A large number of improvements have been made in version 3.4 of S-PLUS, as a result of MathSoft's own testing, and user feedback. The following section summarizes the changes that have been made.
Data Frames rbind.data.frameThis is now a more robust function. data.frameIf this routine is passed a character matrix, it no longer converts each column to a factor.
Distributions dhyper Now accepts NAs without returning an error.
Minor improvements have been made to the following functions: sample, qbinom, qwilcox
Graphics wireframeNo longer fails if there are any NAs, but instead leaves a hole in the surface. levelplotNow handles NAs correctly. motif No longer fails if all the required colors are not available, it now gives you a reduced color set.
Minor improvements were made to the following function: panel.superpose
Input/Output sink Nested calls can now be made to this function.
Minor improvements were made to the following function: scan
Math Functions polyrootNow works correctly when some roots are 0, or when the highest order coefficient is 0. kroneckerIf both matrices input to kronecker have only one row, the result is coerced into matrix format.
Minor improvements have been made to the following functions: is.nan, is.inf
Programming Minor improvements have been made to the following functions: parse.test, options
Statistics step.glmIf the scope argument includes transformations of data that produce NA values, an informative error message is now given. location.mNo longer fails if there are NAs in the vector. anova.survreg anova.survreglistThese two new functions have been added for comparing survreg models. This fixes a bug in earlier versions resulting from using the corresponding functions for a glm model, namely anova.glm and anova.glmlist. summary.survreg print.survregThere have been small modifications to the output of these functions, so their output may differ in detail from examples of their use printed in the documentation.
Minor improvements have been made to the following functions: kmeans, rreg, summary.lm, aov, deriv, factanal
Syntax Minor improvements have been made to the following function: all.equal
Time Series months Adjusts correctly for the origin, similar to years and days. origin.defaultThis is now 01/01/1960, to be consistent with Survival functions.
Minor improvements have been made to the following functions: arima.diag, cut.dates, ts.intersect
Tree Functions cutree If the height argument to cutree is greater than the maximum height of the tree, a correct error message is now given. plot.tree.sequenceThis no longer uses approx, so the axis on side three is changed. prune.tree predict.treeThe handling of missing values in these functions has been changed, as the default use of na.tree.replace was almost always wrong. This behavior can still be obtained by calling na.tree.replace.all. There are modified help pages for na.tree.replace, plot.tree.sequence, predict.tree and prune.tree. Also the tree sequence is computed correctly starting at -Inf, not at 0 (this is important for method="misclass").
Warning Use prune.misclass instead of prune.tree with the function cv.tree, as the method argument is not correctly handled by prune.tree.
Miscellaneous demo library(demo) is now detached when demo has finished.
S-PLUS version 3.4 is available on the following platforms, compiled with the named compilers and specified Motif releases (all platforms use X11R5):
Table 1: Systems used in the development of S-PLUS 3.4
Platform OpSys C Fortran Motif
SunOS4 SunOS4.1.3 SC3.0.1 SC3.0.1 ICS Solaris1 Motif1.2.3
SunOS5 SunOS5.3 SC3.0.1 SC3.0.1 Motif1.2.2 Solaris2
SGI IRIX 5.3 3.19 4.0.2 Motif1.2.3
HP 9000 HP-UX 9.0.5 9.70 9.16 Motif1.2.2 series
RS/6000 AIX 3.2.5 1.3 2.3 Motif1.2
DECAlpha DECUnix 3.11 3.7-063 Motif1.2.3 v3.2(OSF/1)
The primary documentation for this product was written for S-PLUS version 3.2. If you upgraded from version 3.3 to version 3.4, you can ignore the following section. If you upgraded from an earlier version of S-PLUS, or are a new user, the following section details the enhancements to the product that were made with the release of version 3.3.
The principal new features of version 3.3 are the following:
The Survival4 package significantly extends the survival analysis functionality previously available in S-PLUS. New functionality includes parametric survival models, formula-based model specification, a test for proportional hazards in a Cox model, and predicted survival for an age and sex matched cohort of subjects given a matrix of known hazard rates for the population.
All of the version 3.2 functionality is maintained in versions 3.3 and 3.4, but there are some function name changes, and in somes cases functionality has been combined into fewer S-PLUS functions. For example, the functionality of coxreg and agreg in version 3.2 are now contained in the single function coxph. The following table details the differences between the two versions. The version 3.2 survival functions are included in version 3.3 and 3.4, but they should all now be considered deprecated.
Table 2: Survival Analysis functions in version 3.4 and 3.3, compared to 3.2
Function Description 3.4/3.3 3.2
Parametric regression models survregNone Cox proportional hazards models coxphcoxreg Andersen-Gill extension to Cox models coxphagreg Formula representation for failure/status SurvNone Fit a survival curve survfitsurv.fit Compare survival curves survdiffsurv.diff Compute a test for proportional hazards cox.zphNone Plot a survival curve plot.survfitplot.surv.fit Predicted survival for age-sex matched cohortsurvexpNone Handling of missing values naresid, naprintNone
Trellis graphics are plots which contain multiple panels, arranged in a regular grid-like structure (a trellis). Each panel graphs a subset of the data. All panels in a Trellis display contain the same type of graph but these graphs are general enough to encompass a wide variety of 2-D and 3-D displays: histogram, scatter plot, dot plot, contour plot, wireframe, 3-D point cloud and more. The data subsets are chosen in a regular manner, conditioning on continuous or discrete variables in the data, thus providing a coordinated series of views of high-dimensional data. The Trellis software gives flexible control over axes and aspect ratios, and contains "banking" computations that let the data select the aspect ratio. See the Trellis Graphics User's Manual for details.
The Matrix library contains LAPACK-based functions for numerical linear algebra, including methods for matrix decompositions, norms, condition estimation, and determinants, as well as solve methods for a variety of matrix classes and decompositions. These functions are object-oriented, defined as methods for a new class "Matrix" and a number of specialized subclasses including "Hermitian", "Orthonormal", and "Diagonal". The following matrix decompositions are available:
Methods for solving systems of equations and finding matrix inverses and pseudo-inverses are also available, based on these decompositions.
Methods are available for calculating matrix p-norms, condition numbers, and determinants.
Several improvements have been made to the data frame creation functions. These improvements meet the following goals:
Also added are the function merge for joining two data frames and the function by which repeats an analysis for all levels of a specified factor variable. Both of these are similar to SAS statements with similar names, and have been often requested by S-PLUS users.
In addition, the aggregate function, previously limited to time series data, has been made generic, and a new method has been written for data frames.
The read.table function, used to read data frames from text files, has been improved to use much less memory and other system resources.
The new functions ks.gof and chisq.gof provide, respectively, Kolmogorov-Smirnov and Chi-squared goodness-of-fit tests. These tests are useful for assessing whether data have a certain statistical distribution.
The function nnls.fit solves nonnegative least squares problems. The function ivp.ab solves systems of differential equations with given initial values and specified derivatives.
The plot methods for objects of class "lm", "glm", and "gam" have been improved to give a wider, more useful set of default graphical views, with a consistent user interface. All three functions now have an ask argument that defaults to TRUE, controlling whether the function presents a menu of choices to the user. The choices allow the user to select any single view, or all the available plots.
The plot.glm function has the following available plots:
The plot.gam function, as before, plots the estimated relationship between the individual fitted terms and each of the corresponding predictors, with options to add standard errors, residuals, or rugs to the plots.
The Elan License Manager software used to manage S-PLUS licenses has been upgraded from 2.2.x to 3.0.4. This upgrade had a few side effects, including the requirement that all version 3.3 installations will require a new license key. The new license key, however, no longer depends on the host's IP address, so that changes to the IP address will not require the user to secure a new license key. Most of these effects should be transparent to most users, although users and system administrators who used any of the more advanced features of the S-PLUS version 3.2 license manager may notice some differences. A few of the elm* commands were renamed (although the wrapper scripts used by the unixcom Splus LICENSE utility have been preserved, so these can be used to call them by their old names):
elmreport -> elmrpt elmnotify -> elmalert
New wrapper scripts have been written to use the new names: Splus LICENSE rpt Splus LICENSE alert
Other changes include the following:
There are three principal incompatibilities between version 3.3 of S-PLUS and version 3.2:
Over time, some S-PLUS functions become obsolete, either because new functions are developed that replace and extend their functionality, or because of changes in hardware or other software that make them less useful to users. To minimize the impact on users who may be using those functions, we try to move these functions out of S-PLUS in phases. First, we declare a function deprecated, meaning there is a better way to perform the task formerly done by the function. Deprecated functions, however, are still available. Generally after two or three releases as a deprecated function, the function is removed from S-PLUS, and declared defunct.
The following older survival analysis functions are now deprecated:
The openlook graphics device is now deprecated, along with the "openlook" GUI option for data.ed and help.start. Use the motif graphics device and the "motif" GUI option.
Most S-PLUS tools that require either the SunView environment or the Athena widget set have been made defunct and removed from S-PLUS version 3.3. The lone exception is the X11 graphics device, which remains available on all systems on which StatSci was able to build it. The following tools have been made defunct: