Debian Codegraph @

NEWS

raw
1 CHANGES IN VGAM VERSION 1.1-14 (2025-12)
2
3 NEW FEATURES
4
5 o New family functions: loglinb4(), cardioid2(dpqr).
6 Thanks to Arthur Pewsey for his cardioid preprint.
7 o loglinb3() now supports u123 which used to be 0
8 by definition. Yunhao (Harry) Han added this feature.
9 o cops() improved: 'beta.range' is better and
10 'subset' added.
11 o cops3.glm(), copsd3.glm() and wsdm3.glm() written for
12 class "glm" objects. And copsd() is for "vglm" objects.
13 o Arguments 'muxypw = FALSE, roundmux = FALSE'
14 added to depvar.vlm(). The two appear in @infos
15 in some family functions, along with 'Mux4vglmnet'.
16 o Argument 'bhhh' added to vglm.control()
17 and vglm.fit(), and also to some family functions.
18 Argument 'b3h.wz' might accompany this.
19 Function b3hfun() is new.
20 o Argument 'sumcon' added to multinomial() and
21 multilogitlink().
22 o For Tingting Zhan: export S3 method model.frame
23 for "vlm", and ditto to summaryvglm().
24 o acat.deriv() is exported.
25 o trplot(asp = NULL) added for objects of class "qrrvglm".
26 o New data set: eyed.
27 o Functions CM.*() have an 'Intercept' argument.
28 o Tested okay on R 4.5.2.
29
30
31 BUG FIXES and CHANGES
32
33 o No longer exported: cloglog, cauchit, extlogit,
34 fisherz, logc, loge, logneg, logit, logoff, negreciprocal,
35 probit, reciprocal, rhobit, golf, polf, nbolf, nbolf2, Cut.
36 o Renamed data sets: flourbeetle changed to fbeetle.
37 o summaryvglm(HDEtest = FALSE) is the new default,
38 mainly because it may be slow for some models.
39 o gaitdnbinomial(zero = 2) did not work, and
40 neither was 'size' intercept-only by default.
41 o cardioid(dpqr) improved markedly.
42 o constraints(fit, type = "term", matrix = TRUE)
43 had a bug.
44 o seq2binomial()@infos was faulty.
45 o simple.exponential() and better exponential() updated
46 and tested okay.
47 o Better initial values for genpoisson1().
48 o stop() replaces warning() in multinomial() if
49 eta or the fitted values contains NAs. Thanks to
50 Fabrice Rossi for picking this up.
51 o dposnegbin2() added (same as dposnegbin() in VGAMdata).
52 for posnegbinomial(). Thanks to Karl-Joan Alesma
53 for first picking up this bug.
54 o poissonff(imethod = 2) is used if 'weights' is
55 inputted; then weighted.mean() is called.
56
57
58
59
60 CHANGES IN VGAM VERSION 1.1-13 (2025-02)
61
62 NEW FEATURES
63
64 o New family functions: betabinomial.rho(),
65 binom3.or(dr).
66 o Improved functions: pospoisson() (Thanks to Spencer Graves).
67 o Helper function: niters().
68 o VGAMfamilyfunction(link = Link) should work, where
69 Link <- "identitylink", say.
70 Note: weibullR(lshape = logofflink(offset = -2))
71 will no longer work. Instead, use
72 weibullR(lshape = "logofflink(offset = -2)")
73 And something like
74 bvalue <- 0.1
75 vglm(interspike ~ 1, garma(loglink(bvalue = bvalue)))
76 will fail. Instead, use
77 eval(substitute(vglm(interspike ~ 1,
78 garma(paste0("loglink(bvalue = ", .bvalue , ")"),
79 p = 2, coefstart = c(4, 0.3, 0.4))),
80 list( .bvalue = bvalue)))
81 Sorry about this!
82 o linkfun() has a new 'by.var = TRUE' argument.
83 o show.summary.rrvglm() has ... added.
84 o The WSDM function.
85 o Tested okay on R 4.4.1.
86
87
88 BUG FIXES and CHANGES
89
90 o summary(vglm(binom2.or())) failed.
91 o summary(vgamObject) no longer prints the
92 Pearson residuals at all.
93 o hdeffsev() renamed to hdeffsev0(),
94 hdeffsev2() to hdeffsev2() [no change],
95 and hdeffsev() is new and based on WSDM.
96
97
98
99
100 CHANGES IN VGAM VERSION 1.1-12
101
102 NEW FEATURES
103
104 o hdeffsev2() and DDfun() are new.
105 o Some link functions handle up to deriv = 9: e.g.,
106 loglink(), logitlink().
107 o Tested okay on R 4.4.1.
108
109
110 BUG FIXES and CHANGES
111
112 o Passes stricter CHECKing, esp. with Rd
113 \link{} targets that were missing package anchors.
114 And R_Calloc() and R_Free() used.
115 o More \dontrun{} added to .Rd files, to quicken
116 the checking.
117 o Several functions associated with the Laplace
118 distribution moved to VGAMdata.
119 o VGAM now depends on R >= 4.1.0 which was released
120 3 years ago.
121
122
123
124
125 CHANGES IN VGAM VERSION 1.1-11
126
127 NEW FEATURES
128
129 o AIC() and BIC() methods function for "drrvglm" objects.
130 o simulate() works with multinomial(), cumulative(),
131 and other categorical family functions.
132 o New family functions: N1binomial(dr), N1poisson(dr).
133 o Helper functions: is.Identity(), unmaskA(), is.non0().
134 o Tested okay on R 4.4.0.
135
136
137 BUG FIXES and CHANGES
138
139 o "rrvglm" and "drrvglm" objects now fitted under
140 a common framework and algorithm.
141 o summary(vglmObject) no longer prints the
142 Pearson residuals by default. Some other output may
143 be suppressed if they are printed.
144 o Some argument shuffling in binormalcop().
145 Argument 'stdze' means the mean and sd parameters
146 may be estimated.
147 o Some argument shuffling in rrvglm.control(). Also,
148 the following arguments are no longer supported:
149 'Wmat', 'Norrr'", 'Svd.arg', 'Uncorrelated.latvar',
150 'Alpha', 'scaleA'.
151 Also, 'Crow1positive = NULL' for no effect.
152 o summary.rrvglm(h.step = 0.005) and
153 summary.drrvglm(h.step = 0.005) are now defaults.
154 o More checking of links performed if vfl = TRUE.
155 o gaitdnbinomial() now assigns earg[[2]] (needed
156 for rootogram4()).
157 o Coef(drrvglmObject)@A and summary(drrvglmObject) now agree.
158 o The handling of vglm(na.action) is more similar to lm() and
159 glm(). Ditto for vgam().
160
161
162
163
164 CHANGES IN VGAM VERSION 1.1-10
165
166 NEW FEATURES
167
168 o Doubly-constrained RRVGLMs are partially implemented.
169 They have class "drrvglm". More checking
170 is needed to iron out all problems.
171 o plotvgam(..., main = letters[1:3]) will work by
172 having 3 different 'main's.
173 o chisq() has 'squared' for the chi distribution.
174 o New family functions: extbetabinomial(dpqr),
175 gensh(dpqr).
176 o New functions: asinlink(), sqrtlink(), alogitlink(),
177 sloglink(), lcalogitlink(), lcsloglink(), cloglink(),
178 log1plink().
179 o Argument 'ynames = FALSE' added to multinomial(),
180 acat(), [cs]ratio(), propodds() and cumulative(),
181 which can make the output more readable. Thanks to
182 Daniel Smith for the suggestion.
183 o New functions: CM.symm[01](), CM.equid(), CM.free(),
184 CM.ones(), CM.qnorm().
185 o negbinomial(vfl = FALSE) has been added.
186 o Tested okay on R 4.3.2.
187
188
189 BUG FIXES and CHANGES
190
191 o acat(), cumulative(), cratio(), propodds(), sratio() handles
192 argument 'Thresh' which can be assigned 'symm1', 'symm0',
193 'free', 'qnorm', etc.
194 o All remnants of is.R() removed, for R 4.4.0.
195 o Some fortran code now avoids some warnings found by
196 a Debian C compiler.
197 o dgaitdplot() stop()s when the parameters are
198 outside the parameter space.
199 o demo() examples have been updated to run fully.
200 o "rrvglm" objects have new slots "A.est" and "C.est"
201 which simplifies the code.
202 o "rrvglm" objects are estimated using the "alternating"
203 'Algorithm' only, and the "derivative" method withdrawn.
204 o For RR-VGLMs and DRR-VGLMs, 'kill.all' changed to 'omit123'.
205 o rrvglm.control() has some of parameters changed
206 (for more stringent convergence).
207 o Some argument shuffling: genpoisson[12](). That is,
208 the order has been changed.
209 o summary.rrvglm() swapping of pasting block matrices
210 according to 'fixA' was wrong.
211 o replace.constraints() renamed to replaceCMs().
212 o dtruncpareto1() no longer returns 0 at lower and upper.
213 Thanks to Jason Li for picking this up.
214 o summaryvgam() and summarypvgam() calls
215 summaryvglm(HDEtest = FALSE) to
216 simplify things and avoid unnecessary complications.
217 o expected.betabin.ab renamed to Ebbin.ab.
218 o Deprecated: gordlink(), pordlink(),
219 nbordlink(), nbord2link().
220
221
222
223
224 CHANGES IN VGAM VERSION 1.1-9
225
226 NEW FEATURES
227
228 o Soon: VGAM will require R >= 4.4.0 because plot.profile
229 will move from \pkg{MASS} to \pkg{stats}. Sorry!
230 o Yettodo: modify dgaitdlog() to be like
231 dgaitdpois() in terms of safeguarding against too
232 much deflation and inflation (to return NaN).
233 Ditto to dgaitdzeta() and dgaitdnbinom().
234 Ditto for the p-type functions.
235 o New family functions: gammaff.mm(), hurea(d).
236 o New generic and/or methods functions: cops().
237 o acat(), cumulative(), cratio(), propodds(), sratio() handles
238 argument 'thresholds' equalling 'symmetric1' and 'symmetric0'.
239 Note from version 1.1-8 that 'equidistant' was added.
240 o Tested okay on R 4.3.0.
241
242
243 BUG FIXES and CHANGES
244
245 o wald.stat(..., orig.SE = TRUE, values0 = values0)
246 assumed that values0 == 0.
247 o [rq]zinegbin() did not handle the case 'munb = NULL'
248 when 'prob' was specified. Thanks to Giuseppe D for
249 finding this bug.
250 o acat(), cumulative(), cratio(), propodds(), sratio()
251 must have an intercept term.
252 o For vglm.fit(), slot "middle" is now "middle1",
253 "fini" is now "fini1", and a new "start1" slot added.
254 o bigamma.mckay() moved into VGAMdata 1.1-9.
255 o nakagami() has a zero = "shape" argument.
256 o [dp]gaitdpois() have more range checking, e.g., for too
257 much deflation and inflation.
258 o For more speed, several slow-checking .Rd files
259 have been changed by using \dontrun{}.
260 o summary(posbernoulli.tb()) failed to print the confidence
261 interval. Thanks to Carl Schwarz for picking this up.
262 o DFLOAT() changed to DBLE() in some .f files.
263
264
265
266
267 CHANGES IN VGAM VERSION 1.1-8
268
269 NEW FEATURES
270
271 o Warnings from R-devel addressed.
272 o acat(), cumulative(), cratio(), sratio() have arguments
273 'thresholds', 'Tref', 'Treverse'.
274 o hdeffsev() has been edited.
275 o margeff() works for poissonff(), negbinomial() and
276 posnegbinomial() objects.
277 o multilogit() did not handle 1-row arguments correctly when
278 'inverse = FALSE'. Now vglm(y, multinomial) handles a 1-row y,
279 as well as acat and cumulative and [cs]ratio.
280 o rrvglm.fit() is more robust with respect to 'one.more'.
281 o plot.vgam() issues an error if sm.os()/sm.ps() is used with
282 deriv.arg > 0.
283 o New data set: budworm.
284
285
286
287 BUG FIXES and CHANGES
288
289 o rainbow.sky has a new green.
290 o zoabetaR() handles the argument 'constraints'. Thanks
291 to Riccardo Lo Martire for helping to find the bug.
292
293
294
295
296 CHANGES IN VGAM VERSION 1.1-7
297
298 NEW FEATURES
299
300 o Slot 'rqresslot' added to VGAM family functions, for implementing
301 randomized quantile residuals.
302 Available in about 50 family functions such as poissonff() currently.
303 o Slot 'rqresslot' added to VGAM family functions, for implementing
304 o Argument 'label.it' added to model.matrixvlm().
305 o New generic and/or methods functions: rqresid(), rqresiduals().
306 o New data set: lpossums.
307 o Tested okay on R 4.2.1.
308
309
310 BUG FIXES and CHANGES
311
312
313 o cumulative() has been replaced by its version prior
314 to 2021-05. Evidently some so-called improvements
315 created some bugs. Thanks to Johanna Neslehova for
316 alerting me to this.
317 o bistudentt() was documented wrong; thanks to
318 Iago Gine Vazquez for finding this.
319 o fill() now deprecated; replaced by fill4().
320
321
322
323
324 CHANGES IN VGAM VERSION 1.1-6
325
326 NEW FEATURES
327
328 o Most of the GAIT regression code have new arguments 'd.mlm'
329 and 'd.mix' but mainly only the dpqr-type functions work!
330 Some other functions should work such as gaitdpoisson() and
331 dgaitplot(). Also, is.deflated() and goffset() are new.
332 o New family function: gaitdnbinomial().
333 o Upgraded family functions: gaitpoisson() becomes gaitdpoisson(),
334 gaitlog() becomes gaitdlog(), gaitzeta() becomes gaitdzeta().
335 o New generic and/or methods functions: KLD(), Pheapseep().
336 o Some colours are pre-defined for GAITD regression, and the default
337 colour scheme has changed.
338 o trim.constraints() has a new 'intercepts = TRUE' argument.
339 o Many family functions have a 'dpqrfun' component in the @infos slot.
340 Useful to tie in the modelling function with its density, etc.
341 o binormal() has a new 'rho.arg' argument.
342 o Some link functions handle deriv = 4: e.g., loglink(), logitlink().
343 o vglmfit@misc$history gives the convergence history.
344 o Improvements to clogloglink() at the boundary 1.
345 o hdeff.vglm() has a new argument 'theta0'.
346 o wald.stat() cleaned up internally.
347 o Tested okay on R 4.1.2.
348
349
350
351 BUG FIXES and CHANGES
352
353 o All GAIT regression code has arguments 'alt.mlm' changed to
354 'a.mlm', 'inf.mix' changed to 'i.mix', etc. Colours have changed.
355 [dpqr]gaitpois() have been renamed to [dpqr]gaitdpois(),
356 and gaitpoisson() to gaitdpoisson().
357 o [dpqr]zanegbin() no longer supports argument 'prob'.
358 o hdeff() makes a recursive call if any NAs are in the answer and
359 fd.only == FALSE. So the answer should then be a 'proper' one.
360 o Code involving anova() for VGLMs/VGAMs changed to avoid an
361 error when loading in \pkg{semTools}.
362 o spikeplot(..., use.table = TRUE) failed wrt x.use and y.use.
363 o vcov() slightly faster for "vglm" objects. Thanks to Frank Harrell
364 for comments that led to this.
365 o linkfun.vglm() changed to linkfunvglm().
366
367
368
369
370 CHANGES IN VGAM VERSION 1.1-5
371
372 NEW FEATURES
373
374 o New functions: meangait(), spikeplot(), trim.constraints().
375 o iam(i, 1:M, M = M) now works, for selecting a whole row,
376 and similarly for a column.
377 o Tested okay on R 4.0.3.
378
379
380
381 BUG FIXES and CHANGES
382
383 o gait[log,poisson,zeta](): arguments llambda.[ai] and lshape.[ai]
384 equal llambda.p and lshape.p by default. This means only one
385 argument needs be assigned in order to change them all.
386 o [dpqr]genpois[12]() has argument 'mean' changed to 'meanpar'.
387 o gait[log,zeta]() bug fixed wrt missing 'allprobs'.
388 o qfoldnorm() now uses the bisection method. Thanks to
389 Mauricio Romero for a suggestion that led to this.
390 o gait[log,poisson,zeta]() compute the sandwich formula
391 explicitly for the working weights; this has speed gains.
392 Their @deriv avoids nested for() loops too.
393 Their initial value grids have become coarser.
394 o Functions moved to VGAMdata 1.1-5: yip88, bellff(dr), tikuv(dpqr).
395
396
397
398
399 CHANGES IN VGAM VERSION 1.1-4
400
401 NEW FEATURES
402
403 o Functions score.stat() and wald.stat() offer greater flexibility
404 by some new arguments, and 'as.summary' changed to 'orig.SE'.
405 o Influence() computes the influence functions of a "vglm" object.
406 o Methods functions hdeff.matrix() and hdeff.numeric() are new.
407 o Family function slot genpoisson()@ weight was updated with
408 the EIM derived by Easton Huch, and after major changes,
409 genpoisson() was renamed to genpoisson0().
410 o New functions: [dpqr]gaitnbinom(), [dpqr]gaitgenpois[012](),
411 [dpqr]gaitbinom(), dgaitplot(), plotdgait.vglm(), Trunc().
412 o New family functions: genpoisson[12](dpqr), gaitpoisson()
413 so that gaitpoisson.mix() and gaitpoisson.mlm() were removed,
414 gaitlog() so that gaitlog.mix() was removed,
415 gaitzeta() so that gaitzeta.mix() was removed.
416 o Tested okay on R 4.0.3.
417
418
419
420 BUG FIXES and CHANGES
421
422 o Deprecated: dgenpois(), genpoisson(), [dpqr]gaitnbinom.m[ix,lm]().
423 gatnbinomial.m[ix,lm](dpqr), [dpqr]gaitbinom.mlm(),
424 gaitpoisson.mix(), gaitpoisson.mlm(), gabinomial.mlm(),
425 gtbinomial().
426 o 'lhs.prob' changed to 'cdf.max.s'.
427 o Argument changes for all GAIT code, e.g., 'alter' changed to
428 'alt.mix' or 'alt.mlm', 'byrow.arg' to 'byrow.ai', etc.
429 o Functions moved to VGAMdata 1.1-4: oalog, oilog, otlog,
430 oapospois, oipospois, otpospois, oazeta, oizeta, otzeta, oizipf,
431 oiposbinomial(dpqr), [dpqr]posbinom, [dpqr]pospois, [dpqr]posnegbin.
432 o Pearson residuals no longer printed by default in a summary
433 of a "vglm" object: presid = FALSE.
434 o Some changes made to plot(vgam()) code for changes in R 4.0.0,
435 especially if the covariate is a character/factor.
436 Now avoids "coercing argument of type 'double' to logical" warning.
437 o negbinomial(lmu = "nbcanlink") had an error relating to "newemu".
438
439
440
441 CHANGES IN VGAM VERSION 1.1-3
442
443 NEW FEATURES
444
445 o New functions: [dpqr]gaitpois(), [dpqr]gaitlog(), [dpqr]gaitzeta(),
446 dextlogF().
447 o New family functions: gatnbinomial.mix(dpqr),
448 gaitlog.mix(dpqr), gaitpoisson.mix(dpqr), gaitpoisson.mlm(dpqr),
449 extlogF1(d).
450 o New generic and/or methods functions: altered(),
451 inflated(), truncated(), specialsvglm(),
452 fix.crossing(), is.crossing().
453 o Coef.vlm() handles models with contiguous "multilogitlink"s,
454 e.g., gaitpoisson.mlm().
455 o get.offset() is now defined as a generic function (returns the
456 n x M matrix of offsets by default).
457 o Some .f files have been pruned of unnecessary code.
458 o New data set: V2.
459 o Tested okay on R 4.0.0.
460
461
462
463 BUG FIXES and CHANGES
464
465 o Another attempt to .Deprecated() all link function names
466 which do not end in "link", e.g., loglink() replaces loge(),
467 logitlink() replaces logit().
468 o gatpoisson.mix() (now merged into gaitpoisson.mix())
469 reorders its linear/additive predictors and now handles
470 length(alter) == 0 (default, so equivalent to poissonff())
471 and length(alter) == 1.
472 Ditto for gatnbinomial.mlm(), etc.
473 o Three argument mismatches found in .f and .c code.
474 o qlog() and qyules() buggy wrt interval selection prior to
475 applying bisection.
476 o 'Rank mismatches' flagged by gfortran10 fixed, by not
477 calling two subroutines.
478 o [dpqr]gaitpois.mlm() and [dpqr]gaitpois.mix() have been combined into
479 [dpqr]gaitpois().
480
481
482
483 CHANGES IN VGAM VERSION 1.1-2
484
485 NEW FEATURES
486
487 o New generic and/or methods functions: add1.vglm(),
488 dfterms(), drop1.vglm(), extractAIC.vglm(),
489 rootogram4vglm() [based on \pkg{countreg}],
490 step4vglm().
491 o New family functions: gabinomial.mlm(), gatnbinomial.mlm(dpqr),
492 gatpoisson.mix(dpqr), gatpoisson.mlm(dpqr).
493 o New functions: [dpqr]zipfmb(),
494 [dpqr]gaitbinom.mlm(), [dpqr]gaitnbinom.mlm(),
495 [dpqr]gaitpois.mlm(), [dpqr]gaitnbinom.mix(),
496 [dpqr]gaitpois.mix().
497 Note that the naming conventions have changed somewhat, especially
498 regarding the negative binomial distribution.
499 o Decommissioned functions:
500 gapoisson(), gtpoisson(),
501 [dpqr]gabinom(), [dpqr]gapois(),
502 [dpqr]gibinom(), [dpqr]gipois(),
503 [dpqr]gtbinom(), [dpqr]gtpois().
504 o eta2theta() should be able to handle links for, e.g.,
505 gatpoisson.mlm(), gibinomial().
506 o posbernoulli.t() has a 'type.fitted' argument.
507 o gtpoisson.mlm() supports type.fitted = "Pobs.a",
508 gipoisson.mlm() supports type.fitted = "Pstr.i".
509 o New data set: backPain2.
510 o Cleve Moler added to DESCRIPTION as a contributor, to reflect
511 the use of LINPACK.
512 o Tested okay on R 3.6.1.
513
514
515
516 BUG FIXES and CHANGES
517
518 o Use of old link function names now result in a call to
519 .Deprecated(), except for a selected few such as logit() and
520 probitlink()---this will only be allowed for a limited time.
521 o negbinomial()@initialize bug fixed for "nbcanlink()" and
522 multiple responses; thanks to Victor Miranda.
523 o cens.poisson() gave a warning upon a summary().
524 Thanks to Jens Heumann for picking this up.
525 o bisection.basic() returns a warning, not an error, if
526 it cannot find an interval which covers the root.
527 o Rank() has an improved implementation and documentation.
528 o anova.vglm() had some bugs---thanks to Thamron Keowmani for
529 finding them.
530 o posnormal() failed when 'constraints' was inputted---thanks
531 to Michael Zamo for picking this up.
532 o Improvements to yules(dq).
533
534
535
536 CHANGES IN VGAM VERSION 1.1-1
537
538 NEW FEATURES
539
540 o New family functions: gentbinomial(dpr), gentpoisson(dpr),
541 genapoisson(dpr), genipoisson(dpr).
542 o New functions: [dpr]genabinom(), [dpr]genibinom(), care.exp2().
543 o Argument 'parallel' added to benini1(), gamma1(), maxwell(),
544 rayleigh() and topple().
545 o Argument 'parallel = TRUE' for exponential() does not apply to the
546 intercept now.
547 o New link function: loglogloglink().
548 o New variable: rainbow.sky.
549 o Some family functions now support 'type.fitted = "Qlinks"'
550 (quantile link functions in \pkg{VGAMextra}),
551 e.g., benini1(), gamma1(), exponential(), maxwell(), rayleigh(),
552 topple().
553 o Tested okay on R 3.5.3.
554
555
556
557 BUG FIXES and CHANGES
558
559 o All link function names now end in "link", e.g.,
560 loglink() replaces loge(), logitlink() replaces logit().
561 Backward compatability holds for older link function names,
562 probably for quite a while.
563 o gaussianff() reinstated but is effectively deprecated;
564 it calls uninormal() instead after a warning.
565 o Documentation for lerch.Rd improved, as well as zeta()
566 (thanks to Jerry Lewis).
567 o rrar.control(summary.HDEtest = FALSE) is now set.
568 o anova.vglm() gives an error message if it operates on a model
569 with just a single factor as explanatory, when type = 2 or 3.
570 o "summary.vglm" and "summary.vgam" prints out the names of
571 the linear/additive predictors by breaking the output into
572 lines that are not too long.
573 Number of "Fisher scoring" iterations printed.
574 o posbernoulli.tb(ridge.constant = 0.0001) now, it was
575 ridge.constant = 0.01 for a long time. The adjustment is now
576 multiplicative rather than additive.
577
578
579
580 CHANGES IN VGAM VERSION 1.0-6
581
582 NEW FEATURES
583
584 o New functions: [dr]bell(), bellff(), [dr]trinorm(), trinormal(),
585 o New functions: anova.vglm(), R2latvar(), ordsup().
586 And residualsvglm(..., type = "stdres") is implemented for
587 'VGAMcategorical' models and "poissonff" and "binomialff" models.
588 Thanks to Alan Agresti suggesting all these.
589 anova.vglm() works for 'type = "II"' (default) and "I" and "III".
590 o calibrate() has argument 'lr.confint' for likelihood ratio based
591 confidence intervals (for Rank == 1 "rrvglm" and "qrrvglm" objects
592 only). Wald vcov matrix returned in vcov list component for
593 poisson and binomial.
594 calibrate() has argument 'cf.confint' for characteristic function
595 based confidence intervals (for Rank == 1 "rrvglm" and "qrrvglm"
596 objects only).
597 o margeff() works for tobit() objects.
598 o A .control() function for a VGAM family function allows the 'HDEtest'
599 argument in summaryvglm() to be FALSE by default, e.g.,
600 zanegbinomial.control(summary.HDEtest = FALSE) means that
601 summary(zanegbinomialfit) does not conduct any HDE tests.
602 o vcov() for "qrrvglm" objects (but assumes C is given).
603 o New functions: attr.assign.x.vglm().
604 o New documentation: residualsvglm().
605 o Tested okay on R 3.5.1.
606
607
608
609 BUG FIXES and CHANGES
610
611 o By default, dbetabinom() now behaves like dbinom() wrt 'size' and
612 'prob', since rho = 0. This is done by 'limit.prob'.
613 Thanks to S. A. Mousavi for picking this up.
614 The functions [dr]betabinom() have also been modified accordingly to
615 behave like the binomial distribution, by default.
616 o pbetabinom.ab() did not handle negative values of 'q' correctly;
617 NAs now replaced by 0s.
618 o vglm()@predictors did not contain the offsets; now they do.
619 Hence predict() will be affected.
620 o pzeta() was incorrect if the first argument was greater than c.12.
621 Thanks to Mitchell Newberry at the University of Pennsylvania
622 for picking this up.
623 o profilevglm() did not call constraints() correctly.
624 o Deprecated: quasibinomialff(), quasipoisonff(), gaussianff().
625 The models fitted now have a log-likelihood that fully specifies
626 the model.
627 o nbcanlink() has a cosmetically improved label.
628 o lambertW() documentation improved; thanks to Jan Somorcik for this.
629 o dirmultinomial()@validparams had M unassigned [thanks to Feng Zeng].
630 o For model.matrix() for "qrrvglm" objects, model.matrix(type = "lm")
631 is new and returns what used to be model.matrix(type = "vlm").
632 And now model.matrix(type = "vlm") returns the big model matrix
633 of the CQO given the constrained coefficients C.
634 o For uninormal(), if var.arg = TRUE then zero = "var", i.e., then
635 "sd" remains intercept-only by default even if var.arg = TRUE.
636
637
638
639 CHANGES IN VGAM VERSION 1.0-5
640
641 NEW FEATURES
642
643 o extlogit() now handles 'deriv = 3'.
644 o Generic function hdeff() implements an analytic solution
645 for the following families: borel.tanner(), felix(), lindley().
646 For almost all other families, finite-difference approximations
647 to derivatives means that first and second derivatives can be
648 computed, even with models with 'xij' terms.
649 o Generic function wald.stat() implements Wald tests with SEs evaluated
650 at the null values, not at the original MLE, so do not suffer from the
651 Hauck-Donner effect.
652 o "vglm" objects have a new "charfun" slot, for the
653 characteristic function.
654 o "summary.vglm" and "summary.vlm" objects have new
655 "coef4lrt0", "coef4score0", "coef4wald0" slots, for
656 storing the 'Wald table' equivalent of LRTs, score tests and
657 modified Wald tests.
658 The latter has its SEs computed at the null values with the
659 other coefficients obtained by further IRLS iterations, etc.
660 Function summaryvglm() has arguments 'lrt0.arg', 'score0.arg',
661 'wald0.arg'.
662 o TIC() is new, for the Takeuchi's Information Criterion.
663 Thanks to Khedhaouiria Dikra for suggesting this.
664 o mills.ratio() and mills.ratio2() are exported.
665 o New functions: lrt.stat(), score.stat(), wald.stat(),
666 which.etas(), which.xij().
667 o cauchy1() and cauchy() handle multiple responses and have
668 been modernized a bit.
669 o Tested okay on R 3.4.3.
670
671
672
673 BUG FIXES and CHANGES
674
675 o Setting 'deriv.arg' a positive value in plotvgam() when there
676 are no s() terms results in a warning.
677 Thanks to Barry Goodwin for detecting this.
678 o cens.poisson() can better handle large lambda values, at least for
679 left and right censored data (but not for interval-censored data yet).
680 Thanks to Eugenie Hunsicker for picking up deficiencies in the code.
681 o In multinomial.Rd, it was stated that setting parallel = TRUE
682 did not make the intercepts the same. It does make them the same.
683 Thanks to Stuart Coles for picking this up.
684 o binomialff(multiple.responses = TRUE) returned an incorrect deviance.
685 o bilogistic() uses SFS rather than BFGS as its algorithm.
686 o Deprecated: lrp(), normal1() [use uninormal() instead].
687
688
689
690 CHANGES IN VGAM VERSION 1.0-4
691
692 NEW FEATURES
693
694 o This package is now Renjin compatible.
695 o gengamma.stacy() handles multiple responses, and some
696 of its arguments have had their default values changed.
697 o calibrate() has a methods function for "rrvglm" objects.
698 o "vglm" objects have a new "hadof" slot.
699 o "vglm" objects will soon have a new "charfun" slot.
700 o constraints(vglmfit, matrix = TRUE) is now embellished by
701 rownames, by default.
702 o New generic and methods function: hdeff(), lrp().
703 o Many link functions accommodate 'deriv = 3' and sometimes a bit
704 higher. More safety when 'deriv' is assigned out-of-range values.
705 o fitted() and predict() with lms.???() fits, such as lms.bcn(),
706 now handles inputting new 'percentiles' values. Thanks to
707 Sven Garbade for picking up this obvious deficiency.
708 o New function(s): d3theta.deta3().
709 o Tested okay on R 3.4.1. This package now requires R 3.4.0 or higher
710 (not R 3.1.0 as before).
711
712
713
714 BUG FIXES and CHANGES
715
716 o [dqpr]gengamma.stacy() have no default value for 'd' and 'k' args.
717 o negbinomial(lmu = "nbcanlink") now works properly;
718 thanks to Victor Miranda for much help and picking up 3 bugs.
719 o perspqrrvglm() called @linkinv() without passing in @extra.
720 o fisherz(theta, inverse = TRUE, deriv = 1) was out by a factor of 2.
721 o multilogit() did not handle 1-row arguments correctly when
722 'inverse = TRUE' and 'refLevel' was neither 1 or M+1.
723 o Several C and Fortran functions renamed, for R 3.4.0.
724
725
726
727 CHANGES IN VGAM VERSION 1.0-3
728
729 NEW FEATURES
730
731 o vgam() with sm.os() and sm.ps() terms allows G2-VGAMs to be fitted.
732 o plotvgam() has a "shade" argument.
733 o Almost all family functions have been "validparams"-enabled, for
734 greater reliability.
735 o confint() implements the profile likelihood method (in addition
736 to the Wald method).
737 o New family functions: diffzeta(dpqr),
738 oilog(dpqr), oiposbinomial(dpqr), oizeta(dpqr), oizipf(dpqr),
739 otlog(dpqr), otpospoisson(dpqr), otzeta(dpqr),
740 oalog(dpqr), oapospoisson(dpqr), oazeta(dpqr), topple(dpqr).
741 o New functions: [pqr]zeta(), [qr]zipf().
742 o Argument 'zero' now accepts "" or NA and interprets these as NULL,
743 i.e., no linear or additive predictors are intercept-only.
744 o Significance stars added to summary(rrvglm.object), for very crude
745 inference.
746 o zeta() can return the Hurwitz zeta function, via the 'shift' argument.
747 o show.summary.vglm() will only print out any dispersion parameter
748 that is not equal to 1.
749 o type.fitted = "quantiles" is available for
750 gevff(), negbinomial(), poissonff().
751 o Tested okay on R 3.3.2.
752
753
754
755 BUG FIXES and CHANGES
756
757 o mills.ratio1() in tobit() did not handle very negative 'x' values
758 correctly. Thanks to Christoph Nolte for detecting this.
759 o Renamed arguments: zetaff(d) use 'shape', not 'p'.
760 o betabinomialff()@infos was buggy wrt 'lshape1' and 'lshape2'.
761 Thanks to Xiaodong for detecting this.
762 o leipnik() uses logoff(offset = -1) as the default link for
763 lambda now, not "loge".
764 o logff(dpqr) uses 'shape' instead of 'c' as the parameter name.
765 o yules(dpqr) uses 'shape' instead of 'rho' as the parameter name.
766 o hzeta(dpqr) uses 'shape' instead of 'alpha' as the parameter name.
767 o felix(dpqr) uses 'rate' instead of 'a' as the parameter name.
768 o dbetabinom.ab() handles large values of shape1 and shape2 better, via
769 the dbinom() limit. Thanks to Micha Schneider for picking up the bug.
770 o [dpqr]posnegbin() have been improved a little.
771 o logLik(summation = TRUE): prior weights 'w' have been
772 converted to a vector when passed in; this is likely to make it less
773 likely to give an error.
774 o Labelling of the colnames of the fitted values have changed for many
775 family functions, including those for multiple responses,
776 e.g., gevff(). Also "mean" had a bug or two in gevff()@linkinv.
777
778
779
780 CHANGES IN VGAM VERSION 1.0-2
781
782 NEW FEATURES
783
784 o vglm.fit() has been simplified and handles half-stepping better.
785 o AR1() implements the EIM of Porat and Friedlander (1986); this is
786 the work of Victor Miranda. It is specified by type.EIM = "exact"
787 (the default).
788 o Function gevff() replaces egev(). It handles multiple responses
789 like any other ordinary VGAM family function.
790 o A rudimentrary plotvglm() plots the Pearson residuals, firstly
791 versus the predicted values, and secondly, against the hat values.
792 o The 'refLevel' argument of multinomial() accepts a character string,
793 e.g., multinomial(refLevel = "European") for xs.nz$ethnicity as a
794 response.
795 o New family function: oipospoisson(dpqr), zoabetaR().
796 o New functions: grid.search[23](), [dpqr]oiposbinom().
797 o is.buggy() is called by vgam() immediately after estimation;
798 it gives a warning if any constraint matrix corresponding
799 to an s() term is not orthogonal.
800
801
802
803 BUG FIXES and CHANGES
804
805 o vglm.fit() did not handle half-stepping very well.
806 o Some families for counts (i.e., [pos,z[ai]]negbinomial[ff]())
807 have been "validparams"-enabled in order to make estimation
808 near the boundary of the parameter space more stable,
809 especially when a Poisson approximation is suitable.
810 o Other families that have been "validparams"-enabled:
811 gev(), gpd().
812 o Actuarial or statistical size distributions families have
813 been modified with respect to initial values, e.g.,
814 sinmad, dagum, [inv.]lomax, [inv.]paralogistic, [gen]betaII().
815 o rep_len() replaces rep() where possible.
816 o Function gev() has been changed internally and
817 arguments such as 'gshape' have changed.
818 o Function rzipois() may not have handled 0-deflation properly
819 but it does so now.
820 o Function plotvgam() had a bug testing for variable names when
821 the xij facility was used.
822 o multinomial() and multilogit() use "(Last)" to signify
823 the last level of a factor; it used to be "last".
824 o qposbinom() returned 0 (incorrect), and not 1 (correct), for p = 0.
825 o zipoisson() and zipoissonff() no longer store fitted values such
826 as pstr0 in the misc slot. They can be obtained by, e.g.,
827 fitted(fit, type.fitted = "pstr0").
828 o Renamed functions:
829 egumbel() is now called gumbelff().
830 [dqpr]ozibeta() is now called [dqpr]zoabeta().
831 o Renamed parameter names: zetaff() uses 'shape', not 'p'.
832 o qzibinom() did not handle arguments lower.tail and log.p correctly.
833 o Tested okay on R 3.3.0. This package now requires R 3.1.0 or higher
834 (not R 3.0.0 as before).
835
836
837
838 CHANGES IN VGAM VERSION 1.0-1
839
840 NEW FEATURES
841
842 o Argument 'zero' has been programmed to handle (a more
843 intuitive) a character vector. Each value of this
844 vector is fed into grep() with fixed = TRUE. Many
845 VGAM family functions have an equivalent default
846 character value of 'zero'.
847 o New slots: "validparams" and "validfitted" for providing
848 more opportunities for half-stepping.
849 o The "infos" slot of most family functions have a
850 component called "parameters.names", and also
851 "Q1" and "M1".
852 o margeff() works for cratio(), sratio() and
853 acat() models, and is generic (with S4 dispatch).
854 For this, "vcategorical" replaced by "VGAMcategorical",
855 and "VGAMordinal" is also a virtual class.
856 And margeffS4VGAM() is generic.
857 o summaryvglm() calls the generic summaryvglmS4VGAM() in order
858 to compute useful quantities, and it is printed by
859 showsummaryvglmS4VGAM(). Specific examples include
860 the binom2.or() and cumulative() families.
861 o Similarly, show.vglm() calls the generic showvglmS4VGAM() in
862 order to print extra potentially useful output.
863 Ditto for , show.vgam() which calls showvgamS4VGAM().
864 o Similarly, predictvglm() calls the generic predictvglmS4VGAM()
865 in order to allow for family-function-specific prediction.
866 o logitoffsetlink() is new.
867 o [dpqr]ozibeta() and [dpr]ozibetabinom() and
868 [dpr]ozibetabinom.ab() are new;
869 by Xiangjie Xue and Thomas Yee.
870 o coef(..., type = c("linear", "nonlinear")) is available
871 for "vgam" objects.
872 o The following have new 'weights' slots (based on
873 negbinomial()@weight): posnegbinomial(), zanegbinomial[ff](),
874 zinegbinomial[ff](). It is based on the expectation of
875 a difference between 2 trigamma function evaluations being
876 computed using pnbinom(lower.tail = FALSE) and variants.
877 Both functions have some argument defaults tweaked.
878 o log1mexp() and log1pexp(), based on Martin Maechler's 2012 paper,
879 is 'new'.
880 o Many zero-altered and zero-inflated families have additional
881 'type.fitted' choices.
882 Initial values for such families hav been improved
883 (via Init.mu()).
884 o expint(), expexpint(), expint.E1() allow the computation of the
885 first few derivatives.
886 o Tested okay on R 3.2.4.
887
888
889
890 BUG FIXES and CHANGES
891
892 o Order of arguments changed: binom2.rho(lmu, lrho),
893 negbinomial(), posnegbinomial(),
894 zanegbinomial(), zinegbinomial().
895 o pzanegbin() could return -.Machine$double.eps. Thanks to
896 Ryan Thompson for notifying me about this.
897 o pbinorm() used to have a bug wrt Inf and -Inf values in
898 its arguments. Thanks to Xiangjie Xue for picking this up.
899 o plota21() used qchisq(0.95, df = 1) instead of
900 qchisq(0.95, df = 1) / 2 for LRT confidence intervals.
901 Thanks to Russell Millar for picking this up.
902 o A new function Init.mu() is used to initialize several
903 family functions, especially those based on the negative
904 binomial and Poisson distributions.
905 The default for Init.mu() is suitable for 0-inflated
906 data.
907 o The fitted value of polya() was wrong (wasn't the mean).
908 o Default value of argument 'zero' has changed for:
909 bisa(), gumbelII().
910 o zibinomialff()@weight had a bug when calling iam().
911 o [dpqr]nbinom(..., size = Inf) was buggy; it produced many NaNs.
912 Thanks to Martin Maechler for promptly fixing this, for R 3.2.4.
913 o The arguments of interleave.VGAM() have changed: from
914 interleave.VGAM(L, M) to interleave.VGAM(.M, M1, inverse = FALSE).
915 The is a compromise solution with respect to my book.
916 The 'inverse' argument is due to Victor Miranda.
917 o summaryvglm() evidently evaluated the weights slot of an
918 object twice. Now it is only done once.
919
920
921
922 CHANGES IN VGAM VERSION 1.0-0
923
924 NEW FEATURES
925
926 o Official version that goes with the just-released book
927 "Vector Generalized Linear and Additive Models:
928 With an Implementation in R" by T. W. Yee (2015),
929 Springer: New York, USA.
930 o gengamma.stacy() implements a grid search wrt
931 all its parameters.
932 o New functions: [dp]lms.bcn().
933 o New family function: weibull.mean().
934 o triangle.control() slows down the speed of the iterations
935 towards the MLE, because the regularity conditions do not hold.
936 o New arguments: AR1(nodrift = FALSE).
937 o binormal has arguments eq.mean and eq.sd which now operate
938 independently.
939 o confint() should work for objects that are "vglm"s.
940 Thanks to Tingting Zhan for suggesting this.
941 o Tested okay on R 3.2.2.
942 o Methods functions for responseName() and term.names() and
943 has.intercept().
944
945
946 BUG FIXES and CHANGES
947
948 o Link functions have changed a lot!!!
949 They return different values when deriv = 1 and deriv = 2,
950 coupled with inverse = TRUE and inverse = FALSE.
951 Type ?Links to see examples.
952 The first derivatives become reciprocals of each other
953 when inverse = TRUE and inverse = FALSE, however the
954 2nd derivatives are no longer reciprocals of each other.
955 Also affected are dtheta.deta() and d2theta.deta2(), etc.
956 o 'show()' added to importMethods('methods') in NAMESPACE.
957 o The following currently do not work:
958 golf(), nbolf(), polf().
959 o AA.Aa.aa() used the OIM and worked for intercept-only models,
960 but now it uses the EIM.
961 o logneg("a", short = TRUE) has been modified.
962 o posnormal(): the first and second derivatives have been modified
963 for both SFS and ordinary FS,
964 and the default is zero = -2 now ('sd' is intercept-only).
965 Several other improvements have been done.
966 o binomialff()@deviance is assigned all the time now.
967 o dbetabin.ab() better handles extremes in the shape
968 parameters (very close to 0 and larger than 1e6, say).
969 Thanks to Juraj Medzihorsky for picking this up.
970 o Family functions: zigeometric()@weight and zigeometricff()@weight
971 had one element incorrect.
972 o logit("a+b", short = FALSE) was labelled incorrectly, etc.
973 o Family function tobit()@weights implements Fisher scoring
974 entirely. And it handles observations whose fitted
975 values are (relatively) large and positive; thanks to
976 Victor Champonnois for picking up this bug.
977 o S3 methods function df.residual_vlm() also called by
978 df.residual.vlm().
979 This is to avoid a bug picked up by car::linearHypothesis().
980 Ditto for vcovvlm() by vcov.vlm().
981 Also model.matrix() and formula().
982 Thanks to Michael Friendly and John Fox for help here.
983
984
985
986 CHANGES IN VGAM VERSION 0.9-8
987
988 NEW FEATURES
989
990 o Tested okay on R 3.2.0.
991 o is.buggy() tests to see if a fitted VGAM object suffers
992 from known bugs, e.g., a vgam() object with at least one
993 s() term whose constraint matrix does not have orthogonal
994 columns.
995 o New family function: AR1(d).
996 o New function: dgenpois().
997 o The package has been updated to reflect the new
998 J. Stat. Soft. paper by Yee, Stoklosa and Huggins.
999 A vignette based on this paper is now included.
1000 o dgenbetaII() has now been written; and genbetaII() improved,
1001 and about 8 special cases of genbetaII() have
1002 all been modernized to handle multiple responses
1003 and a default grid search over all the parameters
1004 (arguments 'gscale' and 'gshape1.a', etc.). These
1005 families are based on Kleiber and Kotz (2003).
1006
1007
1008 BUG FIXES and CHANGES
1009
1010 o Family function genpoisson() has been modernized, and
1011 should give correct results wrt AIC() etc.
1012 o Argument 'init.alpha' renamed to 'ialpha', for the brat(),
1013 bratt(), and dirmul.old() families.
1014 o Calls to N.hat.posbernoulli() used Hlist = constraints rather
1015 than Hlist = Hlist; this failed for RR-VGLMs.
1016 o Family function tobit() obtains initial values even
1017 in the case when it would otherwise fit an underdetermined
1018 system of equations. Thanks to McClelland Kemp for picking
1019 this up.
1020
1021
1022
1023 CHANGES IN VGAM VERSION 0.9-7
1024
1025 NEW FEATURES
1026
1027 o Tested okay on R 3.1.2.
1028 o linkfun() and nparam() are new generic functions.
1029 o betabinomialff() replaces 'lshape12' with 'lshape1' and 'lshape2'.
1030 Arguments 'i1' and 'i2' are now 'ishape1' and 'ishape2'.
1031 o ABO() has more arguments.
1032 o Arguments lower.tail and log.p have been added to quite
1033 a few pq-type functions (work done by Kai Huang).
1034
1035
1036 BUG FIXES and CHANGES
1037
1038 o Argument 'mv' has been renamed to 'multiple.responses'.
1039 This applies to about 10 family functions such as binomialff().
1040 o Argument 'lss' added to betaII(), dagum(), fisk(),
1041 genbetaII(), inv.paralogistic(), paralogistic(),
1042 sinmad(). Note that the order of the arguments of
1043 these functions will change in the near future, and
1044 consequently the order of the parameters.
1045 The [dpqr]-type functions of all these distributions
1046 have arguments that have been rearranged.
1047 o All d-type functions handle 'x = Inf' and 'x = -Inf'.
1048 Much help from Kai Huang here.
1049 Thanks to Ott Toomet for alerting me to this type of bug.
1050 o vsmooth.spline() has 2 argument name changes, and
1051 a little reordering of its arguments.
1052 o More p-type functions handle 'q = Inf' and 'q = -Inf'.
1053 More q-type functions handle 'p = 0' and 'p = 1'.
1054 Much help from Kai Huang here.
1055 o AA.Aa.aa() and A1A2A3() handled 'inbreeding' the wrong way round.
1056 o pposnorm() returned wrong answers.
1057 o ptobit(log.p = TRUE) was incorrect, as well as some
1058 other bugs in [dpqr]tobit().
1059 The dtobit(Lower) and dtobit(Upper) have changed.
1060 o negbinomial() now computes the EIM wrt the 'size' parameter based
1061 on a finite approximation to an infinite series (provide the
1062 mu and size parameter has values lying in a certain range).
1063 This may be time- and/or memory-hungry, but the user has
1064 control over this via some arguments such as max.mu, min.size
1065 and chunk.max.MB.
1066 o Renamed functions:
1067 elogit() is now called extlogit(),
1068 fsqrt() is now called foldsqrt().
1069
1070
1071
1072 CHANGES IN VGAM VERSION 0.9-6
1073
1074 NEW FEATURES
1075
1076 o All r-type functions handle the 'n' argument the same
1077 way as runif(). This was done with the help of Kai Huang.
1078
1079
1080 BUG FIXES and CHANGES
1081
1082 o Slot "res.ss" changed to "ResSS".
1083 o Some argument reference errors found by valgrind have been fixed.
1084
1085
1086
1087 CHANGES IN VGAM VERSION 0.9-5
1088
1089 NEW FEATURES
1090
1091 o Tested okay on R 3.1.2.
1092 o New argument 'lss' appears on some family functions.
1093 This is important because it changes the order of the
1094 parameters.
1095 o New functions: QR.Q(), QR.R(), [pq]rice() (thanks
1096 to Benjamin Hall for pointing that these are based on the
1097 Marcum-Q function).
1098 o exponential() has a new loglikelihood slot. Thanks to
1099 Neyko Neykov for picking this up this omission.
1100 o Constraint matrices in process.constraints() are checked
1101 that they are of full column-rank.
1102 o New family functions: better.exponential(), polyaR().
1103 o New functions: qvplot() is preferred over plotqvar(),
1104 [dpqr]levy().
1105 o summary() applied to a "vglm" object now prints out
1106 the table of estimates, SEs, test statistics
1107 and p-values very similarly to glm() objects. In
1108 particular, two-tailed p-values in the 4th column
1109 are new; these correspond to the z ratio based on
1110 a normal reference distribution.
1111 o gev(), egev(), and gpd() have a 'type.fitted' argument, which
1112 should be set to "mean" if the mean is desired as
1113 the fitted values.
1114 gpd() has a stop() if the data is negative.
1115 o AA.Aa.aa() and A1A2A3() have a 'inbreeding = TRUE' argument.
1116 If 'inbreeding = TRUE' then an extra parameter is estimated.
1117 If 'inbreeding = FALSE' then the inbreeding coefficient is 0
1118 by definition, and not estimated.
1119 G1G2G3() is now renamed to A1A2A3().
1120 o Decommissioned VGAM family functions:
1121 AAaa.nohw(), matched.binomial().
1122 o deviance() applied to a "qrrvglm" or "rrvgam" object
1123 now has a 'history' argument.
1124 o binomialff(mv = TRUE) is no longer restricted to responses
1125 having 0 and 1 values.
1126 o New data sets: flourbeetle.
1127 o The 'constraints' argument accepts a list with functions
1128 as components, that compute the constraint matrices.
1129
1130
1131 BUG FIXES and CHANGES
1132
1133 o Renamed the order of arguments and linear predictors (now,
1134 'location'-type precedes 'scale'-type, and 'scale'-type precedes
1135 'shape'-type parameters):
1136 benini1(dpqr)
1137 bisa(dpqr)
1138 gumbelII(dpqr)
1139 makeham(dpqr)
1140 nakagami(dpqr)
1141 perks(dpqr)
1142 riceff(dpqr)
1143 genrayleigh(dpqr)
1144 expexpff1(), expexpff()
1145 exppoisson(dpqr)
1146 gammaR()
1147 o Renamed parameter names: poissonff() has "lambda", not "mu",
1148 binomialff() has "prob", not "mu".
1149 o Renamed functions:
1150 plot.vgam() plots "vgam" objects, not plotvgam().
1151 Use plot(as(vglmObject, "vgam")) to plot vglm() objects as
1152 if they were vgam() objects.
1153 plot.vgam(): the user has total control over 'xlim' and 'ylim'
1154 if specified.
1155 o Renamed functions: cm.zero.vgam() has become cm.zero.VGAM(),
1156 cm.nointercept.vgam() has become cm.nointercept.VGAM(),
1157 cm.vgam() has become cm.VGAM(),
1158 process.categorical.data.vgam to process.categorical.data.VGAM,
1159 process.binomial2.data.vgam to process.binomial2.data.VGAM.
1160 o Link loge() returns "loge" as its tag, not "log" anymore.
1161 o Class "cao" changed to "rrvgam".
1162 o dbilogis4() was faulty.
1163 o Renamed arguments: 'location' is now 'scale' in [dpqr]pareto(),
1164 and paretoff().
1165 o gev() and egev() handle working weights better when sigma is close
1166 o gev(zero = 3) has changed to gev(zero = 2:3), by default, and
1167 egev(zero = 3) has changed to egev(zero = 2:3), by default.
1168 That is, only the location parameter is now modelled as
1169 functions of covariates, by default;
1170 the scale and shape parameters are intercept-only.
1171 o bigamma.mckay(zero = 1) has changed to bigamma.mckay(zero = 2:3),
1172 by default.
1173 o rlplot() works for gev() model fits now.
1174 o Renamed functions: subsetc() has become subsetcol(),
1175 my1 has become sc.min1(),
1176 my2 has become sc.min2(),
1177 stdze1() has become sc.scale1(),
1178 stdze2() has become sc.scale2(),
1179 mlogit() has become multilogit().
1180 o Decommissioned VGAM family functions:
1181 AB.Ab.aB.ab2()
1182 o Renamed VGAM family functions:
1183 OLD NAME: NEW NAME:
1184 amh() biamhcop()
1185 bigumbelI() bigumbelIexp()
1186 fgm() bifgmcop()
1187 gammahyp() gammahyperbola()
1188 morgenstern() bifgmexp()
1189 plackett() biplackettcop()
1190 benini() benini1()
1191 cgumbel() cens.gumbel()
1192 cenpoisson() cens.poisson()
1193 cennormal() cens.normal()
1194 double.cennormal() double.cens.normal()
1195 recnormal() rec.normal()
1196 recexp1() rec.exp1()
1197 invbinomial() inv.binomial.exp1()
1198 invlomax() inv.lomax.exp1()
1199 invparalogistic() inv.paralogistic.exp1()
1200 koenker() sc.studentt2()
1201 frechet2() frechet()
1202 hypersecant.1() hypersecant01()
1203 gengamma() gengamma.stacy()
1204 beta.ab() betaR()
1205 betabinom.ab() betabinomialR()
1206 gamma2.ab() gammaR() [see note about reordered arguments]
1207 logistic2() logistic()
1208 lgammaff() lgamma1()
1209 lgamma3ff() lgamma3()
1210 SUR() SURff()
1211 expexp() expexpff()
1212 expexp1() expexpff1()
1213 weibull() weibullR(lss = FALSE). Also 'zero' has changed.
1214 o Functionality has changed:
1215 weibull() weibullR(lss = FALSE). Also 'zero' has changed.
1216 o Data sets renamed:
1217 mmt renamed to melbmaxtemp.
1218 o lms.bcn(): changes in the arguments.
1219 o [log]alaplace[123](): changes in the arguments, e.g.,
1220 'parallelLocation' changed to 'parallel.locat'.
1221 o Argument 'reference' has been changed to 'refResponse' for
1222 CQO objects.
1223 o Argument 'shrinkage.init' has been changed to 'ishrinkage'.
1224 o Argument 'matrix.arg = TRUE' has been changed to
1225 'drop = FALSE' in fittedvlm().
1226 o Bug in dbort(). Thanks to Benjamin Kjellson for picking this up.
1227 o vglm.control()$save.weight changed to vglm.control()$save.weights.
1228 vgam.control()$save.weight changed to vgam.control()$save.weights.
1229 o "ccoef" has been replaced by "concoef".
1230 o Some documentation regarding qvar(se = TRUE) was wrong.
1231 o Argument "alpha" in several bivariate distributions have
1232 been replaced by "apar", for association parameter.
1233 o Arguments "optima" replaced by "optimums",
1234 "maxima" replaced by "maximums",
1235 "logmaxima" replaced by "log.maximums".
1236 o Function getMaxMin() renamed to grid.search().
1237 o lognormal3() withdrawn.
1238 o dfbeta() returns the difference between the coeffs.
1239 o negbinomial(deviance = TRUE) works when fitting the NB-2,
1240 provided criterion = "coef" or half.step = FALSE.
1241 o Argument "a" replaced by "rate" in maxwell(dpqr).
1242 o Arguments "x1" and "x2" replaced by "q1" and "q2" in pbinorm().
1243
1244
1245
1246 CHANGES IN VGAM VERSION 0.9-4
1247
1248 NEW FEATURES
1249
1250 o New data sets: cfibrosis, lakeO, wine.
1251 o New functions: Select().
1252 o negbinomial(deviance = TRUE) works, provided criterion = "coef"
1253 is used too.
1254 o simulate() works with binomialff(), poissonff(), rayleigh() and
1255 several other families. See help(simulate.vlm) for a current listing.
1256 o coef(colon = FALSE) works for VLM objects.
1257 o pslash() has a 'very.negative = -4' argument.
1258 Thanks to Tiago Pereira for picking this up.
1259 o Some family functions have a 'summation = TRUE' argument in
1260 the loglikelihood slot. Can be accessed using, e.g.,
1261 logLik(fit, summation = FALSE). See ?logLik.vlm.
1262 Similarly for deviance(fit, summation = FALSE).
1263 o Tested okay on R 3.1.0.
1264
1265
1266 BUG FIXES and CHANGES
1267
1268 o bs(), ns(), scale() and poly() are no longer smart,
1269 but they will handle simple terms such as bs(x) and scale(x).
1270 The smart version of those functions have been renamed to
1271 sm.bs(), sm.ns(), sm.scale(), sm.poly();
1272 these will handle complicated terms such as sm.bs(sm.scale(x)).
1273 o Renamed functions: identity() has become identitylink().
1274 o Argument names changed:
1275 'ITolerances' renamed to 'I.tolerances' thoughout,
1276 'EqualTolerances' renamed to 'eq.tolerances' thoughout.
1277 o Bug in mix2normal() fixed in @initialize. Thanks to Troels Ring
1278 for finding the bug.
1279 o Upon loading the package, no warnings (such as masking) is
1280 given.
1281 o multinomial(parallel = TRUE) now applies the parallelism
1282 constraint to the intercept.
1283 o If a factor response is ordered then a warning is issued for
1284 multinomial().
1285 o predict(fit, newdata = zdata2, type = "response") used
1286 to fail for z[ai][poisson][,ff]() and z[ai][negbinomial][,ff]()
1287 families. Thanks to Diego Nieto Lugilde for picking this up.
1288 o A bug with offsets and coefstart has been fixed.
1289 Thanks to Giuseppe Casalicchio for picking this up.
1290 o Variables "Blist" replaced by "Hlist".
1291 o Expression new.s.call no longer used in vglm.fit() and vgam.fit().
1292 Musual has been replaced by M1.
1293 o Variable names changed: prinia, Huggins89table1, Huggins89.t1.
1294 o Memory leaks found by valgrind have been patched.
1295
1296
1297
1298 CHANGES IN VGAM VERSION 0.9-3
1299
1300 NEW FEATURES
1301
1302 o New argument:
1303 posbinomial(omit.constant = FALSE), set to TRUE if
1304 comparing M_0/M_h models with M_b/M_t/M_tb/M_bh/M_th/M_tbh.
1305 o rcim() works with family = multinomial; in conjunction with
1306 arguments M and cindex to be specified.
1307 rcim() also had additional arguments and new defaults.
1308 o New arguments:
1309 positive Bernoulli functions have 'p.small' and 'no.warning'
1310 arguments.
1311 o AICc() is new.
1312 o family.name() generic is new.
1313 o New data sets: prinia.
1314 o logLik() methods function for "qrrvglm" objects.
1315 AIC() methods function for "qrrvglm" objects is corrected.
1316 AIC() methods function for "cao" objects is new.
1317
1318
1319 BUG FIXES and CHANGES
1320
1321 o vgam() with nontrivial constraints is giving incorrect
1322 predict(vgam.object) and fitted(vgam.object).
1323 Not yet fixed up but will try soon!
1324 Thanks to Zachary Kurtz for picking this up.
1325 o Argument 'which.lp' changed to 'which.linpred'.
1326 Argument 'which.eta' changed to 'which.linpred'.
1327 Argument 'lapred.index' changed to 'linpred.index'.
1328 Argument 'whichSpecies' changed to 'which.species'.
1329 Argument 'plot.it' changed to 'show.plot'.
1330 Argument 'intervalWidth' in plotqvar() changed to 'interval.width'.
1331 o Decommissioned VGAM family functions: cennormal1().
1332 o posbinomial() returns @extra$N.hat and @extra$SE.N.hat
1333 if the number of trials is constant across observations.
1334 o calibrate() restored to working order.
1335 o Argument names changed:
1336 'szero' renamed to 'str0' thoughout,
1337 'allowable.length' renamed to 'length.arg' in is.Numeric().
1338 o Function uqo() has been withdrawn. Reasons:
1339 (i) It needs to be rewritten in C but unfortunately am too busy...
1340 (ii) It is a very difficult optimization problem, probably too
1341 difficult to solve in general efficiently.
1342 o Arguments in rcqo() have changed.
1343 o Data set Perom withdrawn, but deermice remains.
1344 o Argument 'zero' in binom2.or() had a bug.
1345
1346
1347
1348 CHANGES IN VGAM VERSION 0.9-2
1349
1350 NEW FEATURES
1351
1352 o New family functions: logF(d), biclaytoncop(dr), binormalcop(dp),
1353 bistudentt(d), and a basic normal.vcm(),
1354 zabinomialff(), zageometricff(), zanegbinomialff(), zapoissonff(),
1355 zibinomialff(), zigeometricff(), zinegbinomialff().
1356 o cao.control()suppress.warnings == TRUE is new, and it
1357 suppresses warnings (esp. lack of convergence) by default.
1358 o The convergence criterion now takes into consideration the sample
1359 size, somewhat. It should stop premature convergence for very
1360 large data sets.
1361 o New functions: dpois.points(), log1pexp(), expint(), expexpint(),
1362 expint.E1(), dbinorm(), rbinorm(), kendall.tau(), qvar().
1363 Also, depvar(type = c("lm", "lm2")) has a 'type' argument.
1364 Also, aux.posbernoulli.t() is new.
1365 o New link functions: logneg().
1366 o New data sets: beggs, corbet, deermice, machinists, prats, V1.
1367 o Argument 'form2' added to vgam(), so vgam.fit() has been modified too.
1368 o posbernoulli.tb() seems correct, and works for any
1369 number of sampling occasions. And posbernoulli.[b,t,tb]() have
1370 more argument choices.
1371 o BIC() is now available, it is based on AIC(..., k = log(nobs(object))).
1372 But users need to use it with care.
1373 Also, AICvlm() has a 'corrected = FALSE' argument.
1374 o fittedvlm() now has a 'type.fitted' argument that allows
1375 different fitted values to be computed from a vglm()/vgam()
1376 object. Several family functions such as zi*()
1377 [e.g., zipoisson()] and za*() [e.g., zapoisson()]
1378 have a 'type.fitted' argument that matches it.
1379
1380
1381 BUG FIXES and CHANGES
1382
1383 o Default arguments have changed, esp. wrt 'zero' for:
1384 zibinomial(), zinegbinomial().
1385 o cao() used to crash due to memory problems and segment faults.
1386 o Syntax such as parallel = TRUE ~ 1 is now supported. Hence argument
1387 'apply.parint' has been removed.
1388 o posbernoulli.b() has a new and superior parameterization, & faster .
1389 o Printed output when trace = TRUE has been improved, especially
1390 for large data sets.
1391 o For ordination methods "lv" has been generally replaced by
1392 "latvar".
1393 "latvar()" is supported, "lv()" will become fully
1394 deprecated soon. But "lvplot()" is retained.
1395 Also, this applies to most argument names and list
1396 component names returned, e.g.,
1397 OLD NAME: NEW NAME:
1398 isdlv isd.latvar
1399 varlvI varI.latvar
1400 lvOrder latvar.order
1401 OptimumOrder Optimum.order
1402 maxfitted max.fitted
1403 SD.Ainit sd.Ainit
1404 SD.Cinit sd.Cinit
1405 SD.sitescores sd.sitescores
1406 o For ordination methods "ccoef" has been generally replaced by
1407 "concoef". This applies to most methods functions.
1408 Attributes have changed too, from "ccoefficients"
1409 to "con.coefficients".
1410 o VGAM now suggests \pkg{VGAMdata}.
1411 o Renamed VGAM family functions:
1412 OLD NAME: NEW NAME:
1413 normal1() uninormal()
1414 bivgamma.mckay() bigamma.mckay()
1415 cennormal1() cennormal()
1416 dcennormal1() double.cennormal()
1417 dexpbinomial() double.expbinomial()
1418 explogarithmic() explogff()
1419 frank() bifrankcop(dpr)
1420 [dpr]frank() [dpr]bifrankcop()
1421 fnormal1() foldnormal()
1422 [dpqr]fnorm() [dpqr]foldnorm()
1423 gumbelIbiv() bigumbelI()
1424 mbinomial() matched.binomial()
1425 mix2normal1() mix2normal()
1426 mix2normal1.control() mix2normal.control()
1427 nidentity() negidentity()
1428 normal1() uninormal()
1429 nloge() negloge()
1430 pnorm2() pbinorm(dpr)
1431 pareto1() paretoff()
1432 poissonp() poisson.points()
1433 powl() powerlink()
1434 recnormal1(d) recnormal()
1435 rig() rigff()
1436 skewnormal1() skewnormal()
1437 [dr]snorm() [dr]skewnorm()
1438 tpareto1() truncpareto()
1439 wald() waldff()
1440 o Decommissioned functions:
1441 OLD NEW
1442 dnorm2() dbinorm()
1443 pnorm2() pbinorm()
1444 o Renamed internal functions:
1445 OLD NEW
1446 lv.cao() latvar.cao()
1447 o Renamed arguments:
1448 OLD NEW
1449 equalsd eq.sd
1450 o Internally, variables identifiers with "_" have been replaced
1451 by a ".", e.g., X_vlm becomes X.vlm.
1452 Saved component names follow this change too, e.g.,
1453 @extra$ncols_X_lm becomes @extra$ncols.X.lm.
1454 o Improved:
1455 fgm() has its explicit EIM programmed in.
1456 o summary() applied to a "rcim0" or "rcim" object now works.
1457 o Family functions which have changed: zigeometric().
1458 o Slotname "rss" changed to "res.ss".
1459 o zinegbinomial()@weight continues to use Fisher scoring
1460 until not all the random variates are zeros or nonzeros.
1461 o loglinb2(zero = 3) and loglinb3(zero = 4:6) are defaults
1462 now (used to be zero = NULL).
1463 o Data sets moved: wffc, wffc.nc, etc. moved to \pkg{VGAMdata}.
1464 o stats::print.anova() no longer called directly by lrtest().
1465
1466
1467
1468 CHANGES IN VGAM VERSION 0.9-1
1469
1470 NEW FEATURES
1471
1472 o A companion package, called \pkg{VGAMdata}, is new.
1473 Some large data sets previously in \pkg{VGAM} have
1474 been shifted there, e.g., xs.nz and ugss.
1475 In \pkg{VGAMdata} there is (new) oly12 and students.tw.
1476 o pnorm2() argument names have changed from 'sd1' to
1477 'var1', etc. and 'rho' to 'cov12'.
1478 See documentation.
1479 Warning given if it returns any negative value.
1480 o Introduction of g-type arguments for grid search.
1481 o Improved initial values for: lomax().
1482 o Argument 'bred' works for poissonff().
1483 o latvar() generic available, identical to lv(). But the
1484 latter will be withdrawn soon.
1485 o Rank() generic available for RR-VGLMs, QRR-VGLMs, CAO models.
1486 o New function: pgamma.deriv(), pgamma.deriv.unscaled(),
1487 vlm2lm.model.matrix().
1488 o New VGAM family functions:
1489 posbernoulli.b(), posbernoulli.t(), posbernoulli.tb(tau = 2 or 3).
1490 These provide estimates of N as well as its standard error.
1491 Also, truncgeometric() and truncweibull() are new.
1492 Also, SUR() is new.
1493 Also, binom2.rho.ss() does not work yet.
1494 o New argument 'matrix.out = FALSE' for constraints.vlm().
1495 o cm.vgam() has a few more arguments to provide more flexibility.
1496 But there should be no changes for VGAM users at this stage.
1497 o Renamed functions: confint_rrnb() is now renamed to
1498 Confint.rrnb() and confint_nb1() is now renamed to Confint.nb1().
1499 o Some changes to component names returned by Confint.rrnb() and
1500 Confint.nb1(): $CI. and $SE. are uppercase.
1501 o Some zero-inflated VGAM family functions return a
1502 "vglm" object with @misc$pstr0 for the estimated
1503 probability of a structural zero.
1504 o New data set: olym12.
1505 Note that Students.tw is earmarked for \pkg{VGAMdata}.
1506 o Data sets renamed:
1507 olympic renamed to olym08.
1508 o Qvar() has a 'which.eta = 1' argument specifying which linear
1509 predictor to use. So quasi-variances are now
1510 available to models with M > 1 linear predictors.
1511 o Tested okay on R 3.0.0.
1512
1513
1514 BUG FIXES and CHANGES
1515
1516 o VGAM now depends on R >= 2.15.1.
1517 o Fortran array bounds problems (picked up by
1518 AddressSanitizer) have been fixed.
1519 o All "no visible binding for global variables" warnings
1520 have been suppressed.
1521 o vgam() with a s(spar = myspar) term should run, and if myspar
1522 is extracted from a previous vgam() model then the two models
1523 should effectively be the same.
1524 o summaryvgam() did not calculate or print out all the p-values
1525 for testing linearity.
1526 o fnormal1()@initialize was faulty wrt lm.wfit().
1527 o zageometric() and zigeometric() handle multiple responses.
1528 o mlogit(inverse = TRUE) and mlogit(inverse = FALSE)
1529 were switched. Now multinomial() makes use of mlogit().
1530 mlogit() now calls care.exp() to avoid overflow and underflow;
1531 this stops multinomial() from returning a NA as a fitted
1532 value if abs(eta) is very large.
1533 o arwz2wz() introduced to simplify multiple responses
1534 working weight matrices (wrt construction).
1535 o Renamed functions:
1536 dhuggins91() is now dposbern(),
1537 huber() is now huber2(),
1538 ei() is now eifun(), eij() is now eijfun(),
1539 rss.vgam() is now ResSS.vgam().
1540 o fisherz(theta) was wrong. Corrected, then replaced by
1541 atanh(theta).
1542 o [dpq]dagum(x), [dpq]lomax(x), [dpq]sinmad(x), etc. handled
1543 correctly for x = 0, Inf, -Inf, NaN, NA.
1544 o qdagum(x) failed due to 'Scale' [thanks to Alena Tartalova].
1545 o Arguments renamed:
1546 'intercept.apply' renamed to 'apply.parint',
1547 'Norrr' renamed to 'noRRR' (warning/error message issued),
1548 'nowarning' renamed to 'noWarning' in vglm.control().
1549 o seq2binomial()@loglikelihood includes the binomial
1550 lchoose() constants.
1551 o qgev() bug [thanks to Alex Cannon], and qgpd().
1552 o cao() produces less error/warning messages usually.
1553 o Data sets corrected for errors: chinese.nz.
1554 o Data set changes: gew had incorrect y1 and y2 values,
1555 and variables x1 to x4 have been renamed to value.g, capital.w,
1556 etc. The year variable has been added.
1557
1558
1559 CHANGES IN VGAM VERSION 0.9-0
1560
1561 NEW FEATURES
1562
1563 o Major change: VGAM family functions no longer have
1564 arguments such as earg, escale, eshape, etc. Arguments such
1565 as offset that used to be passed in via those arguments can
1566 be done directly through the link function. For example,
1567 gev(lshape = "logoff", eshape = list(offset = 0.5)) is
1568 replaced by gev(lshape = logoff(offset = 0.5)). The @misc
1569 slot retains the $link and $earg components, however,
1570 the latter is in a different format. Functions such as
1571 dtheta.deta(), d2theta.deta2(), eta2theta(), theta2eta()
1572 have been modified. Link functions have been simplified
1573 somewhat. The casual user will probably not be affected,
1574 but programmers will. Sorry about this!
1575 o New VGAM family functions:
1576 [dpqr]gompertz(), [dpqr]gumbelII(), [dpr]lindley(),
1577 [dpqr]makeham(), [dpqr]perks().
1578 o df.residual() supports a new formula/equation for 'type = "lm"'.
1579 o garma("reciprocal") supported.
1580 o is.parallel() for constraint matrices summary.
1581 o Improved family functions:
1582 these can handle multiple responses:
1583 benini(), chisq(), erlang(), exponential(), gamma1(), geometric(),
1584 gpd(), inv.gaussianff(), logff(), maxwell(), rayleigh(),
1585 yulesimon(), zetaff().
1586 o New data set: hormone
1587 [http://www.stat.tamu.edu/~carroll/data/hormone_data.txt].
1588 o If a factor response is not ordered then a warning
1589 is issued for acat(), cratio(), cumulative() and sratio().
1590 o New dpqr-type functions:
1591 [dpqr]perks(), [dpqr]mperks(), [dpqr]mbeard().
1592 o Argument 'parallel' added to gamma2().
1593 o New link functions: mlogit().
1594
1595
1596 BUG FIXES and CHANGES
1597
1598 o zibinomial() had 1 wrong element in the EIM; one of the
1599 corrections of VGAM 0.8-4 was actually incorrect.
1600 o zibinomial() blurb was wrong:
1601 previously was "(1 - pstr0) * prob / (1 - (1 - prob)^w)" where
1602 prob is the mean of the ordinary binomial distribution.
1603 Now is "(1 - pstr0) * prob".
1604 o betaff() no longer has "A" and "B" arguments; they ar
1605 extracted from "lmu = elogit(min = A, max = B)".
1606 o binom2.rho() has "lmu" as a new argument 2.
1607 o logistic2() has has zero = -2 as default, and can handle
1608 multiple responses.
1609 o gengamma() returned the wrong mean (picked up by Andrea Venturini):
1610 not b * k but b * gamma(k + 1 / d) / gamma(k).
1611 o tobit.Rd nows states vector values for 'Lower' and 'Upper'
1612 are permitted. Also, the @misc$Lower and @misc$Upper are
1613 matrices of the same dimension as the response.
1614 o constraints.vlm(type = c("vlm", "lm")) has been changed to
1615 constraints.vlm(type = c("lm", "term")) [respectively].
1616 o Rcam() renamed to Rcim(), and rcam() renamed to rcim().
1617 Class "rcam" changed to "rcim".
1618 o Days changed from "Monday" to "Mon" in all crash data frames, etc.
1619 o w.wz.merge() written to handle the working weights
1620 for multiple responses.
1621 w.y.check() written to check the integrity of prior
1622 weights and response.
1623 o Argument 'sameScale' changed to 'eq.scale',
1624 'quantile.probs' in negbinomial-type families changed to 'probs.y'.
1625 o No more warnings: dirmultinomial().
1626 o Renamed arguments: benini(earg <- eshape),
1627 binormal(equalmean <- eq.mean),
1628 binormal(equalsd <- eq.sd),
1629 o dirmultinomial() can handle a 1-row response [thanks to Peng Yu].
1630 o weibull() gives improved warnings re. the shape parameter wrt
1631 regularity conditions.
1632 o The 12 most time-consuming examples have been placed in a
1633 \dontrun{} to save time.
1634 o Argument "prob.x" renamed to "probs.x".
1635 o Argument "hbw" removed from iam().
1636 o Argument 'name' is passed into .C and .Fortran() [in dotC()
1637 and dotFortran()] is now okay because the first argument
1638 is unnamed.
1639
1640
1641
1642 CHANGES IN VGAM VERSION 0.8-7
1643
1644 NEW FEATURES
1645
1646 o Modified VGAM family functions:
1647 genbetaII()@initialize has been improved, as well as
1648 those special cases of that distribution (such as sinmad,
1649 lomax, paralogistic, dagum, etc.).
1650 o Argument 'lapred.index' added to model.matrix().
1651 o npred() is now defined as a generic function (returns M).
1652 o hatvalues() and hatplot() written for vglm() objects.
1653 o The argument 'qr.arg' is set TRUE now by default in vglm().
1654 o df.residual() supports the argument 'type = c("vlm", "lm")'.
1655 o Argument 'nowarning' added to vglm.control().
1656 o New data set: ucberk.
1657 o Improved functions: rposbinom(), rposgeom(), rposnegbin(),
1658 rpospois().
1659 o Tested okay on R 2.15.0.
1660
1661
1662 BUG FIXES and CHANGES
1663
1664 o Labelling of the linear predictors for sratio(), cratio()
1665 etc. was faulty.
1666 o pbetabinom.ab() did not recycle shape1 correctly
1667 [found by David Venet].
1668 o Arguments lower.tail and log.p not supported (temporarily) in
1669 pposbinom() and qposbinom().
1670
1671
1672
1673 CHANGES IN VGAM VERSION 0.8-6
1674
1675 NEW FEATURES
1676
1677 o Modified VGAM family functions:
1678 sinmad()@initialize has been improved.
1679
1680
1681 BUG FIXES and CHANGES
1682
1683 o VGAM now depends on R >= 2.14.0.
1684 o Trying to eliminate some residual errors with the NAMESPACE.
1685
1686
1687
1688 CHANGES IN VGAM VERSION 0.8-5
1689
1690 NEW FEATURES
1691
1692 o New VGAM family functions:
1693 negbinomial.size(), zabinomial(dpqr),
1694 zageometric(dpqr), [dpqr]posgeom().
1695 o New link functions: nbcanlink().
1696 o Modified VGAM family functions:
1697 posnegbinomial(), zanegbinomial() and zinegbinomial()
1698 use the nsimEIM argument;
1699 zipoisson() handles a matrix response;
1700 all [dpqr]zi-type functions handle zero-deflation,
1701 normal1() can model the variance too as the 2nd parameter.
1702 o Rudimentary methods functions for lrtest() and
1703 update(), based on packages lmtest and base.
1704 o The VGAM family functions for genetic models have been improved
1705 wrt initial values.
1706 o New data sets: xs.nz.
1707
1708
1709 BUG FIXES and CHANGES
1710
1711 o In anticipation for R version 2.15.0, VGAM imports
1712 from stats4 'coef', 'plot', 'summary', 'vcov'.
1713 Calls to 'print' have been replaced by 'show'
1714 since VGAM uses S4 methods.
1715 Numerous NAMESPACE changes have been made.
1716 No more warnings during checking and installation!
1717 o Labelling in summary() of vglm() objects changed. It
1718 now closely follows glm(). In particular, it has
1719 changed from c("Value", "Std. Error", "t value") to
1720 c("Estimate", "Std. Error", "z value").
1721 Note that "z value" might change later to, e.g., "Wald".
1722 o Zero-inflated and zero-altered functions have renamed and
1723 reordered arguments. Ouch!
1724 These include 'pstr0' for probability of a structural 0 [zero-inflated],
1725 and 'pobs0' for probability of an observed 0 [zero-altered].
1726 For example, argument lpstr0 replaces lphi in zipoisson().
1727 The order of these arguments, including the respective
1728 dpqr-type functions, may have changed too.
1729 o zapoisson() now implements Fisher scoring.
1730 o zipoissonff() had the wrong sign for the non-diagonal EIM element.
1731 o nobs() is now defined as a generic function (needed
1732 for older versions of R---versions 2-12.2 or earlier, actually).
1733 o Data sets renamed:
1734 uscrime and usgrain renamed to crime.us and grain.us;
1735 bminz renamed to bmi.nz,
1736 nzc renamed to chinese.nz,
1737 nzmarital renamed to marital.nz.
1738 o Improved family functions: genbetaII(), betaII(), sinmad(), dagum(),
1739 lomax(), invlomax(), fisk(), invparalogistic(), paralogistic();
1740 wrt fitted values (range checks in place now).
1741 These functions have many argument names changed, e.g.,
1742 link.a is now lshape1.a, init.a is now ishape1.a.
1743 Also, some default initial values have changed from 1 to 2.
1744 o Argument names changed (Ouch!):
1745 q.lag.ma changed to q.ma.lag in garma().
1746
1747
1748
1749
1750 CHANGES IN VGAM VERSION 0.8-4
1751
1752 NEW FEATURES
1753
1754 o VGAM family functions renamed (Ouch!):
1755 1. 'betabinom.ab' renamed to 'betabinomial.ab'.
1756 o Other functions renamed (Ouch!):
1757 1. '[dpr]betabin' renamed to '[dpr]betabinom' etc.;
1758 2. '[dpr]betabin.ab' renamed to '[dpr]betabinom.ab' etc.;
1759 o Slot names changed (Ouch!):
1760 1. 'link' renamed to 'linkfun' for class "vglmff";
1761 2. 'inverse' renamed to 'linkinv' for class "vglmff".
1762 o Extra arguments added to freund61().
1763 o New VGAM family functions:
1764 abbott(), zigeometric(dpqr), huber1().
1765 o New functions: [p]polono(),
1766 depvar() generic for the dependent (response) variable,
1767 Qvar() and explink() for quasi-variances.
1768 o Improved functions: [d]polono(), [dr]betabin() handles rho = 0
1769 (suggested by Peng Yu).
1770 o Improved family functions: normal1() handles matrix 'weights'.
1771 o Defaults changed: [dr]betabin(rho = 0).
1772 o New methods functions: nobs(), nvar(), depvar().
1773 o Renaming: fitted.vlm() is now fittedvlm(),
1774 persp.qrrvglm() is now perspqrrvglm(),
1775 predict.qrrvglm() is now predictqrrvglm(),
1776 predict.vglm() is now predictvglm().
1777 o New data sets: finney44.
1778 o VGAM now depends on R >= 2.11.1.
1779 o Tested okay on R 2.14.0.
1780
1781
1782 BUG FIXES
1783
1784 o zibinomial() had 2 wrong elements in the EIM,
1785 thanks to Alan Welsh for picking this up.
1786 o margeff() for cumulative() was faulty.
1787 o blurb slot of binormal() was faulty.
1788 o betabinomial() did not return the estimated rho in @misc$rho as
1789 did betabinomial.ab().
1790 o kumar() did not initialize well with non-integer prior weights.
1791 o rdagum() did not handle the 'scale' argument correctly.
1792 o codes() in s() is defunct so it has been replaced.
1793
1794
1795
1796 CHANGES IN VGAM VERSION 0.8-3
1797
1798 NEW FEATURES
1799
1800 o Argument names changed (Ouch!):
1801 1. 'method.init' renamed to 'imethod';
1802 2. 'k' renamed to 'size' in negbinomial(), zanegbinomial(),
1803 posnegbinomial(), and zinegbinomial():
1804 e.g., 'lk' renamed to 'lsize', 'ik' renamed to 'isize', etc.
1805 o New data sets: hued, huie, huse, ugss, uscrime;
1806 usagrain renamed to usgrain.
1807 o The "prior.weights" slot is now a "matrix", not "numeric".
1808 o [dpr]betabin.ab() now handles size = 0. Thanks to Robert Wolpert
1809 for picking up this bug.
1810 o New VGAM family functions:
1811 [d]binormal(), [dr]huggins91() but this is not working properly,
1812 [dpqr]explogarithmic(), polya().
1813 o New functions: [dpqr]tobit().
1814 o Improved family functions: tobit() implements the proper EIM for the
1815 standard model, dcennormal1() has some new arguments [and renamed],
1816 cennormal1(), cenrayleigh() renamed.
1817 o VGAM now depends on R >= 2.10.0.
1818 o Rcam(), moffset(), etc. for RCAMs have been improved and modified.
1819 o VGAM family functions currently withdrawn: frechet3().
1820 o Tested ok on R 2.13.0.
1821
1822
1823 BUG FIXES
1824
1825 o tobit()@loglikelihood had omitted the constant in dnorm().
1826 Also, tobit() uses simulated Fisher scoring (for nonstandard model).
1827 o moffset() was buggy.
1828 o ABO() is a little more robust.
1829 o dirichlet()@loglikelihood miscalculated.
1830 Ditto for gaussian()@loglikelihood (constants were omitted).
1831 Thanks to Arne Henningsen for picking up these bugs.
1832 o zipoissonff() did not initialize correctly and labelling was wrong
1833 with matrix responses.
1834
1835
1836
1837 CHANGES IN VGAM VERSION 0.8-2
1838
1839 NEW FEATURES
1840
1841 o Objects of class "vglmff" have a "infos" slot to give
1842 information about the family.
1843 o New functions: lambertW(), rcam(), wffc.P3(),
1844 wffc.P3star(), confint_rrnb(), confint_nb1().
1845 o New VGAM family functions:
1846 binom2.Rho(), [dpqr]expgeometric(),
1847 [dpqr]genrayleigh(), [dpqr]huber(),
1848 [dpqr]koenker(), studentt[23](), zipoissonff().
1849 o Argument 'imethod' changed to 'method.init' for some
1850 families, e.g., cnormal1(), tobit(), weibull().
1851 o Improvements have been made to binom2.rho().
1852 o Improved family functions:
1853 negbinomial() has a new argument 'parallel',
1854 micmen() has more initial value choices and fitting algorithms,
1855 kumar(), studentt() and studentt2() now implement the EIM,
1856 normal1() can handle multiple responses.
1857 o Argument names changed:
1858 'init.rho' renamed to 'irho' in binom2.rho(),
1859 'a' renamed to 'scale' in rayleigh(),
1860 'Structural.zero' renamed to 'szero' thoughout.
1861 o zapoisson() permutes the linear/additive predictors.
1862 o Several families such as negbinomial(), z[ai]negbinomial(),
1863 zapoisson(), gamma2(), handle zero = c(-2, 3), say,
1864 i.e., negative and positive values.
1865 o New data sets: about half a dozen road crash data frames of
1866 2009 NZ data.
1867 o constraints(vglmfit) now has a 'type' argument that can be fed
1868 into the original fit (type = "lm") as the constraints argument.
1869 o vchol() takes drastic action to avoid infinite looping: it sets
1870 the working weights to be proportional to the order-M diagonal
1871 matrix.
1872 o lognormal() and lognormal3() now have zero = 2 as the default
1873 (was zero = NULL).
1874 o Some variable names within grc() changed, e.g., Row. and not Row.
1875 o The smartpred package within VGAM has updated poly(), ns(),
1876 bs() and scale() for R version 2.12.0 (2010-10-15).
1877 Calls to smartpredenv are now VGAM:::smartpredenv.
1878
1879
1880 BUG FIXES
1881
1882 o VGAM:::VGAMenv is used now to avoid problems locating
1883 this environment.
1884 o Input of the mustart, etastart and coefstart arguments at
1885 the solution should results in only one iteration being needed.
1886 o binomialff() and categorical familes (e.g., multinomial) only
1887 accept a factor or non-negative counts as the response.
1888 This allows the 'weights' vector to have any value really.
1889 In the past the denominator of a sample proportion was allowed
1890 via the 'weights' argument.
1891 o wffc.P1() had some rounding problems, e.g., with 0.280 m.
1892
1893
1894
1895
1896 CHANGES IN VGAM VERSION 0.8-1
1897
1898 NEW FEATURES
1899
1900 o Most of the Fortran 77 code has been converted to C.
1901 This change will be largely hidden from most users
1902 but there may be the occasional bug not detected.
1903 Much of the heavy work was done by Alvin Sou.
1904 o lms.bcn()@loglikelihood incorporates the constants in
1905 the log-likelihood.
1906 o Also, no more F90 code! This means less portability/platform
1907 problems.
1908 o bivgamma.mckay, formerly mckaygamma2(), has been modified
1909 substantially.
1910 o Improvements have been made to simplex() and
1911 [dr]simplex() have been written.
1912 o Expectile functions for the uniform, normal and
1913 exponential distributions: [dpqr]-type functions.
1914 o cqo() has EqualTolerances = TRUE and ITolerances = FALSE as
1915 the default now. The result is that cqo() should work without
1916 the environmental variables being scaled.
1917 If it is scaled then setting ITolerances = TRUE will result
1918 in greater speed and requiring less memory.
1919 o Families that deal with proportions, such as binomialff() and
1920 betabinomial(), incorporate weights separately from the weights
1921 generated by the response/counts. So the weights argument can
1922 now have any positive values.
1923 o rrvglm(..., Norrr = NULL) can be used so that the reduced-rank
1924 regression is applied to every variable including the intercept.
1925 o Renaming: ggamma() is now gengamma(), etc.
1926 o Improved functions: negbinomial() has a few new arguments.
1927
1928
1929 BUG FIXES
1930
1931 o Deviance.categorical.data.vgam did not handle small
1932 fitted probabilities.
1933 o binom2.rho() could produce small negative fitted probabilities.
1934 o seq2binomial() did not initialize 'mvector'.
1935 o zeta() crashed on some platforms.
1936 o cqo() appears to be working again with the new C code.
1937 o cao() still not working with the new C code.
1938 o zapoisson() did not implement the elambda argument correctly.
1939 o Tested ok on R 2.11.1.
1940
1941
1942
1943 CHANGES IN VGAM VERSION 0.7-10
1944
1945 NEW FEATURES
1946
1947 o Surv() renamed to Surv4(), class "SurvS4" renamed to "Surv4".
1948 o coef(summary(vglmObject)) returns a 3-column matrix of estimates,
1949 standard errors and Wald statistics, rather than coef(vglmObject)
1950 in the past.
1951 o Improved VGAM family functions: fff() uses simulated Fisher
1952 scoring now and has slightly better initial values.
1953 o New VGAM family functions:
1954 propodds(reverse) is equivalent to
1955 cumulative(parallel=TRUE, reverse=reverse) (for convenience only).
1956 o Compatible with R 2.10.1 and the article "The VGAM package
1957 for categorical data analysis," Journal of Statistical Software,
1958 2010. A vignette based on this paper is included.
1959 o Argument w.als renamed to w.aml in amlnormal().
1960
1961
1962 BUG FIXES
1963
1964 o VGAM family functions: fff() had poor initial values.
1965 o betabinomial()@loglikelihood required 'ycounts' to be integer.
1966 o [dpqr]betanorm() were written but not in the NAMESPACE.
1967
1968
1969
1970
1971
1972 CHANGES IN VGAM VERSION 0.7-9
1973
1974 NEW FEATURES
1975
1976 o New functions: margeff() for marginal effects of a vglm()
1977 "multinomial" or "cumulative" model.
1978 o Almost all VGAM family functions now have a "loglikelihood"
1979 slot that incorporates any constants in the density function.
1980 Hence the fitted likelihood may differ by a constant from
1981 previous results.
1982 In particular, models such as multinomial(), cumulative()
1983 and binom2.or() have this new feature.
1984 o vglm() now has a modified 'xij' argument which implements
1985 eta-specific covariates. Usage now involves the form2' argument,
1986 and the 'xij' argument does not interfere with constraint matrices.
1987 Documentation is supplied on the VGAM website, in particular,
1988 http://www.stat.auckland.ac.nz/~yee/VGAM/doc/xij.pdf
1989 o cases.names() and variable.names() methods functions
1990 written for vglm()/vgam() objects.
1991 o cumulative() has improved initial values, especially for
1992 long data format, i.e., when each row of the data frame is an
1993 observation rather than inputting a matrix of counts.
1994 o rrvglm() handles a factor response without giving a warning.
1995 o New data: olympic.
1996 o testf90.f90 has been renamed to testf90.f95. This may decrease
1997 the incidences of compiler problems on some platforms (f95 seems
1998 more popular than f90).
1999 o For cqo() objects: AIC(), resid() have been written.
2000 o Improved functions: negbinomial() default initial values are
2001 more robust to outliers in the response, betabinomial() and
2002 betabin.ab() have better initialization and "loglikelihood"
2003 slot matches dbetabin.ab(log=TRUE).
2004 o Renamed VGAM family functions: alsqreg() becomes amlnormal().
2005 o Renamed arguments: lmu replaces link.mu in zibinomial().
2006 o dzeta(p) has changed wrt 'p'.
2007 o The functions summary.lms() and summary.rc.exponential() are
2008 no longer distributed to avoid a warning wrt S3 vs S4
2009 methods dispatch.
2010 o The VGAM family functions for genetic models have been improved,
2011 e.g., some basic error checking.
2012 Also some changes in the names of the parameters, e.g.,
2013 "q" to "pB" for ABO(), plus some switching of the order of the
2014 arguments.
2015
2016
2017 BUG FIXES
2018
2019 o VGAM interferes much less in regard to generic functions
2020 such as predict(), fitted(), resid(), wrt other packages and
2021 also including base's lm(), glm(), etc.
2022 o AIC() method for rrvglm() objects was wrong (did not account
2023 for argument 'Structural.zero').
2024 o dzibinom(log=TRUE) was wrong.
2025
2026
2027
2028
2029 CHANGES IN VGAM VERSION 0.7-8
2030
2031 NEW FEATURES
2032
2033 o [dpqr]benf() written for Benford's distribution.
2034 o plog() and dlog() improved.
2035 o multinomial() now has a refLevel argument to specify the
2036 reference or baseline level of the factor.
2037 o binom2.or() has a new argument 'morerobust'.
2038 o Renamed arguments in mix2normal1().
2039 o DeLury() written.
2040 o [dpqr]-type functions related to the negative binomial distribution
2041 have changed wrt argument names and order.
2042 o [pq]posnegbin(), [dpqr]zapois(), [dr]zanegbin() written.
2043 o [dpqr]zinb() renamed to [dpqr]zinegbin().
2044 o lms.bcn(), lms.bcg(), lms.yjn() have zero=c(1,3) as the new default.
2045 This will increase the chances of successive convergence.
2046 o Renamed arguments in lms.bcn(), lms.bcg(), lms.yjn(), e.g.,
2047 link.sigma is now lsigma.
2048 Some new arguments added too, e.g., llambda.
2049 o Works for R 2.5.0 and later (not 2.4.0 and later).
2050 Compatible with R 2.7.2.
2051 o Contains Fortran 90 code (since 0.7-7, actually). This will
2052 cause problems on older Linux machines without such a
2053 compiler. Actually, removing the .f90 file(s) will not be too
2054 much of a problem as there is very little F90 code in use by
2055 the package at the moment.
2056 o New functions: dbinom2.rho(), rbinom2.rho(), dposnegbin().
2057 o New data: wffc, wffc.indiv, wffc.teams, wffc.nc.
2058 o Improved functions: binom2.rho(), micmen(), negbinomial(),
2059 poissonff(), posnegbinomial(), zanegbinomial(),
2060 o A new form2 argument has been added to vglm().
2061 VGAM family functions such as micmen() have the regressor inputted
2062 using form2 now, rather than the regressor argument.
2063 The resulting usage is a more elegant.
2064 Fitted objects have a few more slots and formulas put in set
2065 places on the object.
2066 o AIC() methods functions has been modified.
2067
2068
2069 BUG FIXES
2070
2071 o The big bug whereby vgam(... ~ s(x), ... ) did not work under
2072 Windows was due to a single array element that was not initialized.
2073 Evidently, Linux compilers seemed to have set it to zero.
2074 Funny, the code has worked for decade or so...
2075 o dposbinom() was buggy at x=0. Also it now handles size=0 and
2076 prob=0 or prob=1.
2077 o pzipois() was buggy at x<0.
2078 o dbetabin.ab(log=T) was incorrect outside its support.
2079 o zipf() did not handle 0 < s < 1.
2080 o data(ruge) was faulty.
2081 o summary(rrvglmObject) failed.
2082 o persp.cao(ylim=) did not work.
2083 o plotvgam() failed when se=TRUE and which.cf was specified.
2084
2085
2086
2087
2088 CHANGES IN VGAM VERSION 0.7-7
2089
2090 NEW FEATURES
2091
2092 o Labelling changes: binom2.or() uses "oratio" instead of "OR" (stands
2093 for the odds ratio).
2094 o New VGAM family functions:
2095 zipebcom().
2096 o New functions: dbinom2.or(), rbinom2.or().
2097 o binom2.or() has new arguments 'imu1, 'imu2' and 'ioratio' for
2098 inputting optional marginal probabilities and odds ratio.
2099 The third element of the score vector uses a new formula.
2100 o [dpqr]zinb() has arguments prob and munb set to NULL by default.
2101 o Compatible with R 2.7.0.
2102
2103
2104 BUG FIXES
2105
2106 o gaussianff()@loglikelihood was buggy.
2107 o all(trivial.constraints(Blist)) changed to
2108 all(trivial.constraints(Blist) == 1) to avoid a warning in R 2.7.0.
2109 Ditto for 'all(findex)' and 'any(diff(Alphavec))'.
2110 o qtriangle(0.3, theta=0.3) used to fail.
2111 o gharmonic() handles a negative argument s.
2112
2113
2114
2115 CHANGES IN VGAM VERSION 0.7-6
2116
2117 NEW FEATURES
2118
2119 o dpolono() has a new argument 'bigx' which implements an
2120 approximation. It is for handling large values of x.
2121 o vglm() and vgam() now create the response and model matrices etc.
2122 in the same way as glm(). A consequence is that the response
2123 does not have to be "numeric" as in lm(), e.g., a factor response
2124 is now permitted.
2125 o New VGAM family functions:
2126 alaplace1(), alaplace2(), alaplace3(dpqr),
2127 amlbinomial(), amlexponential(), amlpoisson(), amh(),
2128 lqnorm(), mbinomial(), scumulative().
2129 o Other VGAM family functions with argument names changed or added:
2130 lms.yjn2().
2131 o These VGAM family functions have been improved:
2132 alsqreg() [parallel option, w argument can be a vector, link function
2133 for the expectiles].
2134 o The data set "aml" has been renamed "leukemia".
2135 o Previously laplace(zero=NULL), now laplace(zero=2).
2136
2137 BUG FIXES
2138
2139 o deplot() applied to a "lms.yjn2" object gave an unnecessary warning.
2140 o In the leukemia and toxop data sets 1L is replaced by 1 and 2L by 2 etc.
2141
2142
2143
2144
2145 CHANGES IN VGAM VERSION 0.7-5
2146
2147 NEW FEATURES
2148
2149 o New VGAM family functions:
2150 betaff(), cardioid(dpqr), cauchy(), felix(d), fnormal1(dpqr),
2151 invbinomial(), kumar(dpqr), lms.yjn2(), mix2exp(), plackett(dpr),
2152 riceff(dr), skellam(dr), zinegbinomial(dpqr).
2153 o These VGAM family functions have been improved:
2154 frank(), genpoisson(), hzeta(), mix2normal1(), mix2poisson(),
2155 pospoisson(), studentt().
2156 o These VGAM family functions have had their default arguments changed:
2157 genpoisson(), mix2normal1().
2158 o New documentation: borel.tanner(dr).
2159 o expm1() used whenever possible.
2160 o Renamed VGAM family functions: betaff() changed to beta.ab().
2161 o cauchy1() now returns the location estimates as the fitted values
2162 instead of NA (for the mean).
2163
2164
2165 BUG FIXES
2166
2167 o cumulative(), sratio(), cratio(), acat() had response-matrix
2168 column names which got lost.
2169 o lms.yjn() failed if there was not enough data.
2170
2171
2172
2173
2174 CHANGES IN VGAM VERSION 0.7-4
2175
2176 NEW FEATURES
2177
2178 o weibull() does not handle any censored observations at all.
2179 The function cenweibull(), which will handle censored
2180 observations, is currently being written and will use Surv()
2181 as input; it should be distributed with version 0.7-5 of VGAM.
2182 o bisa() now implements full Fisher scoring. No numerical
2183 integration is needed.
2184 o Certain functions from the smartpred package are no longer
2185 distributed with the VGAM package. These are
2186 lm, glm, predict.lm, predict.mlm, predict.glm.
2187 This is done because many users have found they interfere with
2188 the VGAM package in unpredictable ways.
2189 o The following VGAM family functions have improved initial values:
2190 betabinomial(), cauchy1(), mccullagh89(), negbinomial(),
2191 tpareto1(), zipoisson().
2192 o New family functions: alsqreg(), dexpbinomial(), laplace(),
2193 poissonp(), seq2binomial(), triangle(dpqr).
2194 o VGAM family functions currently withdrawn:
2195 cexpon().
2196 o A new class called "SurvS4" has been prepared. It will be used
2197 later to handle VGAM family functions beginning with "cen" that
2198 use Surv() as input.
2199 o log1p() used whenever possible.
2200
2201
2202 BUG FIXES
2203
2204 o bisa() did not make use of ishape.
2205 o cao(..., family=gaussianff) failed. It now works, although the
2206 dispersion parameter is computed using a slightly different
2207 formula.
2208
2209
2210
2211 CHANGES IN VGAM VERSION 0.7-3
2212
2213 NEW FEATURES
2214
2215 o gpd() now does not delete any data internally. The user should
2216 use the subset argument of vglm() and vgam() in order to select
2217 any subset of a data frame.
2218 o zapoisson() has a zero argument, and this can be assigned
2219 a negative value.
2220 o "partial for" is added to the ylabel of linear terms of a vgam()
2221 object when it is plotted.
2222 o When a vgam() object is plotted with se=TRUE and if there are
2223 linear terms then the mean of x is added to the plot (this makes
2224 the standard error curves meet there).
2225 o This package has been tested (somewhat) under R 2.5.0.
2226
2227 BUG FIXES
2228
2229 o plotvgam() did not work for vgam() objects using the subset argument.
2230 o cao() objects would not show() or print(), at least under R 2.4.1.
2231 o summary(vgam.object) failed if vgam.object was a totally linear
2232 model (i.e., no s() term in the formula). Now the "R" slot is
2233 assigned for all vgam() objects.
2234 o preplotvgam() had a bug regarding $se.fit of an atomic pred$se.fit.
2235
2236
2237
2238 CHANGES IN VGAM VERSION 0.7-2
2239
2240 NEW FEATURES
2241
2242 o Almost all VGAM family functions now have an earg-type argument
2243 to support each link function. This allows parameters specific
2244 to each link to be passed in,
2245 e.g., VGAMfamilyfunction(link="logoff", earg=list(offset=1))
2246 o rinv.gaussian() is new.
2247 o New VGAM family functions: morgenstern(), fgm(), gumbelIbiv(),
2248 ordpoisson().
2249 o New documentation: powl(), fsqrt().
2250
2251 BUG FIXES
2252
2253 o zanegbinomial()@last had wrong names in misc$link.
2254 o summary(vgam.object) failed to print the anova table.
2255 o summary(cao.object) failed.
2256
2257
2258 CHANGES
2259
2260 o binom2.or() has argument names changed from "lp" to "lmu" etc.
2261 This is partly to make it in keeping with other VGAM family
2262 functions for binary responses.
2263 o Other VGAM family functions with argument names changed: frank().
2264 o lms.bcn(), lms.bcg(), lms.yjn() arguments have changed order.
2265 o hyper() renamed to hyperg().
2266 o plotvgam() uses ylim if it is inputted.
2267
2268
2269
2270
2271 CHANGES IN VGAM VERSION 0.7-1
2272
2273 NEW FEATURES
2274
2275 o VGAM family functions now require full (name) specification
2276 of parameter link functions. For example, binomialff(link=probit)
2277 is ok, as is binomialff(link="probit"), but binomialff(link="pr")
2278 isn't.
2279 VGAM family functions no longer offer a fixed set of link
2280 functions but the user can invoke any, as well as write their
2281 own link function.
2282 o Working residuals for vglm() objects are now the default.
2283 They used to be deviance residuals but they are not defined
2284 for most VGAM family functions.
2285 In the future the default may become "pearson" residuals.
2286 For safety, use the type argument, e.g.,
2287 resid(vglmobject, type="response").
2288 o ITolerances=TRUE is now the default for qrrvglm.control(),
2289 consequently, equal tolerances CQO models are fitted. The
2290 rationale for this change that setting ITolerances=TRUE
2291 provides the fast computational speed as well as the easiest
2292 interpretation of the results. Also, rcqo() matches this by
2293 having EqualTolerances=TRUE as its default. However, having an
2294 equal tolerances assumption should be checked.
2295 o New VGAM family functions: tikuv(dpqr), [dpqr]naka(), [dpr]log(),
2296 [dpqr]tpareto1(), betabinomial().
2297 o VGAM family functions which have been renamed (and often improved):
2298
2299 New name Old name
2300 -------- --------
2301 dirmultinomial() dirmul()
2302 negbinomial() negbin.mu()
2303 negbinomial.ab() negbin.mn()
2304 posnegbinomial() posnegbin.mu()
2305 zanegbinomial() zanegbin.mu()
2306 rposnegbin() rposnegbin.mu()
2307 gamma2() gamma2.ab()
2308 gamma2mu() gamma2()
2309
2310
2311 o New functions: lerch(), rcqo().
2312 o In the smartpred package smart.mode.is(mode.arg) now requires
2313 mode.arg, if given, to be exactly one of 3 character strings.
2314 Also, is.smart(object) handles object being a smart function
2315 or a fitted object.
2316 o The VGAM package comes with modified lm, predict.lm, predict.glm,
2317 predict.mlm, glm functions---these implement smart prediction, and
2318 are current to R version 2.3.1 (2006-06-01).
2319 o The order of the linear/additive predictors for expexp()
2320 have been switched.
2321 o weibull(zero=2) is the default now.
2322 o negbin.mu(), posnegbin.mu(), zanegbin.mu: these have a few added
2323 arguments for further flexibility, and some arguments have changed
2324 names, e.g., 'k.init' has been changed to 'ik' and 'link.mu' to
2325 'lmu'.
2326 o Negative binomial random variates are now generated using
2327 rnbinom() in the stats package rather than rnegbin() in the MASS
2328 package.
2329 o binom2.or() and binom2.rho() have more choices for some arguments
2330 such as lor and lrho.
2331 o Initial values have been improved for logff(), zipf() and zetaff().
2332 o This package should work for R 2.4.0 after additional tweaks to
2333 handle changes in show().
2334
2335
2336 BUG FIXES
2337
2338 o pbetabin() had a bug.
2339 o studentt() has a mean of 0 only if df > 1.
2340 o garma() failed for link="loge". It now works for binary
2341 data with the "logit" link.
2342 o Internally, wz <- matrix(NA, ...) changed to
2343 wz <- matrix(as.numeric(NA), ...).
2344 Ditto for rep(NA, ...) to rep(as.numeric(NA), ...).
2345 o tobit() had a bug in the initialize slot.
2346 o rposnegbin.mu() now calls the MASS library function rnegbin()
2347 explicitly.
2348 o gammaff() now works.
2349 o Working residuals for cao() objects were wrong.
2350 o lvplot() for cao() objects have a whichSpecies argument
2351 which allows selective plotting of the species' curves.
2352 o gaussianff() did not work with rrvglm(). It now has a
2353 loglikelihood slot, and returns deviance residuals for M>1.
2354
2355
2356 CHANGES
2357
2358 o gaussianff(), studentt() have the order of its arguments changed.
2359 o eta2theta(), theta2eta(): if these have a matrix "theta" then it
2360 no longer calls the VGAM link function one column at a time.
2361 Hence VGAM link functions must handle matrix "theta" using one
2362 value of "earg" argument.
2363 o The earg argument has changed for many VGAM link functions. It
2364 is now a list, with component names that are specific to each
2365 link function.
2366 See the online help files for the list component names.
2367 Soon, every VGAM family function that allows a link function
2368 will have an earg argument to match it, thus giving maximum
2369 flexibility.
2370
2371
2372
2373
2374 CHANGES IN VGAM VERSION 0.6-9
2375
2376
2377 NEW FEATURES
2378
2379 o New VGAM family functions: lino(dpqr), recexp1(),
2380 posnormal1(dpqr), betageometric(dpr), [dr]polono(),
2381 [dpr]betabin(), gamma2mu(), bisa(dpqr), zipf(dp).
2382 There is a new dirmul() (the old one is renamed to dirmul.old())
2383 but it hasn't yet be completed.
2384 o Renamed VGAM family functions: beta2() changed to betaff().
2385 o Renamed VGAM functions: is.a.number() changed to is.Numeric().
2386 o The Windows crossbuild was done under R 2.3.0.
2387
2388 BUG FIXES
2389
2390 o Family functions lognormal(), lognormal3() now include the
2391 1/sqrt(2*pi) constant in @loglikelihood because of its use of
2392 dnorm(..., log=TRUE) and dlnorm(..., log=TRUE).
2393 o [dpqr]lognormal() withdrawn as they exist in R already.
2394 o Documentation for betaff() contained mistakes.
2395 o summary() of a betabin.ab() object used to fail.
2396 o The assign statement has been removed from some FORTRAN code.
2397
2398
2399
2400 CHANGES IN VGAM VERSION 0.6-8
2401
2402 NEW FEATURES
2403
2404 o New VGAM family functions: recnormal1(), recexp1(),
2405 paretoIV(dpqr), paretoIII(dpqr), paretoII(dpqr),
2406 gammahyp(), benini(dpqr).
2407 However, the fitted value (mean) for benini() may be faulty.
2408 o Decommissioned VGAM family functions: gpdold(), ogev(),
2409 zipoissonX().
2410 o gumbel.block() renamed to gumbel(), and gumbel() renamed to egumbel().
2411 o Argument names and defaults have changed for: gpd(), egev(), gev(),
2412 ogev(), cgumbel(), egumbel(), gumbel(), and weibull().
2413 Also, gpd(), gev() and egev() have some improvements done internally.
2414 Also, rlplot() is new.
2415 o Several family functions have been converted to a new convention
2416 whereby ilocation, iscale, ishape arguments are used, and also
2417 llocation, lscale, lshape arguments for the link functions etc.
2418 o New link function: nidentity(theta) for negative-identity: -theta.
2419 o New argument "untransform" in predict() and vcov() for VGLMs.
2420 o For intercept-only models, Coef(fit) returns more user-friendly
2421 labelled output.
2422
2423
2424 BUG FIXES
2425
2426 o ppareto() had a bug.
2427 o gpd() had an incorrect second derivative.
2428
2429
2430
2431 CHANGES IN VGAM VERSION 0.6-7
2432
2433 NEW FEATURES
2434
2435 o New VGAM family functions: bilogistic4(dpr), frechet2(),
2436 frechet3(), freund61(), frank(dpr), mccullagh89().
2437 o For cao(), df1.nl has a default of 2.5, changed from 2.0 before.
2438 o For vglm(), vgam() etc., diagonal elements of the working weight
2439 matrices that are less than .Machine$double.eps^0.75 are replaced
2440 by this value. The arguments checkwz and wzepsilon support this
2441 feature.
2442 o More documentation on: fill() [re. the xij argument], cauchy1().
2443 o logistic2() now uses Fisher scoring.
2444 o Argument init.method changed to method.init in several family
2445 functions.
2446 o Any non-smart-prediction use of smartpredenv has been changed
2447 to VGAMenv.
2448
2449
2450 BUG FIXES
2451
2452 o rayleigh() was not in NAMESPACE.
2453 o logistic1() and logistic2() had wrong first derivatives and
2454 loglikelihood function.
2455 logistic1() offers some choice of link function for the location
2456 parameter.
2457
2458
2459
2460 CHANGES IN VGAM VERSION 0.6-6
2461
2462 NEW FEATURES
2463
2464 o New functions: zibinomial(), zibinom(dpqr), posbinom(dpqr),
2465 mix2normal1(), mix2poisson(), dsnorm(), rsnorm(),
2466 cexpon(), cgumbel(), cnormal1(), hyper().
2467 o New generic functions: is.bell() works for RR-VGLMs, QRR-VGLMs
2468 and RR-VGAMs (CLO, CQO and CAO, respectively).
2469 o normal1() has a new (first) argument: lmean for the mean.
2470 o Documentation for skewnormal1() and snorm(dr).
2471
2472
2473 BUG FIXES
2474
2475 o tobit() now implements Fisher scoring properly.
2476 o Coef.vlm() needed to test for trivial constraints.
2477 o skewnorm1() had a bug in it. It has been fixed and renamed to
2478 skewnormal1().
2479 o cao() had a problem with the variable "usethiseta" when it had
2480 possible NAs.
2481 o An appropriate error message is given if residuals=TRUE in the
2482 call to @loglikelihood, for all VGAM family functions.
2483 o Two unneeded lines in rgam.f have been removed.
2484
2485
2486
2487 CHANGES IN VGAM VERSION 0.6-5
2488
2489 NEW FEATURES
2490
2491 o New functions: guplot(), meplot(), ggamma(dpqr), fff(),
2492 vonmises(), lgamma3ff, lgamma(dpqr), prentice74, tobit,
2493 zipoisson(dpqr), [dpqr]pospois(), [dpqr]laplace but there is
2494 no laplace().
2495 o cqo() has been largely rewritten. It now sports a new algorithm
2496 for ITolerances=TRUE. It can handle large data sets (e.g., 1000
2497 sites with 100 species). Compared to other cqo() options, it is
2498 the fastest. There are a few things to learn though to take full
2499 advantage of the new algorithm, e.g., centering the variables.
2500 o Windows version is cross built with R 2.2.0. The Linux version
2501 has been tested with R 2.2.0.
2502 o cao() has been largely rewritten. It now should not hang in the
2503 the windows cross build version.
2504 o .Init.Poisson.CQO() has been renamed .Init.Poisson.QO(), and also
2505 improved (however, it uses more memory by default).
2506 o Modelling functions such as vglm(), vgam() and cao() have
2507 qr.arg=FALSE now. This means object sizes can be a lot smaller.
2508 o The functions positive.poisson(), positive.binomial() etc. have
2509 been renamed pospoisson(), posbinomial() etc.
2510 o The functions [dpqr]gpd now have a location=0 argument.
2511 o Some VGAM family functions will be adapted later to use the
2512 BFGS quasi-Newton update for their working weight matrices.
2513 o The link function logoff() now works, for constant offsets.
2514 Link functions had the argument "extra"; now called "earg"
2515 to avoid confusion with the argument "extra" used in vglm() etc.
2516 Also, elogit() is new, which allows a parameter to lie between
2517 two values (A,B), say.
2518
2519
2520 BUG FIXES
2521
2522 o plotvgam() was incorrect if one of the terms (but not the first)
2523 was "x" or a function of "x" such as bs(x).
2524 o smart.expression now handles multiple 'arguments' by choosing
2525 the first, which is the smart function name.
2526 o lv(rrvglm.object) failed.
2527
2528
2529
2530 CHANGES IN VGAM VERSION 0.6-4
2531
2532 NEW FEATURES
2533
2534 o New family functions: betabin.ab(), betaprime(), dcnormal1(),
2535 erlang(), expexp(), inv.gaussianff(), maxwell(), mckaygamma2(),
2536 nakagami(), pareto1(), rayleigh(), wald().
2537 Of these, Pareto, Rayleigh and Maxwell have random number
2538 generation etc.
2539 o If criter="coef" and trace=TRUE, then the number of
2540 decimal places used to print the estimated coefficients at
2541 each iteration is proportional to the control constant
2542 epsilon.
2543 o tanl() has been named to cauchit(), and appropriate
2544 family functions reflect this change, i.e.,
2545 link="cauchit" instead of link="tanl".
2546 o size.binomial() has been improved.
2547 o Documentation for gamma1(), gamma2().
2548
2549
2550 BUG FIXES
2551
2552 o The documentation for the reverse argument in cumulative(),
2553 cratio(), etc. was incorrect.
2554 o vcov() didn't work on the windows version.
2555 o cao() still hangs under the windows version, so hopefully
2556 this bug will be fixed soon!
2557
2558
2559
2560
2561 CHANGES IN VGAM VERSION 0.6-3
2562
2563 NEW FEATURES
2564
2565 o Built with R 2.1.0 for the .zip file (Windows version) and
2566 deposited in the right directory at www.stat.auckland.ac.nz.
2567 o More documentation, e.g., fitted(), yeo.johnson(),
2568 dirmul().
2569 o zeta() and zetaff() have been improved and/or corrected.
2570 o The family functions binomial, poisson, quasibinomial,
2571 quasipoisson, gaussian, inverse.gaussian, Gamma
2572 have been withdrawn because of inteference with glm().
2573
2574
2575
2576 CHANGES IN VGAM VERSION 0.6-2
2577
2578 NEW FEATURES
2579
2580 o model.frame() and model.matrix() are roughly working for
2581 objects that inherit from "vlm"s, e.g., "vglm" objects.
2582 Both of these methods functions accept a "data"
2583 argument etc.
2584 Also, for these, smart prediction works.
2585 o A methods function for the generic function weights() has
2586 been written for VGLM objects. It returns
2587 either the prior or working weights.
2588
2589
2590 BUG FIXES
2591
2592 o The Crow1positive argument in cao() did not function correctly.
2593 o The family functions dagum, fisk, lomax, invlomax, paralogistic,
2594 invparalogistic, lognormal were not exported in the NAMESPACE
2595 file.
2596 o Functions in gaut.c and mux.c used "long" to represent integers.
2597 In R, these should be "int". Although these are equivalent on
2598 32-bit machines, they differ on 64-bit machines and crash.
2599 The files are now renamed to gautr.c and muxr.c in R.
2600 o summary(cao.object) failed.
2601
2602
2603
2604 CHANGES IN VGAM VERSION 0.6-1
2605
2606 NEW FEATURES
2607
2608 o New functions: cao() for "constrained additive ordination",
2609 and uqo() for "unconstrained quadratic ordination".
2610 Both of these are unfinished but will hopefully be
2611 completed in the forseeable future.
2612 o The function cgo() has been renamed to cqo(). Ouch!
2613 CQO stands for "constrained quadratic ordination", and is
2614 better than the old name cgo(), for
2615 canonical Gaussian ordination.
2616 o The inverse() link function has been renamed to reciprocal().
2617 o More documentation: loglinb2() and loglinb3().
2618 o zipbipp() renamed to zapoisson(), where "za" stand for
2619 "zero-altered". This is more in line with the literature.
2620 New families: zanegbin.mu, positive.negbin.mu.
2621 New random variates: rposnegbin.mu, rpospois.
2622 o negbin.mu() works now for cgo(). The subsequent methods
2623 functions have been adapted to work on it too.
2624 However, negbin.mu() is not recommended because maximum
2625 likelihood estimation of the index parameter is fraught
2626 numerically. It is better to use quasipoissonff().
2627 o cgo() now uses the function .Init.Poisson.CGO() to obtain
2628 initial values for the canonical coefficients, C.
2629 The argument Use.Init.Poisson.CGO in qrrvglm.control()
2630 now controls this feature.
2631 o Lazy loading has been enabled for the VGAM package.
2632 o Name spaces has been introduced into the VGAM package.
2633 The consequencies of this might be far reaching for
2634 code heavily based on the internals of the VGAM package.
2635 o The application of name spaces means "ff" can be dropped
2636 from certain family functions. In particular, poisson() can
2637 be used instead of poissonff(), and binomial() instead
2638 of binomialff(). Ditto for quasipoissonff() and
2639 quasibinomialff().
2640 o names.of() changed to namesof(). Many other function names
2641 have been changed, particularly those of the S3 classes
2642 such as coef. something, e.g., coef.vlm to coefvlm.
2643 In general, S3 methods functions such as print.summary.vlm have
2644 the first "." deleted, but classes such as "summary.vlm" retain the
2645 ".", and the function is printsummary.vlm.
2646
2647 BUG FIXES
2648
2649 o Some documentation regarding the negative binomial
2650 distribution was wrong.
2651 o The digamma function in FORTRAN was buggy.
2652 o gumbel.block() now handles a vector response (equivalently,
2653 a one column matrix) and the deviance has been decommissioned.
2654 Instead, the log-likelihood is computed.
2655
2656
2657
2658
2659 CHANGES IN VGAM VERSION 0.5-24
2660
2661 NEW FEATURES
2662
2663 o zipbipp() and zipoissonX() are new alternatives to yip88().
2664 They fit a zero-inflated Poisson distribution.
2665 Both can handle covariates for both parameters (p0 or
2666 phi, and lambda.)
2667 zipbipp() is recommended over the others.
2668 zipoissonX() is experimental at this stage
2669 and should be used with caution.
2670 rpospois() is new.
2671 o More documentation: rhobit and binom2.rho.
2672 o binom2.or() now has lp1 and lp2 arguments, which allow
2673 a different link function for each of the two marginal
2674 probabilities.
2675 o bratt() is a new family function. It fits the Bradley Terry
2676 model with ties.
2677 o flush.console() is used if it exists. This will make
2678 Windows version more nicer for large data sets and when
2679 trace=TRUE is used.
2680 o wweights() extracts the working weights of an object.
2681 Used to be called vweights().
2682
2683
2684
2685
2686 CHANGES IN VGAM VERSION 0.5-23
2687
2688
2689 NEW FEATURES
2690
2691 o The package works under the latest version, viz. 2.0.0.
2692 There are fewer warning messages when checking :)
2693 o persp() for CGO objects now handles Rank=1 models.
2694
2695
2696
2697 CHANGES IN VGAM VERSION 0.5-22
2698
2699 BUG FIXES
2700
2701 o plot.vgam(..., overlay=TRUE, which.cf=1:2) was incorrect.
2702
2703 NEW FEATURES
2704
2705 o demo files now are avaible for VGAM. These include lmsqreg,
2706 distributions, and cgo. More will be added later.
2707
2708
2709
2710 CHANGES IN VGAM VERSION 0.5-21
2711
2712 BUG FIXES
2713
2714 o .Rd files adapted to reflect new changes in the library names.
2715 o persp.qrrvglm() argument whichSpecies was faulty.
2716 o gpd()@inverse returned erroneous centiles.
2717 o Coef(cgo(..., FastAlgorithm=TRUE)) produced wrong results.
2718
2719 NEW FEATURES
2720
2721 o cgo(..., FastAlgorithm=TRUE) has been fined-tuned to give
2722 greater speed and accuracy.
2723 o lms.yjn() uses FORTRAN code to implement the Gauss-Legendre
2724 algorithm. This results in greater accuracy.
2725 o More documentation, especially for family functions for
2726 extreme values modelling.
2727
2728
2729
2730 CHANGES IN VGAM VERSION 0.5-20
2731
2732 BUG FIXES
2733
2734 o vglm(y ~ x, binomialff(link=tanl)) used to fail.
2735 o The CHECK command failed previously, but now it only
2736 gives 5 warnings.
2737
2738 NEW FEATURES
2739
2740 o persp.qrrvglm() has been written to apply persp() to
2741 a rank-2 CGO model.
2742 o cgo(..., FastAlgorithm=TRUE) now has a logical argument
2743 GradientFunction, which if TRUE (default), computes the
2744 derivatives by using finite-difference approximations.
2745 The default will cause the speed to generally increase.
2746
2747
2748
2749 CHANGES IN VGAM VERSION 0.5-19
2750
2751 BUG FIXES
2752
2753 o garma() did coerce the model matrix into the correct class
2754 o fisherz() could not work out the inverse.
2755
2756 NEW FEATURES
2757
2758 o trplot() is a new generic function, and for objects of
2759 class "qrrvglm" (a cgo() object), it produces a trajectory plot
2760 for species.
2761 o vcov.qrrvglm() now computes standard errors and returns the
2762 variance-covariance matrix for rank-1 QRR-VGLMs.
2763 o A new fast algorithm is implemented for cgo(..., FastAlgorithm=TRUE)
2764 which only works under windows. It is a new undocumented algorithm.
2765 o New family functions: lognormal(), lognormal3(), weibull().
2766 o New family functions: genbetaII(), betaII(), sinmad(), dagum(),
2767 lomax(), invlomax(), fisk(), invparalogistic(), paralogistic().
2768 Additionally, d*, r* p* and q* forms of the
2769 density/random-generation etc. functions for all of these
2770 except for betaII and genbetaII.
2771 o New link function for (0,1) parameters: tanl() for tan link.
2772 It has a heavier tail and corresponds to a Cauchy distribution
2773 (cf. probit for normal).
2774 o New family function: brat() for the Bradley Terry model
2775 (intercept model only).
2776
2777
2778
2779 CHANGES IN VGAM VERSION 0.5-18
2780
2781 NEW FEATURES
2782
2783 o I've changed deplot.lmscreg() so that the "at" argument is now
2784 "y.arg", and the density is returned with name "density" instead
2785 of "y". That is, "at" is now "y", and "y" is now "density".
2786 o lvplot.rrvglm() and biplot.rrvglm() have been merged and are now
2787 equivalent.
2788
2789
2790
2791 CHANGES IN VGAM VERSION 0.5-17
2792
2793 BUG FIXES
2794
2795 o Bestof argument in cgo() and rrvglm() was faulty.
2796 o Bug in plot.vgam(type.resid) fixed.
2797
2798 NEW FEATURES
2799
2800 o Updated to work under R 1.8.1
2801 o logLik() and AIC() methods functions supported for many VGAM objects.
2802 o lms.bcn.control(), lms.bcg.control(), lms.yjn.control() now
2803 have trace=TRUE because monitoring LMS quantile regression models
2804 is a good idea.
2805 o lms.bcn(), lms.bcg(), lms.yjn() now improved.
2806
2807
2808
2809 CHANGES IN VGAM VERSION 0.5-16
2810
2811 BUG FIXES
2812
2813 o biplot.rrvglm() had a internal bug with @C.
2814 o Runs under R 1.8.0 now, having a fix with "qr" slot.
2815 o etastart, coefstart, mustart arguments were not functional in vgam().
2816 o vchol() did not replace the correct elements; sometimes the index
2817 was out of subscript range.
2818 o residuals.vlm() tried to evaluate a deviance slot in a "vglmff" object
2819 even when it was empty.
2820 o Documentation links to functions in other packages now work.
2821
2822 NEW FEATURES
2823
2824 o lvplot.qrrvglm() has been renamed biplot.qrrvglm().
2825 Argument Equal.tolerances changed to EqualTolerances.
2826 Argument Circular changed to ITolerances.
2827 rrvglm.control() now split into qrrvglm.control() and itself.
2828 o cgo() now performs canonical Gaussian ordination.
2829
2830
2831
2832 CHANGES IN VGAM VERSION 0.5-15
2833
2834 BUG FIXES
2835
2836 o Coef.qrrvglm() failed wrt Equal.tolerances and Circular when
2837 Rank>2.
2838
2839 NEW FEATURES
2840
2841 o gco() is now an easier interface for fitting Gaussian canonical
2842 ordination models. gco(...) is essentially rrvglm(..., Quadratic=TRUE).
2843 o Documentation for deplot.lmscreg(), qtplot.lmscreg(), cdf.lmscreg()
2844 and related functions. Also for positive.poisson(),
2845 positive.binomial() and yip88().
2846 o lvplot.qrrvglm() improved to handle non-diagonal tolerance matrices,
2847 and a new Rotate option is available for QRR-VGLMs.
2848 o By default, QRR-VGLMs now have the constraint that the latent
2849 variables are uncorrelated and have unit variances, i.e.,
2850 their variance-covariance matrix is diag(Rank).
2851 Also, the Crow1positive argument allows ordinations to be reflected
2852 across axes.
2853
2854
2855
2856 CHANGES IN VGAM VERSION 0.5-14
2857
2858 BUG FIXES
2859
2860 o vgam() with s() terms and subset= used to give a bug because the
2861 attributes of s() were lost.
2862 o summary() of a gaussianff was faulty because control.gaussianff()
2863 was called gaussianff.control().
2864
2865 NEW FEATURES
2866
2867 o install.packages("VGAM", CRAN="http://www.stat.auckland.ac.nz/~yee")
2868 now works for PC and Linux/Unix, i.e., the distribution of the
2869 VGAM package allows for this type of download.
2870 o poissonff(), quasipoissonff(), binomialff() and
2871 quasibinomialff() now handle multiple dispersion parameters when
2872 mv=TRUE and onedpar=FALSE.
2873 o Generic function predictx(), with methods function for "qrrvglm"
2874 objects. This solves (with limited functionality) the calibration
2875 problem.
2876 o predict.qrrvglm() and predict.rrvglm() written (but don't work 100%)
2877 o Coef.rrvglm() now returns an S4 object, which can be printed nicely.
2878 o summary.qrrvglm() has been improved.
2879 o Documentation for poissonff(), quasipoissonff(), binomialff() and
2880 quasibinomialff().
2881
2882
2883
2884 CHANGES IN VGAM VERSION 0.5-13
2885
2886 BUG FIXES
2887
2888 o Code with T and F now use TRUE and FALSE.
2889
2890 NEW FEATURES
2891
2892 o Documentation for lms.bcn(), lms.bcg(), lms.yjn(), and bmi.
2893 Additionally, the overall documentation has been improved
2894 throughout.
2895 o print.Coef.qrrvglm prints the contents of Coef(qrrvglm.object)
2896 in a nicer format. It uses S4 features.
2897
2898
2899 CHANGES IN VGAM VERSION 0.5-12
2900
2901 BUG FIXES
2902
2903 o The package now works under R 1.7.1. This includes the smart
2904 prediction library.
2905 o dirichlet(), skewnorm1(), geometric(), gamma2() and erlang()
2906 had a bug that has been fixed.
2907
2908 NEW FEATURES
2909
2910 o documentation for beta2(), and dirichlet().
2911 o Easier installation; use something like
2912 "R CMD INSTALL -l ./myRlibs VGAM_0.5-12.tar.gz"
2913 for a local library.
2914
2915
2916 CHANGES IN VGAM VERSION 0.5-11
2917
2918 BUG FIXES
2919
2920 o The code has been upgraded to work under R 1.7.0 because
2921 of the calls to LAPACK and object oriented features.
2922
2923 NEW FEATURES
2924
2925 o levy() added, plus grc() documentation.
2926 o constraints added to binomialff() and poissonff() since
2927 they both handle multivariate responses.
2928
2929
2930 CHANGES IN VGAM VERSION 0.5-10
2931
2932 BUG FIXES
2933
2934 o Many univariate family functions had a faulty loglikelihood slot.
2935 o negbin.mu() was faulty causing very slow convergence.
2936 o Coef.vglm() had a bug due to "fit" rather than "object"
2937
2938 NEW FEATURES
2939
2940 o logff() added.
2941 o The undocumented backchat facility now works for Splus 6.x.
2942 This should increase the efficiency of vglm() in particular.
2943 Thanks to Insightful and Dr J. Chambers for helping to get it
2944 going under the S4 engine.
2945
2946
2947
2948 CHANGES IN VGAM VERSION 0.5-9
2949
2950 BUG FIXES
2951
2952 o binomialff() had a bug in @weight.
2953 o binomialff(mv=T) used to fail.
2954 o gev(), ogev() and egev() had @loglikelihood that was faulty.
2955
2956 NEW FEATURES
2957
2958 o .Rd documentation included for vglm(), vgam(), rrvglm(), and
2959 associated control and class functions, plus smart prediction.
2960
2961
2962
2963 CHANGES IN VGAM VERSION 0.5-8
2964
2965 NEW FEATURES
2966
2967 o rrvglm() now has a Quadratic argument to implement the class of
2968 Quadratic Reduced-rank VGLMs, which gives maximum likelihood
2969 solutions to Gaussian canonical ordination problems.
2970 Documentation is in rrvglm.pdf
2971
2972
2973
2974 CHANGES IN VGAM VERSION 0.5-7
2975
2976 NEW FEATURES
2977
2978 o vglm() now has a xij argument which implements eta-specific
2979 covariates. Documentation is supplied on the VGAM website.
2980 o grc() has been written for Goodman's RC association model for a
2981 contingency table. Documentation is in rrvglm.pdf
2982
2983
2984