2013-05-22 Martin Maechler
* R/lmrob.MM.R (Mpsi, Mchi, Mwgt): now all work via .Call():
is 9 x faster for n ~ 4000, and a few NAs.
2013-05-18 Martin Maechler
* R/lmrob.MM.R (MrhoInf): new, providing rho(Inf), i.e.,
the scaling constant (rho |--> rho~ = chi)
* R/lmrob.MM.R (Mpsi, Mchi, Mwgt): renamed, exported and documented,
from former hidden lmrob.psifun(), lmrob.chifun(), lmrob.wgtfun().
* TODO: now has an extensive section on psi/rho/chi etc
2013-03-27 Martin Maechler
* DESCRIPTION (Version): 0.9-8, released to CRAN on ..........
2013-03-06 Martin Maechler
* DESCRIPTION (Version): 0.9-7, released to CRAN on 2013-03-06
2013-03-01 Manuel Koller
* R/lmrob.R (lmrob): Adding weights and offset arguments.
2013-02-26 Manuel Koller
* R/lmrob.R (class lmrob): Objects of class lmrob now store
the robustness weights in $rweights (used to be in $weights).
* R/lmrob.R (weights.lmrob): specialized weights() function for
lmrob objects. Returns prior weights by default. Robustness weights
are available via the "type" argument.
2013-02-14 Manuel Koller
* R/lmrob.R (lmrob): setting class "lmrob" also for unconverged
initial estimates.
* R/lmrob.R (print.summary.lmrob, print.lmrob, print.lmrob.S): improved
handling of unconverged estimates and exact fits.
2013-02-07 Manuel Koller
* R/lmrob.R (lmrob): compatibility to lm:
'assign': labels of terms used for fitting.
* R/lmrob.R (summary.lmrob): compatibility to summary.lm for:
'df': degrees of freedom, a 3-vector (p, n-p, p*), the last being
the number of non-aliased coefficients (used to be just (n-p)).
'aliased': named logical vector showing if the original
coefficients are aliased (was not returned before).
2012-09-11 Manuel Koller
* DESCRIPTION (Version): 0.9-4, released to CRAN on 2012-09-11
2012-05-14 Martin Maechler
* R/lmrob.MM.R (.vcov.avar1): fix typo in "posdefify"
2012-03-05 Martin Maechler
* R/lmrob.MM.R (.vcov.avar1): "posdefify" FIXME (negative eigen values)
2012-02-27 Martin Maechler
* DESCRIPTION (Version): 0.8-1-1, released to CRAN on 2012-03-02
2012-02-24 Manuel Koller
* R/lmrob.R (lmrob): added init argument: string, function or list.
* R/lmrob.MM.R (lmrob.fit): added init argument.
* man/lmrob.Rd: Added documentation about init argument.
* man/lmrob.fit.Rd: see above.
2012-02-22 Martin Maechler
* R/nlrob.R (print.summary.nlrob): use full call instead of just
formula. --> shows non-default psi()
2012-02-19 Martin Maechler
* R/nlrob.R (nlrob): now use nls(...., weights = *, ..), no longer
need hack "put everything to the RHS".
2011-12-23 Martin Maechler
* R/nlrob.R (nlrob): for zero weights 'w', the residuals
accidentally where NaN.
2011-12-12 Manuel Koller
* DESCRIPTION (Version): 0.8-1
* src/mc.c (mc_C_d): fixed bug in iteration; see svn log -r 272 & 274
2011-12-09 Martin Maechler
* R/adjbox.R (adjboxStats): fix the sign-error thinko for the case mc < 0.
* man/adjboxStats.Rd: document; test the reflection invariance now.
2011-12-08 Martin Maechler
* R/adjbox.R (adjbox.default): new 'doReflect' argument; default
'TRUE' ensures that 'adjbox(x)' behaves symmetrically in 'x'.
* DESCRIPTION (Version): 0.8-0, released to CRAN on 2011-12-09
2011-10-24 Martin Maechler with Andreas in train to Fribourg
* R/glmrob.R (residuals.glmrob): provide a version of
residuals.glm() which works
* NAMESPACE: residuals.glmrob
2011-10-11 Manuel Koller
* DESCRIPTION (Version): 0.7-8, released to CRAN on 2011-10-26, r270
* man/lmrob.Rd: See also extended with lmrob..M..fit.
* man/lmrob..M..fit.Rd: Extended example with function lmrob.custom.
* R/lmrob.MM.R (lmrob.tau): moving calculation of h to the inside
of the function, so that obj$qr is not required.
* R/ltsPlot.R (ltsPlot, myqqplot): removed superfluous title().
2011-10-11 Martin Maechler
* src/rfltsreg.f: add 'implicit none' and declare everything;
new gfortran gave (correct) warnings.
* R/nlrob.R, man/nlrob.Rd: finally fix "fitted.values"
* tests/nlrob-tst.R (new): start more testing of nlrob()!
2011-08-09 Manuel Koller
* R/plot.lmrob.R (plot.lmrob): Fixing bug reported by Andreas
Papritz. is.null(x$x) is always FALSE since there is always
x$xlevels present.
* inst/doc/lmrob_simulation.Rnw (f.gen): load(file) was missing the
proper path.
2011-05-24 Manuel Koller
* DESCRIPTION (Version): 0.7-6, released to CRAN today
* inst/doc/lmrob_simulation.Rnw: cleaned up chunk headers.
* inst/doc/Makefile: added Makefile that runs qpdf after texi2pdf.
2011-05-23 Manuel Koller
* DESCRIPTION (Version, Date): 0.7-5
* man/lmrob..D..fit.Rd: updated reference to Koller and Stahel
2011 (now with volume, issue and page numbers).
* man/lmrob.control.Rd: updated reference.
* man/lmrob.Rd: update reference.
* inst/doc/lmrob_simulation.bib: updated reference.
* inst/doc/estimating.functions.R (robustness.weights.lmrob.S):
fixed typo.
2011-03-17 Manuel Koller
* DESCRIPTION (Version, Date): 0.7-4
* inst/doc/lmrob_simulation.Rnw: Fixed problem with paths in
source(), save() and load().
2011-03-08 Manuel Koller
* DESCRIPTION (Version, Suggests): 0.7-3 and added xtable and
ggplot2 to Suggests.
* inst/doc/lmrob_simulation.Rnw: re- or moved dependencies that are
not required or just needed for data generation.
* inst/doc/error.distributions.R: removed dependency on skewt.
2011-02-10 Manuel Koller
* DESCRIPTION (Version): Version: 0.7-2
* inst/doc/lmrob_simulation.Rnw: updated title.
* inst/doc/lmrob_simulation.bib: updated references.
* man/lmrob.control.Rd: as above.
* man/lmrob.Rd: as above.
2011-01-28 Martin Maechler
* R/glmrobMqle.R (Huberprop2, ...),
* R/psi-rho-funs.R (huberPsi): replace pmin2/pmax2 by
pmin.int()/pmax.int().
2011-01-26 Manuel Koller
* R/psi-rho-funs.R (psiFunc, chgDefaults): default arguments are
now also set for the E... slots
* tests/psi-rho-etc.R: updated comments
* tests/psi-rho-etc.Rout.save: updated accordingly
2011-01-20 Manuel Koller
Updated reference to Koller and Stahel 2011 paper.
* man/lmrob.control.Rd: updated reference. KS2010 -> KS2011.
* man/lmrob.Rd: see above
* inst/doc/lmrob_simulation.bib: updated ks2011 reference.
* inst/doc/lmrob_simulation.Rnw: see above
* R/lmrob.MM.R (lmrob.control): setting argument: KS2011 instead of
KS2010
2011-01-18 Manuel Koller
* tests/psi-rho-etc.R: added tests to document bug in psi_func
class and pmin2/pmax2 functions.
* tests/psi-rho-etc.Rout.save: results as they should be
* inst/doc/lmrob_simulation.Rnw: added table with tuning constants
used in simulation.
2010-12-04 Martin Maechler
* R/qnsn.R (Qn): even better finite sample correction,
from an analysis of a Qn simulation.
(Qn.old): provide, if needed for back-compatibility.
2010-12-02 Valentin Todorov
* R/ltsReg.R, man/ltsReg.Rd, tests/LTS-specials.R: the raw weights
'raw.weights' added to the returned object
* R/ltsReg.R, src/rfltsreg.f, man/ltsReg.Rd: options nsamp="exact"
and nsamp="best" fixed; Added 'trace' parameter
2010-12-02 Martin Maechler
* tests/mc-strict.R (adjOutlyingness): adapt tests to the fixed
mc() code; using some tolerance... still need to be tested on Win/Mac/..
* R/nlrob.R (summary.nlrob): do not compute 'se' if not converged;
update help page;
(print.summary.nlrob): print "non-convergence"
* man/summary.nlrob.Rd, man/predict.lmrob.Rd, ..: update and
trivial changes.
* R/qnsn.R (Qn): fixed the consistency constant (thanks to Peter
Ruckdeschel), slightly adapting the finite sample factors for n = 2:9.
2010-12-02 Manuel Koller
* DESCRIPTION (Version): 0.6-9
* R/mc.c: fixed several bugs, consequent use of eps[0] and eps[1]
to avoid numerical problems, fixed memory corruption bug that
caused segfaults from time to time.
* tests/mc-etc.R: see below
* tests/mc-etc.Rout.save: see below
* tests/mc-strict: updated tests and results
* man/mc.Rd: see below
* R/mc.R (mc.default): updated eps1 and eps2 defaults.
* inst/doc/lmrob_simulation.Rnw: minor cosmetic changes
* inst/doc/graphics.functions.R: added some helper functions to
ease working with color palettes used by ggplot2.
2006-10-27 Andreas Ruckstuhl
and
2010-10-28 Martin Maechler
* DESCRIPTION (Version): 0.6-8
* R/lmrobPredict.R, R/glmrobPredict.R: new more careful prediction
methods, now documented in
* man/predict.lmrob.Rd, man/predict.glmrob.Rd:
and no longer in
* man/summary.lmrob.Rd:
2010-10-13 Manuel Koller
* DESCRIPTION (Version): 0.6-7
* R/lmrob.MM.R (lmrob.psifun, lmrob.rhofun, lmrob.wgtfun):
as.double caused input to loose dimension attribute, fixed.
* inst/doc/lmrob_simulation.Rnw: Updated vignette title. Enhanced
aggregation procedure.
* inst/doc/lmrob_simulation.Bib: Fixed entry KS2010, no more
trouble with bibtex.
* inst/doc/aggr_results.Rdata: updated simulation results. n = 400
missing for now.
* inst/doc/error.distributions.R: added cskt distribution: centered
skewed t-distribution.
* inst/doc/simulation.results.R: small changes in proclist generation.
2010-10-11 Manuel Koller
* DESCRIPTION (Version): 0.6-6
(Depends): Depends on R 2.9.0 (use of grepl)
* man/lmrob.Rd: Design Adapted Scale estimate renamed to Design
Adaptive Scale estimate
* man/lmrob.fit.Rd: ditto
* man/lmrob..D..fit.Rd: ditto
2010-10-08 Manuel Koller
* DESCRIPTION (Version): 0.6-5
* inst/doc: added directory.
* inst/doc/lmrob_simulation.Rnw: vignette containing simulation
study of Koller and Stahel (2010).
* inst/doc/lmrob_simulation.bib: bibtex file for vignette
* inst/doc/graphics.functions.R: R code used in vignette, graphics
helper functions.
* inst/doc/error.distributions.R: R code used in vignette, custom
distribution functions.
* inst/doc/simulation.functions.R: R code used in vignette,
functions used in simulations.
* inst/doc/estimating.functions.R: R code used in vignette, extra
estimating functions.
* inst/doc/asymptotic.max.bias.Rdata: Cache of calculations, to
speed up vignette making.
* inst/doc/aggr_results.Rdata: Cache of calculations, to
speed up vignette making.
* .Rbuildignore: Omitting temporary Sweave output files from the
build.
* src/init.c: added function to process simulation output (used in
vignette).
* src/lmrob.c (R_calc_fitted): ditto
* src/robustbase.h: ditto
2010-10-08 Manuel Koller
* DESCRIPTION (Version): 0.6-4
* inst/CITATION: Added my name.
* man/lmrob.control.Rd: Added more information to
"setting"-Argument.
* man/lmrob.Rd: ditto. Added example for "setting" argument.
* R/lmrob.R (lmrob): Moved call of lmrob.control to function code.
* R/lmrob.MM.R (lmrob.control): Changed setting KS2010: uses lqq
instead of ggw. Fixed bug when using unknown setting.
* R/lmrob.MM.R (lmrob.fit): Check for initial estimator. Issues
warning if neq S.
2010-08-13 Martin Maechler
* DESCRIPTION (Version): 0.6-3
* R/lmrob.MM.R (lmrob.psifun, ..., ghq): cosmetic changes; cleaner
* src/lmrob.c (rho, psi, psip, ...): cleaner; partly faster; also
ok for x = +-Inf....
* tests/lmrob-psifns.R: test psi(), rho(), etc -- also nice plots:
tests/rob-psifns.pdf after .. check.
2010-08-13 Manuel Koller
* R/lmrob.MM.R: changed the naming of lgw to the more accurate lqq
"linear, quadratic, quadratic" (by construction of psi prime).
* tests/lmrob-psifuns.R: changed the naming of lgw to the more
accurate lqq
* man/lmrob.control.Rd: ditto
* man/summary.lmrob.Rd: updated documentation to account for "cov"
argument of vcov.
* R/lmrob.R (lmrob): warning and check for method argument improved.
2010-08-05 Manuel Koller
* DESCRIPTION (Version): 0.6-2
* R/lmrob.MM.R (lmrob.const): replaced lmrob.ggw.const by
lmrob.const, which now supports ggw and lgw.
(lmrob.control, lmrob.lgw.findc, lmrob.conv.cc, lmrob.psi2ipsi):
support for lgw psi function.
(lmrob.efficiency, lmrob.bp): functions to calculate the efficiency
and breakdown point of an M-estimator.
(lmrob.tau, lmrob.tau.fast.coefs): Updated constants in all.equal
check for Hampel psi function. Updated constants for all the
supported psi functions.
(lmrob..D..fit): switched order of updating covariance matrix and
updating control$method. This caused the correction factors to be
set incorrectly in some cases.
* src/lmrob.c (normcnst, rho_lin, psi_lin, psip_lin, wgt_lin)
(rho, psi, psip): Added support lgw psi function.
* man/lmrob.control.Rd: updated to account for lgw.
* tests/lmrob-psifns.R: added tests lgw psi function
* tests/lmrob-psifns.Rout.save: updated accordingly
* tests/lmrob-methods.Rout.save: updated, since constants for
calculation of taus changed slightly
* R/lmrob.R (vcov.lmrob): added possibility to calculate covariance
matrix with another estimation method.
2010-07-12 Manuel Koller
* DESCRIPTION (Version): 0.6-1
* src/lmrob.c (normcnst, rho_ggw, psi_ggw_vec, psi_ggw, psip_ggw)
(wgt_ggw): Added support for custom constants for psi ggw.
* R/lmrob.MM.R (.vcov.w): Modified constants for ggw psi function.
(lmrob.control): Added method to calculate constants for ggw psi
function.
(lmrob.conv.cc, lmrob.ggw.mx, lmrob.ggw.ms, lmrob.ggw.ac)
(lmrob.ggw.bp, lmrob.ggw.finda, lmrob.ggw.findc, lmrob.ggw.const):
Methods to calculate constants for ggw psi.
(lwgt): control argument optional.
* tests/lmrob-psifns.R: added tests for custom tuning constants
* tests/lmrob-psifns.Rout.save: added tests for custom tuning
constants
* experi-psi-rho-funs.R: added ggwPsi function, standardized
s.t. Dpsi(0) = 1. Putting this into R/psi-rho-funs.R results in an
error while "preparing for lazy-loading".
2010-06-25 Manuel Koller
* DESCRIPTION (Version): 0.6-0
* NAMESPACE: dropped lmrob.control.sfs
* R/lmrob.R (lmrob): "method" argument
* man/lmrob.Rd: "method" argument
* R/lmrob.MM.R (lmrob.control, lmrob.control.sfs): argument
setting="KS2010" replaces lmrob.control.sfs. Setting a method
involving a D step, sets default psi to ggw.
* man/lmrob.control.Rd: "setting" argument
* tests/lmrob-methods.R: "methods" and "psi" argument
* tests/lmrob-methods.Rout.save: call output in summary changed
* tests/lmrob-psifns.Rout.save: call output in summary changed
* man/lmrob.control.Rd: Documentation details for tuning constants
and alternative constants for .85 efficiency at the normal.
2010-06-24 Manuel Koller
* DESCRIPTION (Version): 0.5-1.1
* tests/lmrob-psifns.R: added Hampel psi function example and
updated coefficients of ggw.
* src/lmrob.c: replaced "gws" with "ggw" in function names, "gwgt"
with "welsh", replaced pow() with R_pow(), implemented support for
coefficient vectors. Fixed problem with "ggw" on 32-bit machines.
* R/lmrob.MM.R: changed the way coefficients for psi functions are
handled: coefficient vectors are now supported.
* R/lmrob.MM.R (lmrob..M..fit): class attribute for init entry in
lmrob..M..fit was missing.
2010-06-23 Manuel Koller
* tests/lmrob-methods.R: tests for compatibility between manual
construction of different methods with specifying method argument
in lmrob.
* tests/lmrob-methods.Rout.save: expected results
* tests/lmrob-psifns.R: tests for the support of different psi
functions.
* tests/lmrob-psifns.Rout.save: expected results
* tests/lmrob-data.Rout.save: updated results to reflect slightly
modified summary output (more options)
* man/lmrob.Rd: updated to reflect changes in options
* man/lmrob..D..fit.Rd: added documentation
* man/lmrob..M..fit.Rd: updated example
* man/lmrob.control.Rd: updated to reflect changes in options
* man/lmrob.fit.Rd: updated from lmrob.fit.MM to reflect changes in
options
* man/lmrob.fit.MM.Rd: moved to lmrob.fit.Rd
* man/lmrob.S.Rd: coef -> coefficients in example
* R/anova.lmrob.R (length.tl, anovaLmrobPair, anovaLmrobList):
Added checks to ensure soundness of result for methods other than
"MM".
2010-06-21 Manuel Koller
* R/lmrob.MM.R (lmrob.control, lmrob.control.sfs, lmrob.fit.MM)
(.vcov.w, .vcov.avar1, lmrob..M..fit, lmrob.S, lmrob..D..fit)
(lmrob.kappa, lmrob.tau, lmrob.hatmatrix, lmrob.leverages)
(lmrob.psi2ipsi, lmrob.psifun, lmrob.chifun, lmrob.wgtfun)
(residuals.lmrob.S, lmrob.E, ghq): Support for psi functions
other than bisquare as well as SMDM estimates.
* R/lmrob.R (lmrob): Support for psi functions other than bisquare
as well as SMDM estimates. Replaced lmrob.fit.MM by lmrob.fit.
* src/robustbase.h: declarations of new functions in lmrob.c
* src/init.c: definitions of new functions in lmrob.c
* src/lmrob.c: added support for psi functions other than
bisquare, including vectorized accessor functions ("R_psifun",
"R_chifun" and "R_wgtfun"). "wgt" is now always used to calculate
robustness weights. Function to calculate scale estimate
iteratively ("R_find_D_scale").
* NAMESPACE: added lmrob.fit, lmrob..D..fit and lmrob.control.sfs
to export.
* DESCRIPTION: added my name to the list.
2010-04-01 Martin Maechler
* DESCRIPTION (Version): 0.5-1
2009-11-19 Martin Maechler
* R/covMcd.R (.fastmcd): nLarge = 100'000 (instead of previous 5000),
for now. nsamp <= integer.max is asserted now.
* R/covMcd.R (covMcd): nmini = 300 is now an optional argument.
* R/rrcov.control.R (rrcov.control): 'nmini = 300' ...
* R/ltsReg.R (ltsReg.default): .fastmcd(..., nmini = 300)
* src/rffastmcd.f (rffastmcd): new argument nmini
2009-11-18 Martin Maechler
* DESCRIPTION (Version): 0.5-0-1, released to CRAN
* tests/mc-strict.R: don't use the longmemo example for now.
(platform dependence !)
2009-11-11 Martin Maechler
* tests/glmrob-1.R: add moe explicit glmrob() tests;
including the simple puzzling 1-outlier problem.
* R/glmrobMqle.R: do use getRversion() !
2009-06-27 Martin Maechler
* data/wagnerGrowth.rda: add the "wagner data" (continuous + categorical)
* man/wagnerGrowth.Rd: ditto
2009-06-05 Martin Maechler
* R/glmrob.R (glmrob): add the option start = "lmrobMM", using
a *robust* start for {essentially} glm().
* man/glmrob.Rd: add an example for that <<<<<______________ FIXME
2009-06-04 Martin Maechler
* DESCRIPTION (Version): 0.5-0; *not* released
* tests/glmrob-1.R: more glmrob() examples
2009-06-03 Martin Maechler
* R/glmrobMqle.R (mFormat): make 'trace=TRUE' also depend on
getOption("digits").
* R/glmrobMqle.R, R/glmrob.R: changes from Andreas Ruckstuhl,
to support family = "Gamma".
Unfortunately, these at first also very slightly change binomial,
poisson. ==> few small changes by MM.
* R/glmrob.R (glmrob): allow y (in "y ~ ...") to be *factor*
2009-01-17 Martin Maechler
* DESCRIPTION (Version): 0.4-5 : bug-fixing release
* tests/tmcd.R: add a regression test for the bug
* src/rffastmcd.f: rfrdraw() w/o 'seed'
* src/rfltsreg.f: ditto
2009-01-10 Martin Maechler
* src/rf-common.f (rfrangen): drop unused 'seed' argument
(rfrdraw): ditto
* src/rffastmcd.f: get rid of TABs (gfortran -Wall complains)
* man/exAM.Rd, man/possumDiv.Rd, ...: Rd_parse fixes
2008-11-28 Martin Maechler
* R/OGK.R (covOGK): also return n.iter and weight.fn.
* man/covOGK.Rd: note about the "silly" weight.fn
2008-10-15 Martin Maechler
* R/lmrob.MM.R (lmrob.S): fix error message s/larger/smaller/
thanks to Keith Ponting.
2008-10-01 Martin Maechler
* src/wgt_himed_templ.h: fix leading comment
2008-08-29 Martin Maechler
* DESCRIPTION (Version): 0.4-3 for CRAN release
* tests/mc-strict.R: define 'isMac' and tweak the tests;
thanks to reports from Rory Winston.
2008-08-09 Martin Maechler
* NAMESPACE: import stats::cov {since others hide it!}
2008-08-05 Martin Maechler
* R/huber.R (huberM): replace `s' by 's' in errors and warnings.
* R/qnsn.R, R/plot.lmrob.R, R/ltsReg.R, R/ltsPlot.R: ditto
2008-08-04 Martin Maechler , really from
Valentin Todorov
* R/ltsReg.R (ltReg.default): if(mcd) call covMcd() with correct alpha
* R/covMcd.R (covMcd): drop (n-1)/n correction, as cov.wt()
contains that per default (in R, not in S-plus!).
* R/tolEllipse.R (tolEllipsePlot): no text() for id.n == 0
* DESCRIPTION (Version): 0.4-2 released to CRAN.
2008-08-04 Martin Maechler
* src/wgt_himed.c: renamed wgt_himed.c_templ to
* wgt_himed_templ.h:
2008-08-02 Martin Maechler
* DESCRIPTION (Version): 0.4-1 ready for release to CRAN.
* R/ltsReg.R (ltsReg.default): don't add artificial "Y" y-name
* src/lmrob.c (rwls): first call to sum_rho() is only needed for
tracing (was used for lambda iterations).
* R/glmrob.R (glmrob),
* R/glmrobMqle.R (glmrobMqle): introduce 'trace' option to trace
the robustness iterations {*not* part of control: does *not*
influence result}
* man/CrohnD.Rd: new data example robust poisson regression
* data/CrohnD.rda:
2008-01-25 Martin Maechler
* R/covMcd.R (.fastmcd): fix nsamp="exact" to use "all"
* src/rffastmcd.f: krep=0 <==> nsamp="exact" : all := TRUE;
new argument i_trace; and use intpr() and dblepr()
* man/covMcd.Rd: pass 'trace' to .fastmcd() and Fortran
* src/rf-common.f (rfncomb): give "error" message for very
large 'comb'
* tests/tmcd.R: add test for nsmap = "exact"
2008-01-05 Martin Maechler
* man/ambientNOxCH.Rd: new data set from RenĂ© Locher,
* data/ambientNOxCH.rda: showing some lmrob-nonconvergence
2007-12-22 Martin Maechler
* src/lmrob.c (rwls): if (trace_lev >= 3) show beta vector.
2007-12-13 Martin Maechler
* man/summarizeRobWeights.Rd: add toy example
* R/lmrob.R (summarizeRobWeights): cosmetic change; notably
finishing line when weights were practically 0/1.
* tests/MCD-specials.Rout.save {adapt from change of 11-07}
2007-11-07 Martin Maechler
* R/covMcd.R (covMcd): if we have singularity, also say so, even
if trace is FALSE.
* R/covMcd.R (singularityMsg): for "on.hyperplane", concatenate
long coefficient vector.
2007-10-25 Martin Maechler
* man/pulpfiber.Rd: New data set (p=8 = 4 + 4)
* data/pulpfiber.tab: from "Robust Multivariate Regression (2004)".
* tests/lmrob-ex12.R: use versions of predict[.lmrob]()
2007-10-24 Martin Maechler
* man/condroz.Rd: set latin1 encoding (and fix typo)
* R/lmrob.R (predict.lmrob): define predict() and model.matrix()
methods, working via "lm" methods.
* NAMESPACE: export vcov()
* man/summary.lmrob.Rd: some docu of new methods
2007-08-02 Martin Maechler
* R/adjoutlyingness.R (adjOutlyingness): small improvements,
getting rid of loops; fix an obvious typo (in original code).
Leave away the Inf and NaN that from abs(Y[] - m) / ( tup | tlo)
2007-07-21 Martin Maechler
* DESCRIPTION (Author): add Tobias
* src/rmc.c (h_kern): revert to absolute test; many more experiments
* tests/mc-etc.R (x3): a smallish "extreme" case of "non-convergence"
2007-07-20 Martin Maechler
* tests/mc-etc.R: new testing of mc() and related
* src/rmc.c (mc_C_d): changed work[] and weight[] to 0-indexing;
this is hopefully the last bug ...
* R/adjbox.R (adjbox.formula): use adjbox(), not boxplot()!
* man/adjboxStats.Rd: added
2007-07-20 Tobias Verbeke
* R/adjbox.R: Skewness-adjusted boxplot ported from Matlab LibRA
* man/adjbox.Rd:
* data/los.rda, man/los.Rd:
* data/condroz.rda, man/condroz.Rd: two dataset related to medcouple
2007-07-19 Martin Maechler
* R/adjoutlyingness.R: new (also from the Antwerpen MC collection)
* DESCRIPTION (Depends): R >= 2.3.1, so we can use
* NAMESPACE (useDynLib): .registration = TRUE
* R/*.R (.C, .Fortran): now can use name variable instead of string
and drop 'PACKAGE = ".."'
2007-07-18 Tobias Verbeke and Martin Maechler
* src/rmc.c, src/robustbase.h (rmc): code for medcouple (MC);
needs debugging: infinite loops and segmentation faults
* R/mc.R (mc): new mc() for MedCouple
2007-07-16 Martin Maechler
* DESCRIPTION (Version): 0.4-0 - definitely made progress
* R/plot.lmrob.R (plot.lmrob): recompute robust Mahalanobis
distances and cache them with the object.
2007-07-10 Martin Maechler
* R/lmrob.R (summarizeRobWeights): new defaults; work but warn
when 'eps' and 'eps1' lead to weights both close to 0 and 1.
* R/covPlot.R: don't warn for which = "all"
* man/covPlot.Rd: example
2007-06-28 Martin Maechler
* data/kootenay.tab: add "famous" data set
* man/kootenay.Rd:
* data/cushny.R: add the other "famous" simple data set
* man/cushny.Rd:
2007-06-20 Martin Maechler
* DESCRIPTION (Version): 0.2-8 ready to be released
* src/lmrob.c (rwls): patch from Matias Salibian: Don't do any
"lambda iterations" anymore, they are remnants from old "experiments".
2007-06-19 Valentin Todorov
* R/covMcd.R: usage of simulated finite sample correction factors fixed:
* - case p=1 fixed
* - simulated corrections used only when the Pison et.al.
* formula is definitely wrong (negative or very large)
2007-06-09 Martin Maechler
* tests/lmrob-data.R, *.Rout.save: new consistency tests for lmrob()
* R/biweight-funs.R (tukeyPsi1, tukeyChi): renamed from
lmrob.Psi() and lmrob.Chi(); also renamed
* man/tukeyPsi1.Rd, man/tukeyChi.Rd: updated, also mentioning "psiFunc"
* src/lmrob.c (rwls): now controlling *relative* error in MM
iterations, i.e., convergence happens when ||b1 - b2|| < eps * ||b1||.
The above 'eps' is now part of lmrob.control() instead of
hard-wired in C code.
* R/lmrob.MM.R (lmrob.MM): new argument 'trace.lev'
* src/robustbase.h (R_lmrob_MM): add trace_lev argument
2007-06-08 Martin Maechler
* src/lmrob.c (refine_fast_s): slight change in warning()
[make message nicer for future R >= 2.5.1]
2007-04-21 Martin Maechler
* R/ltsReg.R: replace 'quan' by 'h' internally
* R/covMcd.R: ditto
* R/covMcd.R (h.alpha.n): renamed from "quan.f"() and now exported:
* NAMESPACE: and hence
* man/h.alpha.n.Rd: documented, and linked to from here
* man/ltsReg.Rd, man/covMcd.Rd:
2007-04-18 Valentin Todorov
* R/covMcd.R: use simulated finite sample correction factors {FIXME!}
2007-04-11 Martin Maechler
* R/ltsReg.R (ltsReg.default): 'ans$X' now has "intercept first"
to match coefficient vector.
* R/ltsReg.R (summary.lts): no need to reorder 'R' (cholesky) anymore
2007-04-02 Valentin Todorov
* R/covMcd.R: the (repeated) calculation of the consistency
* correction factor for the raw and reweighted MCD covariance
* replaced by a call to a function MCDcons()
2007-03-30 Valentin Todorov
* R/covMcd (print.mcd): solve a conflict with fastmcd() in
package robust: both return an object of class "mcd"
2007-03-28 Martin Maechler
* R/covMcd.R (.fastmcd): subsample size myk <- p+1 (was 'p')
2007-03-27 Martin Maechler
* R/covMcd.R: cosmetic changes; comments
* src/rffastmcd.f: ditto
* tests/tmcd.R: added very small sample (n < 2p) examples.
2007-03-26 Martin Maechler
* R/ltsReg.R (print.summary.lts): signif.stars: instead of FALSE,
use same default as for lm(.).
* man/summary.lts.Rd
2007-03-26 Martin Maechler
* R/covMcd.R (covMcd): be more tolerant about small n,
notably n < 2p, now only requiring n >= p + 2
* src/rf-common.f (rfnbreak): really unused
* src/rfltsreg.f, src/rffastmcd.f: comment use of rfnbreak()
2007-03-24 Martin Maechler
* R/covMcd.R (covMcd): ans$raw.cov should be matrix even for p==1
2007-03-21 Martin Maechler
* R/ltsReg.R (ltsReg.default): fix long-standing ("rrcov") bug of
wrong coefficient order in ltsReg(x,y, intercept=FALSE).
* tests/LTS-specials.R: add regression-test
2007-02-08 Martin Maechler
* R/covMcd.R (singularityMsg): moved the remaining cases out of
covMcd().
2007-01-24 Martin Maechler
* R/covMcd.R (covMcd): save 'singularity' info as list, and don't
append it as string to 'method'.
(singularityMsg): new utility used for printing singularity of MCD
(print.mcd): using strwrap() instead of "\n" such that print()
obeys options("width").
2006-12-28 Martin Maechler
* DESCRIPTION (Version): 0.2-7 released to CRAN
* src/rfltsreg.f: get rid of warnings, notably some extraneous args.
* src/rffastmcd.f: ditto
* inst/test_MCD.R (mortality, mort3): add example {near singular}
* tests/tmcd.R: test "near singular" example, using new tolSolve:
* R/rrcov.control.R (rrcov.control): new tolSolve = 1e-14
* R/covMcd.R (covMcd): use 'tolSolve' instead of 1e-10 for
mahalanobis' solve(*, tol).
2006-12-21 Valentin Todorov
* R/ltsPlot.R (ltsPlot): for "rqq", use *standardized* residuals
2006-10-20 Martin Maechler
* src/rfltsreg.f (rfltsreg): more comments; and minor cleanups
* R/ltsReg.R (.fastlts): slightly simplified 'nsamp' checking and
setting
* DESCRIPTION (LazyData): yes
* man/heart.Rd: mention survivals' "heart" data
2006-10-18 Martin Maechler
* src/rfltsreg.f: less "if() l1,l2,l3" ; better indenting; etc
worked pretty hard on C translation but that still seg.faults (!)
2006-10-04 Martin Maechler
* INDEX: added a "manual" INDEX which has all the data sets at
the end.
* R/OGK.R (covOGK): more default arguments: n.iter = 2, and
weight.fn = hard.rejection
* R/OGK.R (s_mad, s_IQR): more "scale functions" for (O)GK
* tests/OGK-ex.R: added examples for new s_*() scales;
* tests/OGK-ex.Rout.save: new
2006-10-03 Martin Maechler
* R/qnsn.R (s_Qn, sSn): new (trivial) wrapper functions for easier
use in covOGK().
* man/Sn.Rd, man/Qn.Rd: ditto
* man/covOGK.Rd: example with the above
2006-09-30 Martin Maechler
* DESCRIPTION (Version): 0.2-6
* R/covPlot.R (covPlot): added 'labels.id', 'cex.id' and
'label.pos' arguments, "parallel" to plot.lm() and improved the
labeling accordingly.
* man/covPlot.Rd: 'ask=TRUE' gives problems in R <= 2.3.1
2006-09-29 Martin Maechler
* DESCRIPTION (Version): 0.2-5
* data/radarImage.rda: add 'radarImage' data set from MMY-book
* man/radarImage.Rd: ditto
* data/toxicity.rda: added 'toxicity' data set (from MMY)
* man/toxicity.Rd: ditto
* src/rf-common.f (rfishsort): swap integer declaration order
2006-09-23 Martin Maechler
* man/covPlot.Rd: document covPlot() as well, since that is more
generally usable.
* src/rf-common.f (rfrangen): declare unifrnd() as double prec.;
also in other places; now using 'implicit none' to catch such
things more easily.
* R/tolEllipse.R (tolEllipsePlot): classic=TRUE: do not prepare
a side-by-side plot, since we *over*plot.
Rename 2nd argument from 'mcd' to 'm.cov', since it really only
needs to have a mean ('center') and covariance component.
2006-09-05 Martin Maechler
* DESCRIPTION (Version): 0.2-3 uploaded to CRAN
* tests/tlts.R: rename test functions and move to new file
* inst/test_LTS.R:
* tests/tmcd.R: rename test functions and move to new file;
do not 'time' them, since we want to use *.Rout.save files.
* inst/test_MCD.R:
* src/Makevars: add long overdue $(FLIBS)
2006-09-04 Martin Maechler
* R/ltsReg.R (ltsReg.default): add argument name: sort(*, partial = )
* R/covMcd.R (covMcd): ditto
2006-06-27 Martin Maechler
* DESCRIPTION (Depends): no longer depend on "MASS" which is suggested.
* NAMESPACE, man/summarizeRobWeights.Rd: export and document
summarizeRobWeights().
* R/lmrob.R (summarizeRobWeights): improvement (singular/plural etc)
* R/ltsReg.R (ltsReg.formula): add 'subset' etc; and follow
guidelines in developer.r-project.org/model-fitting-functions.txt.
* tests/tmcd.R: typo; add timing comparison
* R/glmrob.R (glmrob): family = "gaussian" now dispatches to lmrob()
2006-06-24 Martin Maechler
* R/glmrob.R (print.summary.glmrob): as for lmrob(), now use
summarizeRobWeights() and printControl().
* R/lmrob.R (printControl): more flexible
* man/glmrob.Rd: update examples with 'weights.on.x'
2006-06-23 Andreas Ruckstuhl
* R/glmrobMqle.R (wts_HiiDist): *row*Sums()!
2006-06-13 Martin Maechler
* DESCRIPTION (Version): 0.2-2
* src/Makevars (PKG_LIBS): need this (in particular for Windows)
2006-06-08 Martin Maechler
* src/lmrob.c (refine_fast_s): made sure 'conv *is* called as *Rboolean
2006-06-07 Martin Maechler
* src/lmrob.c (refine_fast_s): print warning in case of non-convergence
* DESCRIPTION (Version): 0.2-1 - pre-release
* R/lmrob.MM.R (lmrob.control): decrease refine.tol "back" to 1e-7
* man/lmrob.control.Rd: ditto
* src/lmrob.c (rho_biwgt): fix embarassing sign typo/thinko
2006-06-06 Martin Maechler
* DESCRIPTION (Version): 0.2-0 "pre-released" to Andreas & Matias
* src/lmrob.c (rwls): return *max_it = #{iterations used}
* R/lmrob.MM.R (lmrob.MM): return #{iterations} back to R level
* R/lmrob.R (summary.lmrob): and make sure they are printed.
2006-06-05 Andreas Ruckstuhl
* R/anova.lmrob.R (anova.lmrob): new, for model comparison
both "Wald" and "Deviance" tests.
* man/anova.lmrob.Rd: docu + example
2006-05-31 Martin Maechler
* R/covMcd.R (covMcd): use R' RNG
* R/covMcd.R (print.summary.mcd): is new; also print.mvd and
summary.* are now here, and improved.
* src/rfltsreg.f (rfltsreg): use R's RNG
* src/rffastmcd.f (rffastmcd): ditto
* src/rf-common.f (rfrangen): use unifrnd(), and comment out
the previous uniran()
* src/R-rng4ftn.c: use R's RNGs also for Fortran, in
2006-05-29 Martin Maechler
* src/lmrob.c: use R's unif_rand() instead of C's rand()
* R/lmrob.MM.R: set/save R's .Random.seed, possible from 'seed'
2006-05-29 Martin Maechler
* DESCRIPTION: version 0.1-7; the last one with C's rand()
* R/lmrob.MM.R: it's "$seed" from init.S , not M(M)-estimate
2006-04-25 Martin Maechler
* R/biweight-funs.R (lmrob.Psi): unification, rescaling, such
that Psi'(0) = 1
2006-04-22 Martin Maechler
* R/lmrob.MM.R (lmrob.S): new (control) option 'best.r.s = 2';
was hardcoded in C
* src/lmrob.c (R_lmrob_S): 'best_r' is now argument; further,
seed_rand is passed to the fast_s*() sub functions, so we
can call them from R {and decide ourselves "if fast".
* src/rf-common.f: new file for functions identical
in rffastcmd.f and rfltsreg.f.
2006-04-21 Martin Maechler
* data/NOxEmissions.Rd: add the large dataset from RenĂ© Locher
* man/NOxEmissions.Rd:
2006-04-21 Martin Maechler
* R/lmrob.MM.R (lmrob.MM): return robustness weights as 'wt'
2006-04-18 Martin Maechler
* R/lmrob.MM.R (lmrob.control): change default for 'compute.rd'
to FALSE ==> robust (Mahalanobis) distances are *not* computed by
default. This prevents singularity errors which happen frequently
as soon as X contains (large) factors (with few levels).
2006-04-15 Martin Maechler
* man/pension.Rd: add plots and simple 1st aid trafo (!)
* R/glmrobMqle.R (glmrobMqle): weights.on.x = "hat" couldn't have
worked(!) - still needs work __TODO__
* R/lmrob.MM.R (lmrob.S, lmrob.MM): simplified computations
achieving ~ 10% faster execution for a (n,p) = (500,20) example.
2006-04-01 Martin Maechler
* src/lmrob.c: new lmrob() code from Matias' roblm package
* R/lmrob.*.R: renamed s/roblm/lmrob/ but also refactored
* man/lmrob.*.Rd: and added a bit
* TODO: section on lmrob()
2006-03-20 Martin Maechler
* DESCRIPTION (Version): 0.1-4
* data/vaso.rda: one value, vaso[32,2], had a typo; it was '0.3'
but should have been '0.03'
2006-03-16 Martin Maechler
* R/OGK.R (scaleTau2): add consistency correction (for OGK).
* man/scaleTau2.Rd: added
* R/glmrobMqle.R: ni=0 special casing (not finished yet).
* tests/binom-ni-small.R: more testing of ni=0
2006-03-14 Martin Maechler
* TODO, Done, DESCRIPTION: updated, ready for release
2006-03-14 Andreas Ruckstuhl
* R/anova-glmrob.R (anova.glmrob): new function
* man/anova.glmrob.Rd: and documentation, replacing previous modsel.*()
2006-02-24 Martin Maechler
* R/print.lts.R: moved to this file (and simplified slightly) to
* R/ltsReg.R (ltsReg): more cleanup; fix (y ~ 1) and (y ~ 0) properly
2006-02-23 Martin Maechler
* tests/LTS-specials.R: new file
2006-02-22 Martin Maechler
* tests/MCD-specials.R: new tests
* tests/huber-etc.R: tests for huberM() moved from pkg 'sfsmisc'
2006-02-21 Martin Maechler
* R/glmrob.R (summary.glmrob): "Std. Error" (with blank!)
2006-02-18 Martin Maechler
* R/glmrobMqle.R (glmrobMqle): make it work for ncoef == 0
* tests/glmrob-specials.R: new tests
* R/glmrob.R (print.summary.glmrob): the same as print.glmrob + more
2006-02-17 Martin Maechler
* R/glmrob.R (vcov.glmrob): added
(print.glmrob): fix typo
* tests/binom-ni-small.R: add from Martin's old "robGLM1" package
* R/glmrob.R (glmrob): '...' passed to glmrob.control(...)
2006-02-17 Valentin Todorov
* FIXED - .fastmcd and .fastlts no more return everything
* fixed problems in ltsReg in case of location model Y~1 (i.e. x
is missing in ltsReg.default())
* ltsReg & covMcd - added options 'best' and 'exact' for nsamp
* ltsReg & covMcd - added parameter for controlling the small
sample correction factors - use.correction
* rrcov.control - added parameter for controlling the small sample
correction factors - use.correction
* ltsReg & covMcd - output of the consistency and the small sample
correction factors
2006-02-09 Martin Maechler
* DESCRIPTION (Version): 0.1-2 ready for CRAN "baby release"
* man/vaso.Rd: fix longstanding typo: con*s*triction
* R/glmrob.R (glmrob): Oops! the *.control() function must only
have one "." and must be *called*
2006-02-08 Martin Maechler
* R/glmrob.R,......: added 'glmrob' and 'nlrob' from Andreas Ruckstuhl
2006-02-02 Martin Maechler
* R/covMcd.R: all moved from 'rrcov' after Valentin's
* R/ltsReg.R: "green light"
..........
2006-01-25 Martin Maechler
* DESCRIPTION (Version): 0.0-1 "prerelease" ready
* man/psi_func-class.Rd: version "checked in" and ready for pre-release
* man/chgDefaults-methods.Rd: ditto
2006-01-17 Martin Maechler
* NAMESPACE: added name space
* R/OGK.R (covOGK): added this; even though, the default scale
estimate is *not* consistent
* tests/tests-OGK.R: minimal 'test'
2006-01-16 Martin Maechler
* man/starsCYG.Rd: clean up of documentation for the Rousseeuw data
* man/aircraft.Rd: from Valentin
* ....
--------------- all these are for those things from 'rrcov' -------------------
2005-12-28 Valentin Todorov
* 0.2-11
* added more data sets from Rousseeuw & Leroy: telef, lactic,
pension, pilot, cloud, education, airmay
* fixed codoc discrepancies in the data sets stars and wood
* ltsReg & covMcd - added control parameter for the small sample correction factors
* ltsReg & covMcd - output of the consistency and the small sample correction factors
2005-10-24 Valentin Todorov
* 0.2-10
* minor corrections in the help of covPlot
* fixed bug in covPlot in case of class=TRUE
* tolellipse - both robust and classical ellipse are superimposed
* added directory inst/bm containing benchmarks comparing covMcd and ltsReg
to the corresponding functions in MASS, S-PLUS and Matlab
2005-10-17 Valentin Todorov
* 0.2-9
* minor corrections in the help of covMcd
* minor correction in ltsReg (false corrected to FALSE)
* covMcd - fixed the limitation on the number of variables <= 50
* ltsReg - fixed the limitation on the number of variables <= 50
* added function summary.mcd which prints (additionally to the output
of print.mcd) the correlation matrix (if requested), the eigenvalues
of the covariance or correlation matrix and the robust distances.
* added control object for the estimation options rrcov.control and
used in covMcd and ltsReg
2005-09-20 Valentin Todorov
* 0.2-8
* ltsReg: added formula interface
* ltsReg: adde generic functions summary.lts and print.summary.lts
* ltsReg: fixed a problem with reordering of the coeficients even in case
without intercept
2005-04-16 Valentin Todorov
* 0.2-7
* ltsReg: fixed a bug related to nsamp -> it was hard-coded = 500
in Fortran
* ltsPlot: default for id.n changed - instead of 3, now it is the
number of identified outliers
* ltsPlot: help enhanced
* covMcd, covPlot, tolellipse: tol.inv parameter changed to tol,
according the change in mahalanobis() in 2.1.0
2004-12-26 Valentin Todorov
* 0.2-6
* a bug in ltsReg (rsquared) fixed
* fixed CRLF line endings in FORTRAN sources
* fixed a problem in covMcd: in case of p=1 and cov=[0], the cov matrix was
a double instead of a matrix, which resulted in errors in subsequent
calls (e.g. determinant(mcd$cov) expects a matrix)
* fixed a problem in ltsReg when p==1 and Intercept==FALSE - the vectors of
coefficients ans$coefficients and ans$raw.coefficients were of size 2
* error handling added in ltsReg in case of scale=0
2004-09-16 Valentin Todorov
* 0.2-5
* several errors in doc fixed (ltsPlot.Rd, covPlot.Rd, aircraft.Rd)
2004-09-16 Valentin Todorov
* 0.2-4
* added Regression Diagnostic Plots - function ltsPlot()
* ...added Normal QQ Plot of the residuals
* ...added Standardized Residuals versus index plot
* ...added Standardized Residuals versus fitted values plot
* ...added Regression diagnostic plot
* ltsReg: the responce variable Y added to the result object
* covMcd: fixed a bug related to nsamp -> it was hard-coded = 500 in Fortran
* covMcd: fixed a bug - in case of alpha=1
* added S3 methods plot.mcd and plot.lts
* the S3 methods print.mcd and print.lts moved to separate R files
* added the stars data set (Hertzsprung-Russell diagram)
2004-07-13 Valentin Todorov
* 0.2-3
* improved documentation of the datasets
* added datasets aircraft and delivery
* added Covariance Plots - function covPlot()
* ...added Distance Plot - function distplot()
* ...added Distance-Distance Plot - function ddplot()
* ...added Chisquare QQ-Plot - function chi2qqplot()
* ...added Tolerance Ellipse Plot - function tolellipse()
* added function print.lts (for ltsReg result); included in the test tlts.R
* added function print.mcd (for covMcd result); included in the test tmcd.R
2004-06-26 Valentin Todorov
* 0.2-2
* fixed bug in Fortran: rfltsreg.f, xrfnbreak()
* Depends >= 1.8 (it was >= 1.9, because of the bug above)
* Parameter, controlling whether to perform intercept adjustment at each step added to ltsReg and its default value set to FALSE