2021-05-31 version 0.51
bugfix:
- fixing memory issues in C++ code
- reverting Schall algorithm to R code
2019-06-09 version 0.50
new features:
- speed is increased due to translation of key parts into Rcpp
- arguments of rb(): "penalizedpart_pspline" fitting only the penalized part of a pspline; modification of number of basis function for B-splines possible
- new function AIC
- update expectreg.ls results
- expectreg.ls has now argument smooth="ocv" and LAWSmaxCores for interactivly defining the maximal number of used cores in unix systems.
- prediction of values without use of intercept is now possible
- plot function contains more arguments to individualize plots and also plots results of factor and parametric covariates
- expectreg.boost includes now arguments BoostmaxCores for number of cores and quietly to run quietly
changes:
- names of the covariate specific output now contains the type of the covariate.
- psplines, 2dsplines, markov, random are now centered around 0
- radial is out of order
- grid of cvgrid and lcurve is modified
- aic and bic have changed formulas
- mono has changed penalty for flatend
- class of expectreg now includes the smoothing type
bugfixes:
- print of y~1 model without error
- psplines, 2dsplines, markov and random are centered around 0
- plot(): ask is now working
factor covariates are generally checked
plot of markov without bnd data does not throw an error anymore
markov fields are now grey to avoid confusion with colorshemes of the other plots
psplines are centered around their intercept - intercept of asymmetry 50% (if possible)
par() is saved and reset before each plot
- psplines and 2dpsplines have now a bases degree of 3
- penalty of krig is modified
- schall algorithm uses only the penalized part of the covariates
- quantboost has modified maximum number of cores
2015-05-25 version 0.40
new features:
- argument for penalty on random intercept in longitudinal qp sheets
- added new example data set Gasoline as copied from package plm
bugfixes:
- fixed package Description for R>=2.8.0
- fixed quantile bundle estimation for specific examples from crashing on Linux
- fixed definition of AIC / BIC for expectiles
2014-03-05 version 0.39
new features:
- longitudinal data in quadratic programming sheets with random intercepts
changes:
- efficient algorithm to determine expectiles of distributions, e.g. enorm
- set maximum number of used kernels to 2 according to CRAN policy
bugfixes:
- fixed handling of boundary class regions
- fixed plotting of maps
- fixed boosting of Markov random fields (prediction)
2013-07-20 version 0.38
new features:
- least squares smoothing parameter choice by GCV, AIC, BIC, L-curve, GCV grid search
changes:
- optimized Schall algorithm for extreme expectiles
bugfixes:
- fixed prediction and plotting of 2dim bases
2013-04-10 version 0.37
bugfixes:
- fixed print method for boosting
- fixed covariate name usage for parametric effects
- fixed intercept output of quantile bundle
changes:
- removed function "interp" from plot in order to remove dependency from akima
2013-01-15 version 0.36
changes:
- plots for confidence intervals are now added
bugfixes:
- fixed prediction of parametric effects
2012-08-15 version 0.35
new features:
- introducing asymptotic results for weighted least squares estimation
- methods summary and confint added
changes:
- expectreg.qp now works with semiparametric additive models
- nonlinear effects can be plotted with a rug plot
- plot function now adapts to setting of mfrow and pdf creation
- parametric bases now accept more flexible categorical variables by implementing model.matrix
bugfixes:
- additive bases now work with varying coefficients
- p-spline bases and schall algorithm now work better with NA in data
2012-02-13 version 0.30
changes:
- optimised basis selection in kriging
- changed parallelisation to 'parallel' package
bugfixes:
- choosing asymmetries in quantile bundle fixed
- parametric effects in boosting fixed
- fixed plotting options
2011-11-30 version 0.29
new features:
- function mono added that constructs shape constraint p-spline bases
(not yet available for expectreg.qp)
changes:
- renamed function base to 'rb' to avoid name conflict with base package
- includes boosting baselearners bmono and brad
- package structure changed:
- expectreg.ls now contains laws, restricted, bundle and sheets method
- renamed expectreg.boot (prev. expectile.boost) and expectreg.qp (expectile.noncross)
- renamed 'bundle.density' and 'expectile.cdf' to cdf.bundle and cdf.qp (returning the same class)
bugfixes:
- building and plotting of radial / kriging bases fixed
2011-09-08 version 0.26
new features:
- functions eeplot and eenorm added
- functions penorm, pet, pebeta, peunif etc. compute asymmetries from expectile values
changes:
- univariate expectile computation and estimation optimised
- help pages use simpler, faster examples
- added "..." support for plot function
bugfixes:
- predicting sheets fixed
2011-05-09 version 0.25
new features:
- function expectile.noncross now also works for additive parametric models
- adding methods for class 'expectreg': coef, resid, fitted, effects and predict
- function expectile added for univariate expectiles corresponding to 'quantile' function
changes:
- functions can now handle formulas y ~ 1 and y ~ .
- variable names are now saved throughout the regression functions
bugfixes:
- minor fixes regarding sheets and plotting them
2011-03-22 version 0.21
new features:
- function expectile.noncross added that uses quadratric programming with constraints
- function expectile.cdf added that estimates a CDF from an 'expectreg' object
changes:
- optimising kriging and radial basis computation
- expectile plots only with the use of S3method
- formula evaluation not based on GlobalEnv anymore
bugfixes:
- correcting penalty matrices for non-centered basis
- sheets now plot correctly and return an appropriate object
- correcting sheet intercept in additive models and centered bases
2011-02-24 version 0.20
new features:
- functions (d,p,q,r,e)emq replace the 'koenker' distribution functions, being more general
- function plot.expectreg now also works for boosting
- introducing a 'by' argument for varying coefficients in function base
- function quant.bundle added
- print method for class 'expectreg' added
- predict method for base and therefore for class 'regbase' added
changes:
- boosting results now really seperate for each covariate
- not centering bases is now recognised by expectile.(laws,restricted,bundle,sheets)
- base now returns an object of class 'regbase'
- fixed smoothing option renamed from 'none' to 'fixed'
- schall algorithm returns warning if no convergence is achieved
- vignette is now available via call: 'vignette("expectreg")'
bugfixes:
- correcting the plot function for the 'expectreg' class
- correcting theoretical expectiles of gamma (and therefore exponential, chi squared), lognormal
- user defined covariate basis of type 'special' added to output
2010-09-07 version 0.16
new features:
- function expectile.sheets added
- function bundle.density and plot method added
- functions for theoretical expectiles of: student-t, gamma, exponential, chi squared, lognormal added
changes:
- functions expectile.restricted and expectile.bundle ready for density estimation using bundle.density
- centering bases in function base is now optional
- adding parametric effects to the regression functions
- possibility to choose set of expectiles that are computed
bugfixes:
- correcting schall smoothing for additive models
- fixing schall and ACV smoothing for bundle
- covariates are now returned in correct sorting
2010-03-16 version 0.1
* initial release.