   • an additional guide has been created to showcase the different graphs in the npde package; it contains many examples of graphs and shows how to fine-tune graphical options
   • both guides can be found in the inst directory of the package
2. New features (see details in the PDF documentation and the online help files):
   • new methods to handle the data below the LOQ
   • prediction bands: added prediction bands to scatterplots, distribution plots
   • covariate models: tests for covariate models; option covsplit added to the scatterplots and distribution plots
   • different methods for decorrelation (default method: Cholesky)
3. Reprogrammed using S4 classes; main changes include:
   • the structure of the output has changed: an object of class npdeObject is now created by a call to npde or autonpde
   • methods (special functions) plot, summary are now available and apply to the npdeObject object
   • all methods can be called simply as e.g. plot(x) where x is an object of class npdeObject (see documentation)
   • options for graphs and methods are stored in lists within the object and can be modified on the fly
   • new methods have been defined
   • the function testnpde() has been changed to the method npde()
4. Additional changes:
   • the default option is now to compute the pd (calc.pd=TRUE)
   • the option "output=TRUE" has been removed; the npde and autonpde return a value which can be assigned to an object, but which remains invisible (not printed) when not assigned
   • a "method" option has been added to the call to autonpde; method="cholesky" uses the Cholesky decomposition to compute a square root of the individual variance-covariance matrix Vi; method="inverse" uses the inverse of Vi obtained through diagonalisation); method="polar" uses a combination of Cholesky decomposition and diagonalisation to obtain the same inverse (more stable)
   • a "continuous" option has been added; if TRUE, the distribution of pd is made continuous (default=XXX)
   • a "centering" option has been added; if TRUE, the distribution of pd is centered by substracting 1/(2K) so that pmin=1/(2K) and pmax=1-1/(2K); if FALSE, pmin=1/K and pmax=1, which was the default in the previous version of npde (default=XXX)

A full list of changes can be found in the CHANGES file.

Version 1.2, November 21st 2007

1. On loading the version and date of the library is printed.
2. A few changes have been made to check user input when using the interactive menu.
3. The standard error of the mean and variance are now reported in the output.

Version 1.1, July 25th 2007

1. The computational speed has been dramatically improved by changes to the computational routines.
2. The user may now specify an additional column to signal missing data in the observed Y (eg MDV column for Nonmem and Monolix users) containing values of 1 to denote missing data and 0 to denote observed data. Calls to the autonpde() function can include this optional column by the new imdv=xxx element.
3. A bug with the computation of npde when calc.pd was FALSE and the dataset included missing values has been corrected.