Fix broken bib file.
Adding scores for discrete mtram
models.
Update mtram
to internal interface changes.
Use mltoptim
also in mtram
.
Ship proper mtram
vignette (so far only the rendered output
was supplied).
Make sure demos are in shape (again).
Gradient check failed on Fedora noLD.
Gradient check failed on MacOS.
mmlt
moved to mlt 1.6-0.
New Mmlt
allowing arbitrary mixes of continuous
and discrete data, including exact and censored observations
for the same variable; implementing three (potentially) convex
approximations; allowing for non-informative missings encoded as
-Inf, Inf
.
Compris
now also uses the parametric likelihood for
the competing risk case.
Allow density predictions for all continuous variables (regardless of censoring in training data).
Add example for Compris
.
Correct simulate.mmlt
.
Change to R\_Calloc
and friends.
Allow larger then needed args$w
argument in mmlt
.
Fix bug in predict.mmlt
.
Add sequential fits to mmlt
.
Add fixed = TRUE/FALSE
argument to coef.mmlt
such that the whole parameter vector can be retrieved.
Handle constraints for fixed parameters in mmlt
correctly (as mlt
always did).
tram
and Polr
allow binary responses (and return
results identical to glm
with appropriate family
).
New demo("survtram")
and demo("hcc")
.
predict
methods for tram
and stram
now offer the same type
argument as
mlt::predict.ctm
offers.
Allow shift-scale transformation models in mmlt
.
Add nonparametric log-likelihood to Compris
.
Improved, simplified, and generalised implementation of mmlt
.
Coxph
, Colr
, and BoxCox
allow unbiased
estimation of time-to-event distributions under dependent
censoring.
All tram
models can be equipped with frailty terms.
Fix some minor issues.
Update demo stram
.
Add simulate.mmlt
.
Survreg
didn't work with constraints, reported by Lucas Kook.
mtram
didn't deal properly with interval censoring in all cases.
Enable stratification for shift-scale transformation models (still experimental).
Fix S3 argument issues.
Try to detect negative SEs.
Stabilise and generalise mmlt
: Alternative link functions,
separate lower triangular matrix storage and computations from
mmlt
.
coef(<mmlt>, type = "Sigma")
and coef(<mmlt>, type = "Corr")
were incorrect for more than three dimensions.
Improved numerical stability in mtram
.
New location-scale functionality, see demo("stram")
.
New Laplace link function.
New cauchit link function.
Minor bug fixed for shift-scale transformation models (still experimental).
bread.tram
added for sandwich
to work.
Update mtram
vignette; move code to demo.
Add shift-scale transformation models (experimental).
Add methods for probabilistic indices and overlap coefficients + friends.
Export Gradient
and Hessian
methods.
Simplify score_test
and profile
: Use fixed
instead of offset
.
Models with standardisation now available in mtram
for exact
continuous outcomes and link functions other than probit.
Remove arguments ignored by mlt
.
Move mtram vignette code to demo as it took too long on Windows for CRAN.
Deal with stratum - covariate interactions, overparameterised coefficients are constraint to zero.
Make sure to use the Schur complement when computing
variances in for linear transformation models in
score_test
and perm_test
.
Fix LaTeX problems.
Add dofit
argument to mmlt
.
Robustify vcov
for mmlt
objects.
Add experimental score_test
and perm_test
methods
for glm
.
Add simulate
method.
Enable nonparametric likelihood estimation by Polr
.
New argument sparse_nlevels
to tram
, requesting
a model Matrix
for ordered factors with at least
sparse_nlevels
levels.
Add experimental score_test
and perm_test
for
survival::coxph
.
Update simulation code.
Robustify computation of score intervals.
Add Spearman correlation to mmlt
example.
Lehmann()
parameter is now labelled as proportional reverse time hazard ratio.
Add mmlt
for multivariate conditional transformation models
(continuous responses only for now).
Reproducibility material for <arxiv:1906.03151> added to package.
Force support to start at time zero and constrain cumulative
hazard to zero for time zero in Aareg
.
Add Aareg
for Aalen additive hazard models.
score_test
generic and method added for computing
score tests and corresponding confidence intervals.
perm_test
generic and method added for computing
permutation score tests.
PI
converts log-odds ratios to probabilistic indices and
vice versa.
profile
method added for computing
profile likelihood confidence intervals.
Allow additional linear inequality constraints for regression coefficients.
Deal with model_only
in Survreg
, spotted by Balint Tamasi.
Fix problem in C header file.
Fix problems with responses of class R
, spotted by Balint Tamasi.
y | s1:s2 ~ x
lead to rank-deficit models, the intercept is
now removed.
Not really a bug, but time-consuming examples were removed (or replaced by static output) in package vignettes to make CRAN happy.
as.mlt
returned an object of class mlt
but not
mlt_fit
.
New marginally interpretable models for clustered data
in mtram
, see mtram
vignette. This is still
experimental and very rough, so please be careful.
Compute vcov.tram
via Schur complement when possible.
Add regression test for count data and log_first = TRUE
.
tram
objects are now in line with multcomp::glht
,
allowing post-hoc multiple comparisons.
log_first = TRUE
is now accepted by tram
and all
model functions calling tram
. This is still experimental.
Silence regression tests on Windows.
New extrapolate = TRUE
option for all models constraints
the second derivate of the Bernstein transformation function to be
zero at support
. This helps to get rid of possible
discontinuous densities at support
.
Add Lehmann
for fitting shift transformation models those
linear predictor can be interpreted as log-Lehmann-alternative.
Add method = "loglog"
to Polr
.
New update
method.
Use default bounds = c(0, Inf)
for Surv
objects (for example
in Coxph
.
Update to mlt 1.0-2.
Likelihood ratio tests with fixed parameters gave an error.
Remove strata from linear predictor (experimental).
Try to implement a better guess for the add
argument
later defining the axes corresponding to the response
in plots.
All models accept add
and bounds
arguments.
tram first published on CRAN.