NEWS | R Documentation |
New function fitdistr.grouped, with pertaining class and methods.
More informative messages are issued when trace=TRUE, at several places.
An experimental way for initializing the parameters of sn.mple fitting is introduced; it only takes effect when it improves over the existing scheme.
Some adjustments in the documenation on request of the CRAN group.
Fixed a bug affecting st.prelimFit when the argument 'w' was used.
Calling rmsn, rmst, rmsc with parameters for distributions with dimension 1 now works.
The same feature applies to mst.prelimFit with one column matrix y.
In psn, improved handling of non-finite x values (NA, NaN, Inf's), which led to crashes; also, streamlined coding of qsn.
Minor fixes in the documentation of zeta and qsn; improved wording elsewhere.
Change of the tuning arguments when numDeriv::hessian is called, to improve computation of the information matrix following a ST model fitting.
Fixed bug which in some cases prevented the use of the 'start' argument.
Re-organization of some internal functions computing SUN summary quantities.
Improved documentation for selm, selm.fit and SUNdistr-base.
Support for the SUN family is introduced, as for probability distribution operations. Two modes of working are envisaged: - using classic-style functions for probability distributions, plus some functions of similar style; - using S4 objects of the new class SUNdist. For mode (i), the functions {d,p,r}sun, sun{Mean,Vcov,Mardia} are provided. For mode (ii), there are {make, marginal, conditional, affineTrans, join, convolution, summary}SUNdistr, and related S4 methods.
Additional facilities include convertSN2SUNdistr, convertCSN2SUNpar and two matrix operations (tr and blockDiag).
Fixed a bug in plot.SECdistr affecting the plot of a subset of the variables.
Fixed wrong computation of standard errors when a multivariate ST model was fitted with the constraint alpha=0; similar fix of function confint.
Change of the algorithm used in rsn: the additive representation is now used both fo r 0 and for non-0 values of tau.
Use of the vignette builder R.rsp.
Some minor changes in the selm function documentation.
New intialization technique for numerical MLE search when function selm is called with family="ST". This is performed by the new functions st.prelimFit and mst.prelimFit, with the aid of galton_moors2alpha_nu.
In st.prelimFit, the package quantreg is employed for preliminary linear predictor estimation.
Related new functions of more general interest are: fournum, pprodn2, pprodt2, qprodt2.
Additional facts: an improved version of function profile.selm.
Improved coding of some internal functions.
In sn.infoMv, new argument at.MLE is introduced.
Arguments of selm are updated to match changes in R.
Fix a bug in dsn when called with a non-scalar argument alpha.
Changed internal checks on 'try' output, to accomplish R changes.
On request from the CRAN group, a modification is inserted to get around numerical problems arising in connection with use of OpenBLAS (version 0.3.5).
For plotting of a multivariate SECdistr when 'range' is not supplied and 'data' is non-NULL, calculation of the plotting range has been modified.
Proper handling of the 'name' argument of marginalSECdistr when the result is a univariate distribution.
In pst, improved implementation of Method 2; slight modification of the automatic selection method when 'method=0'.
Internal function qst_bounds introduced for better initial bracketing of ST quantiles; improved qst coding.
In profile.selm the selected parameter area does no longer need to include the MLE/MPLE point.
In sn.infoMv removed bugs in the stage of parameter parsing, which prevented computing the expected information matrix.
In internal functions st.pdev.gh and mst.pdev.grad, improved computation of 'nu' component of logLik gradient, yielding faster fitting of ST models.
Improved handling of arguments of sn.infoMv.
Improved documentation of modeSECdistr and coding of modeSECdistrMv.
Improved checking of input arguments to lower level fitting procedures (those below selm) and improved handling for those of pst.
Fixed improper handling when not positive-definite information in st.infoUv occurs, pointed out by the CRAN group.
More extensive documentation: - addition of 'overview' entry in standard documentation; - in directory 'doc', inclusion of a PDF file providing a tutorial introduction to the package; - additions and improvements at various places in Rd files.
Fix a minor bug in sn.infoUv causing crash when the DP information matrix is not invertible.
Improved numerical inversion of st.cp2dp in extreme situations.
Tools for user-defined symmetry-modulated (AKA skew-symmetric) distributions are introduced: {d,r}[m]SymmModulated and its bivariate density plotting.
Fixed a bug in dsn affecting the cases (a) x=Inf, alpha=0, and (b) omega<=0.
Introduce methods confint and predict for selm-class objects.
Fix bug in rmst causing some dependence among subsequent samples.
Fix bug of modeSECdistrMv affecting certain ST cases.
Now plot.SECdistrBv allows to overlap plots.
Improved naming of output.
profile.selm can now be called with vector(s) param.values of length 1.
Method profile.selm is introduced.
The object returned by plot.SECdistrMv now includes the coordinates of the contour curves.
Fixed a bug affecting rmsn when called using dp=, also dp[[1]] is named beta instead of xi.
Not released
Output of plot.SECdistr is better structured and documented.
In pmst, handling of case nu=Inf required a fix.
Corrected a bug of internal function msn.dp2dp when called with aux=TRUE and d=1; this affected rmsn and rmst if d=1.
Fixed a bug in evaluation of the feasible CP parameter space of univariate ST.
Fixed a bug which crashed pmst when called with fractional degrees of freedom.
Functions dmsn, pmsn and dmst now expand a single value supplied as 'xi' into a vector or matrix of suitable dimension.
Fixed a bug in extractSECdistr from mselm-class objects.
Fixed a bug that prevented calling low level fitting functions with non-null 'penalty' argument.
Improved documentation of selm.fit and related functions.
Optimization parameters are now passed from selm to sn.mple and st.mple as indicated in the documentation.
Plotting of selm-class and mselm-class objects avoids clash of par('cex') parameters.
Computation of sn.infoMv now takes into account whether method="MPLE" was used at the estimation stage.
Created new functions extractSECdistr and modeSECdistr.
New methods mean and sd for class SECdistrUv, and new methods mean and vcov for class SECdistrMv.
Computation of qst switches to qsn if nu>1e4, instead of nu=Inf as before.
Fixed a bug in st.pdev.hessian (correction in args sequence).
Improved detection of singular distributions in selm output.
Improved handling of component names of SECdistr.
Fixed a bug affecting plotting of mselm objects under certain circumstances.
Fixed a bug affecting function selm when the weights argument contained 0's.
Improved coding in some functions.
More functions are exported and their documentation is added.
Function qsn has an additional argument 'solver'.
Functions pmsn and pmst can now be called with argument 'xi' of matrix type.
More functions are now exported in NAMESPACE.
Fixed a bug about selm.control argument of selm.fit.
Improved documentation of various functions.
Main few feature is the possibility to set the constraint alpha=0 in function selm and in lower level fitting functions.
Other additions or changes are: introduction of OP parameterization; fix a bug in qst; more efficient coding of dmsn and dmst; pmsn can now be called with argument 'x' of matrix type; in pst and qst, new argument method allows to select the algorithm employed.
More detailed documentation of pst and other functions and methods.
This is a major upgrade of the package, with much of the code completely new or largely re-written, leading to changes in the syntax and the user interface.
The key new functions are selm and makeSECdistr, with various related functions and methods.
S4 methods are adopted.
Many existing functions are updated, a few are dropped.
See help(SN) for more information.
(Development of "version 1" was started in June 2007.)
Various minor adjustments, many of them to fulfill CRAN programming standards.
Several changes and additions are included:
many routines allow use of composite parameter 'dp';
multivariate normal and t probabilities are now computed by 'mnormt';
use of NAMESPACE introduced;
some more routines introduced, eg. st.cumulants.inversion;
various fixes/improvements in documentation.
Added some new functions (these include msn.affine, sn.mmle, sn.Einfo, sn.mle.grouped), fix various errors, and other improvements (eg. improved pst).
The main change is the addition of routines for (multivariate) skew-t distribution; also some other routines, e.g. mle for grouped data
Fix error in sn.dev.gh, improved qsn
This is the first public release and distribution, originally from http://azzalini.stat.unipd.it in October 1998, shortly afterwards from CRAN.