** denotes quite substantial/important changes
*** denotes really big changes
Liable to change in future release:
- method="GCV.Cp" as default in gam. Default will
change to "REML". (1.8-24, June 2018)
Currently deprecated and liable to be removed:
- gam.fit and full.score (1.9-0)
Issues:
* openblas 0.3.x x<7 is not thread safe if itself compiled for single thread
use and then called from multiple threads (unlike the reference BLAS, say).
0.2.20 appears to be OK. For 0.3.x x>6 make USE_THREAD=0 USE_LOCKING=1
to make openblas ensures thread safety.
* t2 in bam(...,discrete=TRUE) - not treated as tensor products at
present, and reparameterization needs checking (also for bam).
* bam(...,discrete=TRUE) has no option to use the identical
discretizaton to that used in fitting when predicting
* efsud Vb every time? Fix bam NCV variance comp, to be as gam.
1.9-2
** 'NCV' added as a method option for bam(...,discrete=TRUE). More efficient
than gam(...,method='NCV'), but much less efficient than default REML.
* default margins changed for plot.gam(...,pages=1)
* 'clog' censored logistic family added for AFT models. Written by Chris Shen.
* improved cov matrix estimation under NCV used with correlated errors.
* names in NCV 'nei' structure modified to better match paper and documentation
notation.
* estimate.theta improved for semi-definite case (e.g. infinite DoF in scat).
* extended families now give a useful error if an illegal link is supplied.
* pcls now returns indices of active set constraints as attribute of return
vector.
* fix bam to always return a square QR R factor of the model matrix
(previously it was possible for it not to be square for a rank deficient
model matrix, which could cause a failure in summary). Thanks to Amy Hughes
* Fixed bam/bam.update to pick up weights correctly, rather than failing.
Thanks to Max Moldovan.
* Added sparse matrix Cholesky routine, mchol, a wrapper for
Matrix::Cholesky, giving the same functionality as Matrix::chol prior
to the decision for this not to return the pivot sequence.
* 'mroot' modified so that svd option uses an efficient symmetric eigen routine
to compute SVD (argument is positive semi-definite after all).
* fixed gam.vcomp to avoid failure with extended families and bam. Thanks to
Nathan LaSpina for reporting.
* modification of discrete bam prediction method function to better deal with
extra factor levels in prediction. Slightly differs from predict.gam, in
setting effect to zero, rather than NA.
* fix to NCV handling of dropped NA points - previously could end up with
different numbers of predict and drop neighbourhoods. Thanks to Nicole
Augustin.
* XWXd fix for row col interchange bug introduced with long vector support in
1.9-1.
* fix of gumbls initializtion code bug. Initial parameters were so bad that
fit could fail.
* zoom in bfgs modified so that start values always the same to avoid
occasional step failures from repeatedly changing initial values.
* gevlss improved xi initialization.
* gam(...,discrete=TRUE) no longer ignores starting coefficient values (thanks
to Matteo Fasiolo).
* Sl.setup and ldetS changes to return totla penalty matrices directly
(avoiding a redundant crossproduct)
* Fix to Pearson residuals in gammals family - thanks to gavin Simpson and
Arthur Fendrich
* ldetS slight efficiency improvement by direct matrix diagonal modification
in C code, as suggested by Matteo Fasiolo.
1.9-1
* Revised bam discrete code to allow indexing vectors to be long vectors,
thereby allowing much larger datasets to be used. n limit is still
.Machine$integer.max.
* Revised C code underlying mvn family to allow larger problems, via use of
R long vectors.
* Revised gfam to work correctly with bam(...,discrete=FALSE). Subsetting
in bgam.fit was incorrect for gfam. Thanks to Dave Miller for reporting it.
* Correction of ziplss residual computation - incorrect probability of non
zero used. Thanks to Xiao Liang.
* correction of ginla integration weights (correcting typo from Rue et al.
2009). Thanks to Paul Van Dam-Bates.
* C code modification to remove some new compiler warnings.
1.9-0
*** NCV smoothing parameter estimation now available for most models. See ?NCV.
*** 'gfam' allows reponse variables to be from several different families.
* Restriction that number of coefficients must be fewer than number of data
removed.
* Fix to tensor product constructor to allow matrix factor arguments to be
handled correctly (previously failed). Thanks to Dave Miller.
* removed argument 'pers' from 'plot.gam' (deprecated since 1.8-23, Nov 2017).
* removed "nlm.fd" optimizer (deprecated since 1.8-19, Sept 2017).
* removed discrete method prediction from discrete bam objects fitted
before 1.8-32 (deprecated since then).
* 'multinom' variance fix for K>=3 cases. A counter was initialized in the wrong
place. Thanks to Max Goplerud for finding and reporting this.
* NCV variance estimation improvements + na handling.
* 'in.out' argument added for 'bam'.
* slight internal change to mgcv:::get.var to enable user defined functions
in arguments of smooths to work.
* smooth2random tweak to guarantee same eigenvector sign everytime for same
problem (needed by some developers who don't want to carry the original
transform around).
* Some unregistered methods fixes.
* tests directory removed, as check time was tripping up every CRAN submission.
* coxph memory error fix - could make bfgs fitting fail.
1.8-42
* One remaining old style C declaration fixed.
1.8-41
** 'cnorm' family added for left, right, interval or un censored Gaussian
data. Useful for log normal Accelerated Failure Time models, Tobit
regression, rounded data etc.
** 'sz' factor smooth interaction class added for implementing models with
main effect smooths and difference smooths for levels of a factor.
See ?factor.smooth.
** 'NCV' smoothing parameter method added, but still experimental.
* replacment of some old (K&R) style C function declarations.
* mono.con corrected for cases with upper bounds. Thanks to Sean Wu.
* plot.gam(...,seWithMean=TRUE) modified to only include mean uncertainty
for the linear predictor of which the smooth is a part, when there are
multiple linear predictors. Thanks to Gavin Simpson.
* modifications of sparce matrix coercions, to avoid deprecated direct
coercions. as(as(as(a, "dMatrix"), "generalMatrix"), "XsparseMatrix") in
place of as(a,"dgXMatrix") where 'X' is 'C' or 'T'. Actually this requires
Matrix 1.4-2 to work (will be added to dependencies in future).
* vis.gam now deals properly with models with more than one linear predictor.
* slight change in bgam.fitd to check scale parameter estimate converged when
using bam(...,discrete=TRUE), otherwise scale could be wrong for all fixed
smoothing parameters.
* predict.gam modified so that 'terms' and 'exclude' control all terms, smooth
or parametric, in the same way. Including the "(Intercept)" term.
* Warnings from 'model.matrix' suppressed in 'terms2tensor' called by e.g.
predict.bamd. There is a warning if any extra contrasts are supplied to
model matrix that do not relate to a term in the model (which contradicts
the documentation). Doc vs code bug report also filed.
* Fix of broken rank deficiency handling in gam.fit5. Thanks to Cesko Voeten.
* trind.generator modified to allow return of index functions in place
of index arrays.
* summary.gam (recov/reTest) modified to deal with 'fs' smooths fitted using
'gamm'.
* gam.fit4 convergence testing improved and bug fix in computation of dVkk
matrix used to check for converged 'infinite' smoothing parameters in bfgs.
1.8-40
* Small gam.fit5 convergence change to reduce chance of repeating step
failures.
* ExtractData fix avoiding assumption that 'xt' argument to a smooth is always
a list (could cause setup failure with e.g. fs smooth). Thanks Keith Woolner.
* Minor changes to mat.c and tprs.c to allow DEFS = -DSTRICT_R_HEADERS build.
1.8-39
* gam.fit5 convergence logic tightened up, to avoid pointlessly lengthy
iterations when can't meet convergence tolerance. Faster, more reliable,
but also generates more warnings when tolerances not met.
* gam.fit3 convergence test modified to more reliable version.
* Some modification of warning handling to only print inner optimization
warnings if they occur at final call to inner optimizer.
* newton and bfgs optimizers reset inner loop tolerance to 1e-2 of outer
loop tolerance if it is larger than this, to avoid inner loop being too
inaccurate for outer tolerance.
* minor R 4.2.0 C compatibility change.
1.8-38
* uniquecombs fix to reduce memory footprint for text data.
1.8-37
* minor update for clang compatibility.
* PredictMat fix of bug in which dimensions could be dropped for 1 row matrix
as part of constraint handling (leading to disaster in predict.bamd).
Thanks to Shawn Ligocki.
1.8-36
* 'fs' smooth construction modified to (approximately) orthogonalise
smoothing penalty penalized and unpenalized bases. This makes the
assumption that the associated variance components are independent
more natural. Thanks to Matteo Fasiolo and Harald Bayaan for showing
how the original construction could be problematic.
* correction to 'fs' smooth handling in gamm to allow correlation structures
to be used in the same model without cuasing an error.
* AIC calculation modified for non-Gaussian families with a scale parameter,
so that the model estimated scale parameter is used, rather than the
deviance estimator employed with glm exponential families. This matters for
avoiding substantial bias with very low mean count data (particulalry 'tw'
and 'Tweedie' families).
* ziP family response scale se fix for case when b>0 (thanks to Mark Donoghoe)
* prediction on response scale could fail for models fitted by
bam(...,discrete=TRUE) - fixed.
* bug fix to multi-model anova.gam code to deal properly with extended family
scale parameters (still not the recommended way of testing!)
1.8-35
* Fix to bug in Fletcher scale parameter estimate with weighted data (e.g.
quasibinomial with n>1). Gaussian case not affected. Thanks to John
Maindonald.
* Italian translation updated thanks to Daniele Medri
* German translation updated thanks to Detlef Steuer
* French translation updated thanks to Philippe Grosjean
* gumbls initialization improved.
* Modification so that smooths can insist on not being reparameterized when
called using default bam methods, by having a 'repara' element set to FALSE
in the object returned by the smooth constructor.
* Modification so that smooths can have the summation convention for matrix
arguments turned off, by having an 'xt$sumConv' element set to FALSE in the
'xt' argument of 's'.
1.8-34
* 'gam.mh' proposal df typo fix and help file '%*%' -> '\%*\*' fix. Thanks
Len Thomas.
* various random number generation calls in rd functions in families could
fail if simulating one point - fixed.
* gp smooths can now be specified to be strictly stationary (see ?gp.smooth)
* fix to predict.bamd bug in handling fixed effect contrasts (wrong contrast
could be used for prediction, if non-default used in fitting). Thanks to
Jalal Al-Tamimi.
* predict.bamd modified to catch case where all prediction data is NA
* predict.gam modified to allow NA to be a factor level.
1.8-33
* 'inline' -> 'static inline' error fix - caused installation failure on some
platforms.
1.8-32
** bam(...,discrete=TRUE) now uses discretization on the parametric model
components as well as the smooths.
* 'gumbls' family added for Gumbel location scale models.
* 'shash' location scale and shape family added.
* psum.chisq function added to compute c.d.f. of weighted sums of chi-squared
r.v.s using method of Davies, 1980. This is now used for computing p-values
in summary.gam in place of the Liu et al, 2009 approximation.
* score and Schoenfeld residuals added for cox.ph to facilitate PH assumption
checking (see ?cox.ph).
* added 'gam.mh' for posterior dampling from models fitted with 'gam'.
* 'ocat' upgraded to not ignore weights.
* added multivariate t and Gaussian densities and multivariate t generation.
* fix to predict.bamd to correct handling of `terms' and `exclude' arguments.
* A recent lme change broke gamm offset handling. Fixed.
Thanks Aaron Benjamin Shev.
* dgesvd LAPACK calls replaced by faster dgesdd, as some versions of MKL
BLAS/LAPACK have broken dgesvd.
* Fix to derivative w.r.t. scale parameter calculation for extended families
('tw' is the only one at present) when using 'ML' smoothing parameter
estimation. Could lead to step failure with 'tw'. Thanks Kevin Hawkshaw.
* Fix to prediction/fake formula environment that could cause failure of
gam/bam called from a function. Thanks to Duncan Murdoch.
* Fix to smoothing parameter uncertainy correction for general families.
* Much cleaner design for the discretization of covariates and associated
discretization index matching in bam (predict.bamd and discrete.mf).
* Fix to jagam logic for identifying separable penalties which led to
wrong partioning of penalties for 'fs' smooths used with jagam. Thanks to
Chris Jackson for spotting the problem.
* Internal Sl list handling changes.
* tensor.prod.model.matrix now also works with sparse matrices of class
"dgCMatrix".
1.8-31
* fix in initalization in gammPQL
* fix of some C routines of type void in place of SEXP called by .Call.
1.8-30
* anova.gam now uses GLRT for multi-model comparisons in both extended and
general family cases.
* Fix to bug in bam(...,discrete=TRUE) offset handling introduced in 1.8-29,
which corrupted offset (and generated numerous warnings). Also fixes a
less obvious bug introduced at the same time in predict.gam which could get
the offset wrong when predicting with such models. Thanks to Brian
Montgomery and Sean Wilson.
* Fix to problem in pen.reg (used to initialize location scale models in
particular), which could lead to initialization failure for lightly
penalized models. Thanks Matteo Fasiolo.
* Fix to predict.bamd handling of 'terms' and 'exclude' for models fit by
bam(...,discrete=TRUE).
* Work around in predict.gam for a spurious model.matrix warning when a
contrasts.arg relates to a variable in data not required by object.
* Fix to gammals family which had 'd2link' etc defined with argument 'eta'
instead of the intended 'mu'. Thanks to Jim Stagge.
* 'in.out' modified to allow boundaries specified exactly as for a soap film
smoother.
* soap film smoother constructor modified to check knots are within boundary
on entry and drop those that are not, with a warning. Also halts if it
detects that basis setup is catastrophically ill-conditioned.
1.8-29
* gammPQL modified to use standard GLM IRLS initilization (rather than
glmmPQL method) to improve convergence of `gamm' fits.
* bam(...,discrete=TRUE) now drops rownames from parametric model matrix
components, to save substantial memory.
* All BLAS/LAPACK calls from C now explicitly pass hidden string length
arguments to avoid breakage by recent gfortran optimizations (stack
corruption causing BLAS/LAPACK to return error code).
* predict.gam bug fix - parameteric interaction terms could be dropped for
type="terms" if there were no smooths. knock on was that they were also
dropped for all bam(...,discrete=TRUE) fits. (Thanks Justin Davis.)
* bam(...,discrete=TRUE) indexing bug in setup meant that models containing
smooths with matrix arguments and other smooths with factor by variable
would fail at the setup stage.
* gam.fit4 initial divergence bug fix.
* Gamma location-scale family 'gammals' added. See ?gammals.
* row-wise Kronecker product operator %.% added for convenience.
* changes to general families to allow return of first deriv of penalized
Hessian component more easily.
* ocat bug fix. Response scale prediction was wrong - it ignored the estimated
thresholds. Thanks to Fabian Scheipl.
* bam deviance could be wrongly returned, leading to 100% explained
deviance. Fixed.
1.8-28
* fix of obscure sp naming bug.
* changed some contour default colours from green to blue (they overlay
heatmaps, so green was not clever).
* Tweedie likelihood evaluation code made slightly more robust - for a model
with machine zero scale parameter estimate it could segfault, as series
maximum location could then overflow integer storage. Fixed + upper limit
imposed on series length (warning if it's not enough).
1.8-27
** Added routine 'ginla' for fully Bayesian inference based on an integrated
nested Laplace approximation (INLA) approach. See ?ginla.
* Tweedie location scale family added: 'twlss'.
* gam.fit5 modified to distinguish more carefully between +ve semi definite
and +ve definite. Previously could fail, claiming indefiniteness when it
should not have. Affects general families.
* bam was ignoring supplied scale parameter in extended family cases - fixed.
* work around in list formula handling for reformulate sometimes setting
response to a name in place of a call.
* preinitialize in general families is now a function, not an expression. See
cox.ph for an example.
* added routine cholup for rank one modification of Cholesky factor.
* two stage gam/bam fitting now allows 'sp' to be modified.
* predict.gam could fail with type="response" for families requiring the
response to be provided in this case (e.g. cox.ph). Fixed.
* sp.vcov defaults to extracting edge corrected log sp cov matrix, if
gam(...,gam.control(edge.control=TRUE)) used for fitting.
* gam(...,gam.control(edge.correct=TRUE)) could go into infinite loop if
sp was effectively zero. Corrected.
1.8-26
* LINPACK dependency removed.
* Added service routine choldrop to down date a Cholesky factor on row/col
deletion.
* liu2 had a check messed up when vectorized. Fix to stop vector being
checked for equality to zero.
1.8-25
** bam(...,discrete=TRUE) methods improved. Cross products now usually faster
(can be much faster) and code can now make better use of optimised BLAS.
* fix to 'fs' smooth.construct method and smooth2random method, to allow
constructor to be called without a "gamm" atribute set on the smooth spec
but still get a sensible result from smooth2random (albeit never using
sarse matrices). Useful for other packages using constructors and
smooth2random, for 'fs' smooths.
* The mrf smooth constructor contained an obsolete hack in which the term
dimension was set to 2 to avoid plotting when used as a te marginal. This
messed up side constraints for terms where a mrf smooth was a main effect
and te marginal. Fixed.
* extract.lme.cov/2 documentation modified to cover NA handling properly, and
routines modified to not require data to be supplied.
* fix of efsudr bug whereby extended families with no extra parameters to
estimate could give incorrect results when using optimer="efs" in 'gam'.
* negbin() corrected - it was declaring the log link to be canonical, leading
to poor convergence and slight misfit.
* predict.bam(...,discrete=TRUE) now handles na.action properly, rather than
always dropping NAs.
* Fix of very obscure bug in which very poor model of small dataset could
end up with fewer `good' data than coefs, breaking an assumption of C code
and segfaulting.
* fix of null deviance computation bug introduced with extended families in
bam - null deviance was wrong for non-default methods.
* liu2 modified to deal with random effects estimated to be exactly 0, so
that summary.gam does not fail in this case.
1.8-24
* Extended Fellner Schall optimizer now avaialable for all families with 'gam'
using gam(...,optimizer="efs").
* Change to default behaviour of plot.gam when 'seWithMean=TRUE', and of
predict.gam when 'type="iterms"'. The extra uncertainty added to CIs or
standard errors now reflects the uncertainty in the mean in all other model
terms, not just the uncertanity in the mean of the fixed effects as before.
See ?plot.gam and ?predict.gam (including for how to get the old behaviour).
* 're' smooths can now accept matrix arguments: see ?linear.functional.terms.
* cox.ph now allows an offset to be provided.
* Fix in smoothCon for bug in case in which only a single coefficient is
involved in a sum-to-zero constraint. Could cause failure in e.g. t2 with
cc marginal.
* Model terms s, te etc are now always evaluated in mgcv workspace explicitly
to avoid masking problems in obscure circumstances.
* 'mrf' smooth documentation modified to make it clearer how to specify 'nb',
and code modified so that it is now possible to specify the neighbour
structure using names rather than indices.
* 'bfgs' fix to handle extended families.
* plot.gam modified to only prompt (via devAskNewPage) for a new page after
the first page is used up.
* export 'k.check'.
* Fix to 'Rrank'. Previously a matrix R with more columns than rows could
cause a segfault.
* Fix to non-finite likelihood handling in gam.fit5.
* Fix in bgam.fitd to step reduce under indefinite deviance and to ensure
penalty evaluation is round off negative proof.
* newton slighty modified to avoid (small) chance of all sp's being dropped
for indef likelihood.
1.8-23
* default plot methods added for smooths of 3 and 4 variables.
* The `gamma' control parameter for gam and bam can now be used with RE/ML
smoothness selection, not just GCV/AIC. Essentially smoothing parameters
are chosen as if the sample size was n/gamma instead of n.
* The "bs" basis now allows multiple penalties of different orders on the same
spline. e.g. s(x,bs="bs",m=c(3,2,0)). See ?b.spline.
* bam(...,discrete=TRUE) can now use the smooth 'id' mechanism to link
smoothing parameters, but note the method constraint that the linked bases
are not forced to be identical in this case (unlike other fitting methods).
* summary.gam now allows random effects tests to be skipped (in some large
models the test is costly and uninteresting).
* 'interpret.gam0' modified so that masked 's', 'te', etc from other packages
does not cause failure.
* coxph fix of prediction bug introduced with stratified model (thanks
Giampiero Marra)
* bam(...,discrete=TRUE) fix to handle nested matrix arguments to smooths.
* bam(...,discrete=TRUE) fix to by variable handling with fs and re smooths
which could fail during re-representation as tensor smooths (for
discretization purposes).
* bam extended family extension had introduced a bug in null deviance
computation for Gaussian additive case when using methods other than fREML
or GCV.Cp. Fixed.
* bam(...,discrete=TRUE) now defaults to discrete=FALSE if there are no
smooths, rather than failing.
* bam was reporting wrong family under some smoothing parameter selection
methods (not default).
* null deviance computation improved for extended families. Previous version
used an approximation valid for most families, and corrected the rest - now
replaced with exact computations for all cases.
* scat initialization tweaked to avoid -ve def problems at start.
* paraPen handling in bam was broken - fixed.
* slight adjustment to sp initialization for extended families - use observed
information in weights if possible.
1.8-22
* Fix of bug whereby testing for OpenMP and nthreads>1 in bam, would fail if
OpenMP was missing.
1.8-21
* When functions were added to families within mgcv some very large
environments could end up attached to those functions, for no good reason.
The problem originated from the dispatch of the generic 'fix.family.link'
and then propagated via fix.family.var and fix.family.ls. This is now avoided,
resulting in smaller gam objects on disk and lower R memory usage.
Thanks to Niels Richard Hansen for uncovering this.
* Another bug fix for prediction from discrete fit bam models with an offset,
this time when there were more than 50000 data. Also fix to bam fitting when
the number of data was an integer multiple of the chunk size + 1.
* check.term was missing a 'stop' so that some unhandled nesting structures
in bam(...,discrete=TRUE) failed with an unhelpful error, instead of a
helpful one. Fixed.
1.8-20
* bam(,discrete=TRUE) could produce garbage with ti(x,z,k=c(6,5),mc=c(T,F))
because tensor re-ordering for efficiency failed to re-order mc (this is
a *very* specialist bug!). Thanks to Fabian Scheipl.
* plot(...,residuals=TRUE) weighted the working residuals by the sqrt working
weights divided by the mean sqrt working weight. The standardization by the
mean sqrt weight was non standard and has been removed.
* Fix to bad bug in bam(...,discrete=TRUE) offset handling, and predict.bamd
modified to avoid failure predicting with offset. Thanks to Paul Shearer.
* fix of typo in bgam.fit, which caused failure of extended families
when dataset larger than chunk size. Thanks Martijn Wieling.
* bam(...,discrete=TRUE)/bgam.fitd modified to use fisher weights with
extended families if rho!=0.
1.8-19
** bam() now accepts extended families (i.e. nb, tw, ocat etc)
* cox.ph now allows stratification (i.e. baseline hazard can differ between
groups).
* Example code for matched case control in ?cox.ph was just plain wrong. Now
fixed. Thanks to Patrick Farrell.
* bam(...,discrete=TRUE) tolerance for judging whether smoothing parameters
are on boundary was far too low, so that sps could become so large that
numerical instability set in. Fixed. Thanks to Paul Rosenfield.
* p.type!=0 removed in summary.gam (previously deprecated)
* single penalty tensor product smooths removed (previously deprecated).
* gam(...,optimizer="perf") deprecated.
* extra divergence check added to bam gam default gam fitting (similar to
discrete method).
* preinitialize and postproc components of extended families are now functions,
not expressions.
* coefficient divergence check was missing in bam(...,discrete=TRUE) release
code - now fixed.
* gaulss family link derivatives modified to avoid overflow. Thanks to
Kristen Beck for reporting the problem.
* redundant 'n' argument removed from extended family 'ls' functions.
* Convergence checking can step fail earlier in fast.REML.fit. If trial step
is no improvement and equal to previous best (to within a tolerance), then
terminate with step failure after a few step halvings if situation persists.
Thanks to Zheyuan Li for reporting problem.
1.8-18
* Tweak to 'newton' to further reduce chance of false convergence at
indefinite point.
* Fix to bam.update to deal with NAs in response.
* 'scat' family now takes a 'min.df' argument which defaults to 3. Could
otherwise occasionally have indefinite LAML problems as df headed towards 2.
* Fix to `gam.fit4' where in rare circumstances the PIRLS iteration could
finish at an indefinite point, spoiling implicit differentiation.
* `gam.check' modified to fix a couple of issues with `gamm' fitted models, and
to warn that interpretability is reduced for such models.
* `qq.gam' default method slight modification to default generation of reference
quantiles. In theory previous method could cause a problem if enough
residuals were exactly equal.
* Fix to `plot.mrf.smooth' to deal with models with by variables.
* `plot.gam' fix to handling of plot limits when using 'trans' (from 1.8-16
'trans' could be applied twice).
* `plot.gam' argument 'rug' now defaults to 'NULL' corresponding to 'rug=TRUE'
if the number of data is <= 10000 and 'rug=FALSE' otherwise.
* bam(...,discrete=TRUE) could fail if NAs in the smooth terms caused data rows
to be dropped which led to parametric term factors having unused levels
(which were then not being dropped). Fixed (in discrete.mf).
* bam(...,discrete=TRUE,nthreads=n) now warns if n>1 and openMP is not
available on the platform being used.
* Sl.addS modified to use C code for some otherwise very slow matrix
subset and addition ops which could become rate limiting for
bam(...,discrete=TRUE).
* Parallel solves in Sl.iftChol can speed up bam(...,discrete=TRUE) with
large numbers of smoothing/variance parameters.
* 'gamm' now warns if called with extended families.
* disasterous 'te' in place of 'ti' typo in ?smooth.terms fixed thanks to
John McKinlay.
* Some `internal' functions exported to facilitate quantile gam methods
in separate package.
* Minor fix in gam.fit5 - 1 by 1 matrix coerced to scalar, to prevent failure
in some circumstances.
1.8-17
* Export gamlss.etamu, gamlss.gH and trind.generator to facilitate user
addition of new location-scale families.
* Re-ordering of initialization in gam.fit4 to avoid possible failure of
dev.resids call before initialization.
* trap in fast.REML.fit for situation in which all smoothing parameters
satisfy conditions for indefinite convergence on entry, with an
immediate warning that this probably indicates iteration divergence (of bam).
* "bs" basis modified to allow easier control of the interval over which the
spline penalty applies which in turn allows more sensible control of
extrapolation behaviour, when this is unavoidable.
* Fix in uniquecombs - revised faster code (from 1.8-13) could occasionally
generate false matches between different input combinations for integer
variables or factors. Thanks to Rohan Sadler for reporting the issue that
uncovered this.
* A very bad initial model for uninformative data could lead to a negative
fletcher estimate of the scale parameter and optimizer failure - fixed.
* "fREML" allowed in sp.vcov so that it works with bam fitted models.
* 2 occurances of 'return' replaced by (correct) return().
1.8-16
* slightly improved intial value heuristics for overlapping penalties in
general family case.
* 'ocat' checks that response is numeric.
* plot.gam(...,scale=-1) now changes scale according to 'trans' and 'shift'.
* newton optimizer made slightly more cautious: contracts step if reduction
in true objective too far different from reduction predicted by quadratic
approximation underlying Newton step. Also leaves parameters unchanged
in Newton step while their grad is less than 1% of max grad.
* Fix to Fisher weight computation in gam.fit4. Previously a weight could
(rarely) evaluate as negative machine prec instead of zero, get passed to
gdi2 in C code, generate a NaN when square rooted, resulting in a NaN passed
to the LAPACK dgeqp3 routine, which then hung in a non-interuptable way.
* Fix of 'sp' argument handling with multiple formulae. Allocation to terms
could be incorrect.
* Option 'edge.correct' added to 'gam.control' to allow better correction
of edge of smoothing parameter space effects with 'gam' when RE/ML used.
* Fix to setting of penalty rank in smooth.construct.mrf.smooth.spec.
Previously this was wrong, which could cause failure with gamm if the
penalty was rank deficient. Thanks Paul Buerkner.
* Fix to Vb.corr call from gam.fit3.post.proc to ensure that sp not
dropped (wrongly treated as scale estimate) when P-REML or P-ML used.
Could cause failure depending on BLAS. Thanks Matteo Fasiolo.
* Fix in gam.outer that caused failure with "efs" optimizer and fixed sps.
* Fix to `get.var' to drop matrix attributes of 1 column matrix variables.
* Extra argument added to `uniquecombs' to allow result to have same row
ordering regardless of input data ordering. Now used by smooth constructors
that subsample unique covariate values during basis setup to ensure
invariance to data re-ordering.
* Correction of scaling error in spherical correlation structure GP smooth.
* qf and rd functions for binomial family fixed for zero n case.
1.8-15
* Fix of survival function prediction in cox.ph family. Code used expression
(8.8.5) in Klein and Moeschberger (2003), which is missing a term. Correct
expression is, e.g., (10) from Andersen, Weis Bentzon and Klein (1996)
Scandinavian Journal of Statistics.
* Added help file 'cox.pht' for Cox PH regression with time dependent
covariates.
* fix of potential seg fault in gdi.c:get_bSb if single smooth model
rank deficient (insufficient workspace allocated).
* gam.fit5 modified to step half if trial penalized likelihood is infinite.
* Fix so that bam works properly with drop.intercept=TRUE.
1.8-14
* bug fix to smoothCon that could generate NAs in model matrix when using bam
with numeric by variables. The problem was introduced as part of the
bam(...,discrete=TRUE) coding.
1.8-13
* Added help file ?one.se.rule on the `one standard error rule' for obtaining
smoother models.
* bam(...,discrete=TRUE) no longer complains about more coefficients than data.
* 's', 'te', 'ti' and 't2' modified to allow user to specify that the smooth
should pass through zero at a specified point. See ?identifiability.
* anova.gam modified to use more appropriate reference degrees of freedom
for multiple model call, where possible. Also fixed to allow multiple
formulae models and to use -2*logLik in place of `deviance' for
general.family models.
* offsets allowed with multinomial, ziplss and gaulss families.
* gevlss family implementing generalized extreme value location, scale and
shape models.
* Faster code used in 'uniquecombs'. Speeds up discretization step in
'bam(...,discrete=TRUE)'. Could still be improved for multi-column case.
* modification to 'smoothCon' to allow resetting of smooth supplied
constraints - enables fix of bug in bam handling of 't2' terms, where
parameterization of penalty and model matrix did not previously match
properly.
* clarification of `exclude' argument to predict.gam in help files.
* modification to 'plot.gam' etc, so that 'ylim' is no longer shifted by
'shift'.
* ylim and ... handling improved for 'fs' plot method (thanks Dave Miller)
* gam.check now recognises RStudio and plots appropriately.
* bam(...,sparse=TRUE) removed - not efficient, because of unavoidability
of dense off diagonal terms in X'X or equivalent. Deprecated since 1.8-5.
* tweak to initial.sp/g to avoid infinite loop in s.p. initialization, in
rather unusual circumstances. Thanks to Mark Bravington.
* bam and gam have `drop.intercept' argument to force the parametric terms not
to include a constant in their span, even when there are factor variables.
* Fix in Vb.corr (2nd order edf correction) for fixed smoothing parameter case.
* added 'all.vars1' to enable terms like s(x$y) in model formulae.
* modification to gam.fit4 to ignore 'start' if it is immediately worse than
'null.coef'.
* cSplineDes can now accept a 'derivs' argument.
* added drop.intercept handling for multiple formulae (mod by Matteo Fasiolo).
* 'gam.side' fix to avoid duplication of smooth model matrices to be tested
against, when checking for numerical degeneracy. Problem could occasionally
cause a failure (especially with bam), when the total matrix to be tested
against ended upo with more columns than rows.
* 4 occurances of as.name("model.frame") changed to quote(stats::model.frame)
* fix in predict.bamd discrete prediction code to be a bit more relaxed about
use of as.factor, etc in formulae.
* fix in predict.gam handling of 'na.action' to avoid trying to get type of
na.action from name of na.action function - this was fragile to nesting
and could cause predict.bam to fail in the presence of NA's.
* fix of gam.outer so that general families (e.g. cox.ph) can have all
their smoothing parameters supplied without then ignoring the penalties!
* fix in multiple formulae handling of fixed smoothing parameters.
* Fix of bug in zlim handling in vis.gam perspective plot with standard
errors. Thanks Petra Kuhnert.
* probit link added to 'jagam' (suggested by Kenneth Knoblauch).
* 'Sl.' routines revised to allow operation with non-linearly parameterized
smoothers.
* bug fix in Hessian computation in gam.fit5 - leading diagonal of Hessian of
log|Hp| could be wrong where Hp is penalized Hessian.
* better use of crossprod in gamlss.gH
1.8-12
** "bs" B-spline smoothing basis. B-splines with derivative based penalties
of various orders.
* 'gamm' now uses a fixed scale parameter in PQL estimation for Poisson and
binomial data via the `sigma' option in lmeControl.
* bam null deviance computation was wrong with prior weights (including
binomial other than binary case), and returned deviance was wrong for
non-binary binomial. Fixed (did not affect estimation).
* improvements to "bfgs" optimizer to better deal with `infinite' smoothing
parameters.
* changed scheme=3 in default 2-D plotting to grey scale version of
scheme=2.
* 'trichol' and 'bandchol' added for banded Cholesky decompositions, plus
'sdiag' functions added for extracting and setting matrix sub- and
super- diagonals.
* p-spline constructor and Predict.matrix.pspline.smooth now allow set
up of SCOP-spline monotonic smoothers, and derivatives of smooths. Not
used in modelling functions yet.
* s(...,bs="re") now allows known precision matrix structures to be defined
using the `xt' argument of 's' see ?smooth.construct.re.smooth.spec for
details and example.
* negbin() with a grid search for `theta' is no longer supported - use
'nb' instead.
* bug fix to bam aic computation with AR rho correction.
1.8-11
* bam(...,discrete=TRUE) can now handle matrix arguments to smooths (and hence
linear functional terms).
* bam(...,discrete=TRUE) bug fix in fixed sp handling.
* bam(...,discrete = TRUE) db.drho reparameterization fix, fixing nonsensical
edf2. Also bam edf2 limited to maximum of edf1.
* smoothCon rescaling of S changed to use efficient matrix norm in place of
relatively slow computation involving model matrix crossproduct.
* bam aic corrected for AR model if present.
* Added select=TRUE argument to 'bam'.
* Several discrete prediction fixes including improved thread safety.
* bam/gam name gcv.ubre field by "method".
* gam.side modified so that if a smooth has 'side.constrain==FALSE' it is
neither constrained, nor used in the computation of constraints for other
terms (the latter part being new). Very limited impact!
* No longer checks if SUPPORT_OPENMP defined in Rconfig.h, but only if _OPENMP
defined. No change in actual behaviour.
1.8-10
** 'multinom' family implemented for multinomial logistic regression.
* predict.bam now defaults to using efficient discrete prediction methods
for models fit using discrete covariate methods (bam(...,discrete=TRUE)).
* with bam(...,discrete=TRUE) terms like s(a,b,bs="re") had wrong p-value
computation applied, as a result of being treated as tensor product terms.
Fixed.
* minor tweak to soap basis setup to avoid rounding error leading to 'approx'
occasionally producing NA's with fixed boundaries.
* misc.c:rwMatrix made thread safe (had been using R_chk_calloc, which isn't).
* some upgrading for 64bit addressing.
* uniquecombs now preserves contrasts on factors.
* variable summary tweak so that 1 column matrices in parametric model are
treated as regular numeric variables.
1.8-9
* C level fix in bam(...,discrete=TRUE) code. Some memory was mistakenly
allocated via 'calloc' rather than 'R_chk_calloc', but was then freed
via 'R_chk_free'. This could cause R to halt on some platforms.
1.8-8
** New "gp" smooth class (see ?gp.smooth) implemeting the Matern
covariance based Gaussian process model of Kamman and Wand (2003), and a
variety of other simple GP smoothers.
* some smooth plot methods now accept 'colors' and 'contour.col' argument to
set color palette in image plots and contour line colors.
* predict.gam and predict.bam now accept an 'exclude' argument allowing
terms (e.g. random effects) to be zeroed for prediction. For efficiency,
smooth terms not in 'terms' or in 'exclude' are no longer evaluated, and
are instead set to zero or not returned. See ?predict.gam.
* ocat saturated likelihood definition changed to zero, leading to better
comprability of deviance between model fits (thanks to Herwig Friedl).
* null.deviance calculation for extended families modified to make more sense
when `mu' is the mean of a latent variable, rather than the response itself.
* bam now returns standarized residuals 'std.rsd' if `rho!=0'.
* bam(...,discrete=TRUE) can now handle 'fs' terms.
* bam(...,discrete=TRUE) now accepts 'by' variables. Thanks to Zheyaun Li
for debugging on this.
* bam now works with drop.unused.levels == TRUE when random effects should
have more levels than those that exist in data. (Thanks Alec Leh)
* bam chunk.size logic error fix - error could be triggered if chunk.size
reset automaticlly to be larger than data size.
* uniqucombs can now accept a data frame with some or all factor columns,
as well as purely numeric marices.
* discrete.mf modified to avoid discretizing a covariate more than once,
and to halt if a model requires the same covariate to be discretized
two different ways (e.g. s(x) + s(x,z)). This affects only
bam(...,discrete=TRUE).
* Some changes to ziP and ziplss families to improve numerical robustness,
and to ziP help file to suggest appropriate checking. Thanks to Keren
Raiter, for reporting problems.
* numerical robustness of extended gam methods (gam.fit4) improved for cases
with many zero or near zero iterative weights. Handling of zero weights
modified to avoid divide by (near) zero problems. Also tests for poor
scaling of sqrt(abs(w))*z and substitutes computations based on w*z if
detected. Also 'newton' routine now step halves if REML score not finite!
* Sl.setup (used by bam) modification to allow more efficient handling of terms
with multiple diagonal penalties with no non-zero elements in common, but
possibly with non zero elements `interleaved' between penalties.
1.8-7
** 'gam' default scale parameter changed to modified Pearson estimator
developed by Fletcher 2012 Biometrika 99(1), 230-237. See ?gam.scale.
** 'bam' now has a 'discrete' argument to allow discretization of covariates
for more efficient computation, with substantially more parallelization
(via 'nthreads'). Still somewhat experimental.
* Slightly more accurate smoothing parameter uncertainty correction. Changes
edf2 used for AIC (under RE/ML), and hence may change AIC values.
* jagam prior variance on fixed effects is now set with reference to data and
model during initialization step.
* bam could lose offset for small datasets in gaussian additive case. fixed.
* gam.side now setup to include penalties in computations if fewer data than
coefs (an exceedingly specialist topic).
* p-value computation for smooth terms modified to avoid an ambiguity in the
choice of test statistic that could lead to p-value changing somewhat between
platforms.
* gamm now warns if attempt is made to use extended family.
* step fail logic improved for "fREML" optimization in 'bam'.
* fix of openMP error in mgcv_pbsi, which could cause a problem in
multi-threaded bam computation (failure to declare a variable as private).
* Smoothing parameter uncertainty corrected AIC calculations had an
indexing problem in Sl.postproc, which could result in failure of bam with
linked smooths.
* mroot patched for fact that chol(...,pivot=TRUE) does not operate as
documented on rank deficient matrices: trailing block of triangular factor
has to be zeroed for pivoted crossprod of factor to equal original matrix.
* bam(...,sparse=TRUE) deprecated as no examples found where it is really
worthwhile (but let me know if this is a problem).
* marginal model matrices in tensor product smooths now stored in
re-parameterized form, if re-parameterization happened (shouldn't change
anything!).
* initial.spg could fail if response vector had dim attributes and extended
family used. fixed.
1.8-6
* Generalization of list formula handling to allow linear predictors to
share terms. e.g. gam(list(y1~s(x),y2~s(z),1+2~s(v)+w-1),family=mvn(d=2))
* New German translation thanks to Detlef Steuer.
* plot.gam now silently returns a list of plotting data, to help advanced
users (Fabian Scheipl) to produce customized plot.
* bam can now set up an object suitable for fitting, but not actually do
the fit, following a suggestion by Fabian Scheipl. See arguments 'fit'
and 'G'.
1.8-5
* Korean translation added thanks to Chel Hee Lee.
* scale parameter handling in edf in logLik.gam made consistent with glm
(affects AIC).
* 'bam', 'gam' and 'gamm' modified to often produce smaller files when models
saved (and never to produce absurdly large files). Achieved by setting
environment of formula, terms etc to .GlobalEnv. Previously 'save' could
save entire contents of environment of formula/terms with fitted model
object. Note that change can cause failure in user written functions calling
gam/bam and then 'predict' without supplying all prediction variables
(fix obvious).
* A help file 'single.index' supplied illustrating how single index models
can be estimated in mgcv.
* predict.gam now only creates a "constant" attribute if the model has one.
* gam.fit4 convergence testing of coefs modified to more robust test of
gradients of penalized dev w.r.t. params, rather than change in params,
which can fail under rank deficiency.
* mgcv_qrqy was not thread safe. Not noticeable on many platforms as all
threads did exactly the same thing to the same matrix, but very noticeable
on Windows. Thread safe mgcv_qrqy0 added and used in any parallel sections.
* Allow openMP support if compiler supports it and provides pre-defined macro
_OPENMP, even if SUPPORT_OPENMP undefined. (Allows multi-threading on
Windows, for example.)
* 'eps' is now an argument to 'betar' allowing some control on how to
handle response values too close to 0 or 1. Help file expanded to
emphasise the problems with using beta regression with 0s and 1s in
the data.
* fix of bug in multi-formula contrast handling, causing failure of prediction
in some cases.
* ziP and ziplss now check for non-integer (or binary) responses and produce
an error message if these are found. Previously this was not trapped and
could lead to a segfault.
1.8-4
** JAGS/BUGS support added, enabling auto-generation of code and data
required to used mgcv type GAMs with JAGS. Useful for complex random
effects structures, for example.
* smoothCon failed if selection penalties requested, but term was unpenalized.
Now fixed (no selection penalties on unpenalized terms.)
* gam.check would fail for tensor product smooths with by variables - fixed.
* predict.gam would fail when predicting for more data than the blocksize
but selecting only some terms. Fixed thanks to Scott Kostyshak.
* smoothCon now has an argument `diagonal.penalty' allowing single penalty
smooths to be re-parameterized in order to diagonalize the penalty matrix.
PredictMat is modified to apply the same reparameterization, making it
user transparent. Facilitates the setup of smooths for export to other
packages.
* predict.bam now exported in response to a request from another
package maintainer.
* 1.8 allows some prediction tasks for some families (e.g. cox.ph) to
require response variables to be supplied. NAs in these then messed up
prediction when they were not needed (e.g. if response variables with
NAs were provided to predict.gam for a simple exponential family GAM).
Response NAs now passed to the family specific prediction code, restoring
the previous behaviour for most models. Thanks Casper Wilestofte Berg.
* backend parallel QR code used by gam modified to use a pivoted block
algorithm.
* nthreads argument added to 'bam' to allow for parallel computation
for computations in the main process (serial on any cluster nodes).
e.g. QR based combination of results from cluster nodes is now
parallel.
* fREML computation now partly in parallel (controlled by 'nthreads'
argument to 'bam')
* slanczos now accepts an nt argument allowing parallel computation of
main O(n^2) step.
* fix to newton logic problem, which could cause an attempt to use 'score2'
before definition.
* fix to fREML code which could cause matrix square root to lose dimensions
and cause an error.
* initial.sp could perform very poorly for very low basis dimensions - could
set initial sp to effective infinity.
1.8-3
* Fix of two illegal read/write bugs with extended family models with no
smooths. (Thanks to Julian Faraway for reporting beta regr problem).
* bam now checks that chunk.size > number of parameters and resets the
chunk.size if not.
* Examples of use of smoothCon and PredictMat for setting up bases
for use outside mgcv (and then predicting) added to ?smoothCon.
1.8-2
* For exponential family gams, fitted by outer iteration, a warning is now
generated if the Pearson scale parameter estimate is more than 4 times
a robust estimate. This may indicate an unstable Pearson estimate.
* 'gam.control' now has an option 'scale.est' to allow selection of the
estimator to use for the scale parameter in exponential family GAMs.
See ?gam.scale. Thanks to Trevor Davies for providing a clear unstable
Pearson estimate example.
* drop.unused.levels argument added to gam, bam and gamm to allow
"mrf" (and "re") terms to have unobserved factor levels.
* "mrf" constructor modified to deal properly with regions that contain no
observations.
* "fs" smooths are no longer eligible to have side conditions set, since
they are fully penalized terms and hence always identifiable (in theory).
* predict.bam was not declared as a method in NAMESPACE - fixed
* predict.bam modified to strip down object to save memory (especially in
parallel).
* predict.gam now has block.size=NULL as default. This implies a block
size of 1000 when newdata supplied, and use of a single block if no
new data was supplied.
* some messages were not printing correctly after a change in
message handling to facilitate easier translation. Now fixed.
1.8-1
* bam modified so that choleski based fitting works properly with rank
deficient model matrix (without regularization).
* fix of 1.8-0 bug - gam prior weights mishandled in computation of cov matrix,
resulting in incorrect variance estimates (even without prior weights
specified). Thanks Fabian Scheipl.
1.8-0
*** Cox Proportional Hazard family 'cox.ph' added as example of general
penalized likelihood families now useable with 'gam'.
*** 'ocat', 'tw', 'nb', 'betar', 'ziP' and 'scat' families added for
ordered categorical data, Tweedie with estimation of 'p', negative binomial
with (fast) estimation of 'theta', beta regression for proportions, simple
zero inflated Poisson regression and heavy tailed regression with scaled t
distribution. These are all examples of 'extended families' now useable
with 'gam'.
*** 'gaulss' and 'ziplss' families, implementing models with multiple linear
predictors. For gaulss there is a linear predictor for the Gaussian mean
and another for the standard deviation. For ziplss there is a linear
predictor controlling `presence' and another controlling
the Poisson parameter, given presence.
*** 'mvn' family for multivariate normal additive models.
** AIC computation changed for bam and gam models estimated by REML/ML
to account for smoothing parameter uncertainty in degrees of freedom
term.
* With REML/ML smoothness selection in gam/bam an extra covariance matrix 'Vc'
is now computed which allows for smoothing parameter uncertainty. See
the 'unconditional' arguments to 'predict.gam' and 'plot.gam' to use this.
* 'gam.vcomp' bug fix. Computed intervals for families with fixed scale
parameter were too wide.
* gam now defaults to the Pearson estimator of the scale parameter to avoid
poor scale estimates in the quasipoisson case with low counts (and possibly
elsewhere). Gaussian, Poisson and binomial inference invariant to change.
Thanks to Greg Dropkin, for reporting the issue.
* Polish translation added thanks to Lukasz Daniel.
* gam.fit3 now forces eta and mu to be consistent with coef and valid on
return (previously could happen that if step halving was used in final
iteration then eta or mu could be invalid, e.g. when using identity link
with non-negative data)
* gam.fit3 now bases its convergence criteria on grad deviance w.r.t. model
coefs, rather than changes in model coefs. This prevents problems when
there is rank deficiency but different coefs get dropped at different
iterations. Thanks to Kristynn Sullivan.
* If mgcv is not on the search path then interpret.gam now tries to
evaluate in namespace of mgcv with environment of formula as enclosing
environment, if evaluation in the environment of the formula fails.
* bug fix to sos plotting method so that it now works with 'by' variables.
* 'plot.gam' now weights partial residuals by *normalized* square root
iterative weights so that the average weight is 1 and the residuals
should have constant variance if all is ok.
* 'pcls' now reports if the initial point is not feasible.
* 'print.gam' and 'summary.gam' now report the rank of the model if it is
rank deficient. 'gam.check' reports the model rank whenever it is
available.
* fix of bug in 'k.check' called by 'gam.check' that gave an error for
smooths with by variables.
* predict.gam now checks that factors in newdata do not contain more
levels than those used in fitting.
* predict.gam could fail for type "terms" with no intercept - fixed.
* 'bfgs' now uses a finite difference approximation for the initial inverse
Hessian.
1.7-29
* Single character change to Makevars file so that openMP multi-threading
actually works.
1.7-28
* exclude.too.far updated to use kd-tree instead of inefficient search for
neighbours. This can make plot.gam *much* faster for large datasets.
* Change in smoothCon, so that sweep and drop constraints (default for bam
for efficiency reasons) are no longer allowed with by variables and matrix
arguments (could lead to confusing results with factor by variables in bam).
* 'ti' terms now allow control of which marginals to constrain, via 'mc'.
Allows e.g. y ~ ti(x) + ti(x,z,mc=c(0,1)) - for experts only!
* tensor.prod.model.matrix re-written to call C code. Around 5-10 times
faster than old version for large data sets.
* re-write of mini.mf function used by bam to generate a reduced size
model frame for model setup. New version ensures that all factor levels
are present in reduced frame, and avoids production of unrealistic
combinations of variables in multi-dimensional smooths which could occur
with old version.
* bam models could fail if a penalty matrix was 1 by 1, or if multiple
penalties on a smooth were in fact seperable into single penalties.
Fixed. Thanks to Martijn weiling for reporting.
* Constant in tps basis computation was different to published version
for odd dimensions - makes no difference to fit, but annoying if you
are trying to test a re-implementation. Thanks to Weijie Cai at SAS.
* prediction for "cc" and "cp" classes is now cyclic - values outside the
range of knots are wrapped back into the interval.
* ldTweedie now returns derivatives w.r.t. a transform of p as well as
w.r.t log of scale parameter phi.
* gamm can now handle 'varComb' variance functions (thanks Sven Neulinger
for reporting that it didn't).
* fix of a bug which could cause bam to seg fault for a model with no smooths
(insufficient storage allocated in C in this case). Thanks Martijn Weiling.
1.7-27
* Further multi-threading in gam fits - final two leading order matrix
operations parallelized using openMP.
* Export of smooth.construct.t2.smooth.spec and Predict.matrix.t2.smooth,
and Rrank.
* Fix of of missing [,,drop=FALSE] in predict.gam that could cause problems
with single row prediction when 'terms' supplied (thanks Yang Yang).
1.7-26
* Namespace fixes.
1.7-25
* code added to allow openMP based multi-threading in gam fits (see
?gam.control and ?"mgcv-parallel").
* bam now allows AR1 error model to be split blockwise. See argument
'AR.start'.
* magic.post.proc made more efficient (one of two O(np^2) steps removed).
* var.summary now coerces character to factor.
* bugs fixed whereby etastart etc were not passed to initial.spg and
get.null.coefs. Thanks to Gavin Simpson.
* reformulate removed from predict.gam to avoid (slow) repeated parser
calls.
* gaussian(link="log") initialization fixed so that negative data
does not make it fail, via fix.family patching function.
* bug fix in plot method for "fs" basis - ignored any side conditions.
Thanks to Martijn Weiling and Jacolien van Rij.
* gamm now checks whether smooths nested in factors have illegal side
conditions, and halts if so (re-ordering formula can help).
* anova.glmlist no longer called.
* Compiled code now uses R_chck_calloc and R_chk_free for memory management
to avoid the possibility of unfriendly exit on running out of memory.
* fix in gam.side which would fail with unpenalized interactions in the
presence of main effects.
1.7-24
* Examples pruned in negbin, smooth.construct.ad.smooth.spec and bam help
files to reduce CRAN checking load.
* gam.side now warns if only repeated 1-D smooths of the same variable are
encountered, but does not halt.
* Bug fix in C code for "cr" basis, that could cause a memory violation during
prediction, when an extrapolation was immediately followed by a prediction
that lay exactly on the upper boundary knot. Thanks to Keith Woolner for
reporting this.
* Fix for bug in fast REML code that could cause bam to fail with ti/te only
models. Thanks to Martijn Wieling.
* Fix of bug in extract.lme.cov2, which could cause gamm to fail when
a correlation structure was nested inside a grouping factor finer than
the finest random effect grouping factor.
* Fix for an interesting feature of lme that getGroups applied to the
corStruct that is part of the fitted lme object returns groups in
sorted order, not data frame order, and without an index from one order
to the other. (Oddly, the same corStruct Initialized outside lme has its
groups in data frame order.) This feature could cause gamm to fail,
complaining that the grouping factors for the correlation did not appear
to be nested inside the grouping structure of the random effects. A
bunch of ordering sensitivity tests have been added to the mgcv test suite.
Thanks to Dave Miller for reporting the bug.
1.7-23
*** Fix of severe bug introduced with R 2.15.2 LAPACK change. The shipped
version of dsyevr can fail to produce orthogonal eigenvectors when
uplo='U' (upper triangle of symmetric matrix used), as opposed to 'L'.
This led to a substantial number of gam smoothing parameter estimation
convergence failures, as the key stabilizing re-parameterization was
substantially degraded. The issue did not affect gaussian additive models
with GCV model selection. Other models could fail to converge any further
as soon as any smoothing parameter became `large', as happens when a
smooth is estimated as a straight line. check.gam reported the lack of full
convergence, but the issue could also generate complete fit failures.
Picked up late as full test suite had only been run on R > 2.15.1 with an
external LAPACK.
** 'ti' smooth specification introduced, which provides a much better (and
very simple) way of allowing nested models based on 'te' type tensor
product smooths. 'ti' terms are used to set up smooth interactions
excluding main effects (so ti(x,z) is like x:z while te(x,z) is more
like x*z, although the analogy is not exact).
* summary.gam now uses a more efficient approach to p-value computation
for smooths, using the factor R from the QR factorization of the weighted
model matrix produced during fitting. This is a weighted version of the
Wood (2013) statistic used previously - simulations in that paper
essentially unchanged by the change.
* summary.gam now deals gracefully with terms such as "fs" smooths
estimated using gamm, for which p-values can not be computed. (thanks to
Gavin Simpson).
* gam.check/qq.gam now uses a normal QQ-plot when the model has been fitted
using gamm or gamm4, since qq.gam cannot compute corrext quantiles in
the presence of random effects in these cases.
* gamm could fail with fixed smooths while assembling total
penalty matrix, by attempting to access non-existent penalty
matrix. (Thanks Ainars Aunins for reporting this.)
* stripped rownames from model matrix, eta, linear predictor etc. Saves
memory and time.
* plot.soap.film could switch axis ranges. Fixed.
* plot.mgcv.smooth now sets smooth plot range on basis of xlim and
ylim if present.
* formXtViX documentation fixed + return matrix labels.
* fixDependence related negative index failures for completely confounded
terms - now fixed.
* sos smooth model matrix re-scaled for better conditioning.
* sos plot method could produce NaNs by a rounding error in argument to
acos - fixed.
1.7-22
* Predict.matrix.pspline.smooth now allows prediction outside range of knots,
and uses linear extrapolation in this case.
* missing drop=FALSE in reTest called by summary.gam caused 1-D random effect
p-value computation to fail. Fixed (thanks Silje Skår).
1.7-21
** soap film smoother class added. See ?soap
* Polish translation added thanks to Lukasz Daniel.
* mgcv/po/R-mgcv.pot up-dated.
* plot methods for smooths modified slightly to allow methods to return
plot data directly, without a prediction matrix.
1.7-20
* '...' now passed to termplot by plot.gam (thanks Andreas Eckner).
* fix to null deviance computation for binomial when n>1, matrix response
used and an offset is present. (Thanks to Tim Miller)
* Some pruning of unused code from recov and reTest.
* recov modified to stop it returning a numerically non-symmetric Ve, and
causing occasional failures of summary.gam with "re" terms.
* MRF smooth bug. Region ordering could become confused under some
circumstances due to incorrect setting of factor levels. Corrected
thanks to detailed bug report from Andreas Bender.
* polys.plot colour/grey scale bug. Could ask for colour 0 from colour
scheme, and therefore fail. Fixed.
1.7-19
** summary.gam and anova.gam now use an improved p-value computation for
smooth terms with a zero dimensional penalty null space (including
random effects). The new scheme has been tested by full replication
of the simulation study in Scheipl (2008,CSDA) to compare it to the best
method therein. In these tests it is at least as powerful as the best
method given there, and usually indistinguishable, but it gives slightly
too low null p-values when smoothing parameters are very poorly identified.
Note that the new p-values can not be computed from old fitted gam objects.
Thanks to Martijn Wieling for pointing out how bad the p-values for regular
smooths could be with random effects.
* t2 terms now take an argument `ord' that allows orders of interaction to
be selected.
* "tp" smooths can now drop the null space from their construction via
a vector m argument, to allow testing against polynomials in the null space.
* Fix of vicious little bug in gamm tensor product handling that could have
a te term pick up the wrong model matrix and fail.
* bam now resets method="fREML" to "REML" if there are no free smoothing
parameters, since there is no advantage to the "fREML" optimizer in this
case, and it assumes there is at least one free smoothing parameter.
* print.gam modified to print effective degrees of freedom more prettily,
* testStat bug fix. qr was called with default arguments, which includes
tol=1e-7...
* bam now correctly returns fitting weights (rather than prior) in weights
field.
1.7-18
* Embarrassingly, the adjusted r^2 computation in summary.gam was wrong
for models with prior weights. Now fixed, thanks to Antony Unwin.
* bam(...,method="fREML") could give incorrect edfs for "re" terms as a
result of a matrix indexing error in Sl.initial.repara. Now fixed.
Thanks to Martijn Wieling for reporting this.
* summary.gam had freq=TRUE set as default in 1.7-17. This gave better
p-values for paraPen terms, but spoiled p-values for fixed effects in the
presence of "re" terms (a rather more common setup). Default now reset to
freq=FALSE.
* bam(...,method="fREML") made fully compatible with gam.vcomp.
* bam and negbin examples speeded up
* predict.gam could fail for models of the form y~1 when newdata are supplied.
(Could make some model averaging methods fail). Fixed.
* plot.gam had an overzealous check for availibility of variance estimates,
which could make rank deficient models fail to plot CIs. fixed.
1.7-17
** p-values for terms with no un-penalized components were poor. The theory on
which the p-value computation for other terms is based shows why this is,
and allows fixes to be made. These are now implemented.
* summary p value bug fix --- smooths with no null space had a bug in
lower tail of p-value computation, yielding far too low values. Fixed.
* bam now outputs frequentist cov matrix Ve and alternative effective degrees
of freedom edf1, in all cases.
* smoothCon now adjusts null.space.dim on constraint absorption.
* Prediction with matrix arguments (i.e. for models using summation
convention) could be very memory hungry. This in turn meant that
bam could run out of memory when fitting models with such terms.
The problem was memory inefficient handling of duplicate evaluations.
Now fixed by modification of PredictMat
* bam could fail if the response vector was of class matrix. fixed.
* reduced rank mrf smooths with supplied penalty could use the incorrect
penalty rank when computing the reduced rank basis and fail. fixed
thanks to Fabian Scheipl.
* a cr basis efficiency change could lead to old fitted model objects causing
segfaults when used with current mgcv version. This is now caught.
1.7-16
* There was an unitialized variable bug in the 1.7-14 re-written "cr" basis
code for the case k=3. Fixed.
* gam.check modified slightly so that k test only applied to smooths of
numeric variables, not factors.
1.7-15
* Several packages had documentation linking to the 'mgcv' function
help page (now removed), when a link to the package was meant. An alias
has been added to mgcv-package.Rd to fix/correct these links.
1.7-14
** predict.bam now added as a wrapper for predict.gam, allowing parallel
computation
** bam now has method="fREML" option which uses faster REML optimizer:
can make a big difference on parameter rich models.
* bam can now use a cross product and Choleski based method to accumulate
the required model matrix factorization. Faster, but less stable than
the QR based default.
* bam can now obtain starting values using a random sub sample of the data.
Useful for seriously large datasets.
* check of adequacy of basis dimensions added to gam.check
* magic can now deal with model matrices with more columns than rows.
* p-value reference distribution approximations improved.
* bam returns objects of class "bam" inheriting from "gam"
* bam now uses newdata.guaranteed=TRUE option when predicting as part
of model matrix decomposition accumulation. Speeds things up.
* More efficient `sweep and drop' centering constraints added as default for
bam. Constaint null space unchanged, but computation is faster.
* Underlying "cr" basis code re-written for greater efficiency.
* routine mgcv removed, it now being many years since there has been any
reason to use it. C source code heavily pruned as a result.
* coefficient name generation moved from estimate.gam to gam.setup.
* smooth2random.tensor.smooth had a bug that could produce a nonsensical
penalty null space rank and an error, in some cases (e.g. "cc" basis)
causing te terms to fail in gamm. Fixed.
* minor change to te constructor. Any unpenalized margin now has
corresponding penalty rank dropped along with penalty.
* Code for handling sp's fixed at exactly zero was badly thought out, and
could easily fail. fixed.
* TPRS prediction code made more efficient, partly by use of BLAS. Large
dataset setup also made more efficient using BLAS.
* smooth.construct.tensor.smooth.spec now handles marginals with factor
arguments properly (there was a knot generation bug in this case)
* bam now uses LAPACK version of qr, for model matrix QR, since it's
faster and uses BLAS.
1.7-13
** The Lanczos routine in mat.c was using a stupidly inefficient check for
convergence of the largest magnitude eigenvectors. This resulted in
far too many Lanczos steps being used in setting up thin plate regression
splines, and a noticeable speed penalty. This is now fixed, with many thanks
David Shavlik for reporting the slow down.
* Namespace modified to import from methods. Dependency on stats and graphics
made explicit.
* "re" smooths are no longer subject to side constraint under nesting (since
this is almost always un-necessary and undesirable, and often unexpected).
* side.con modified to allow smooths to be excluded and to allow side
constraint computation to take account of penalties (unused at present).
1.7-12
* bam can now compute the leading order QR decomposition on a cluster
set up using the parallel package.
* Default k for "tp" and "ds" modified so that it doesn't exceed 100 +
the null space dimension (to avoid complaints from users smoothing in
quite alot of dimensions). Also default sub-sample size reduced to 2000.
* Greater use of BLAS routines in the underlying method code. In particular
all leading order operations count steps for gam fitting now use BLAS.
You'll need R to be using a rather fancy BLAS to see much difference,
however.
* Amusingly, some highly tuned blas libraries can result in lapack not always
giving identical eigenvalues when called twice with the same matrix. The
`newton' optimizer had assumed this wouldn't happen: not any more.
* Now byte compiled by default. Turn this off in DESCRIPTION if it interferes
with debugging.
* summary.gam p-value computation options modified (default remains the
same).
* summary.gam default p-value computation made more computationally
efficient.
* gamm and bam could fail under some options for specifying binomial models.
Now fixed.
1.7-11
* smoothCon bug fix to avoid NA labels for matrix arguments when
no by variable provided.
* modification to p-value computation in summary.gam: `alpha' argument
removed (was set to zero anyway); computation now deals with possibility
of rank deficiency computing psuedo-inverse of cov matrix for statistic.
Previously p-value computation could fail for random effect smooths with
large datasets, when a random effect has many levels. Also for large data
sets test statistic is now based on randomly sampling max(1000,np*2) model
matrix rows, where np is number of model coefficients (random number
generator state unchanged by this), previous sample size was 3000.
* plot.mrf.smooth modified to allow passing '...' argument.
* 'negbin' modified to avoid spurious warnings on initialization call.
1.7-10
* fix stupid bug in 1.7-9 that lost term labels in plot.gam.
1.7-9
* rather lovely plot method added for splines on the sphere.
* plot.gam modified to allow 'scheme' to be specified for plots, to easily
select different plot looks.
* schemes added for default smooth plotting method, modified for mrfs and
factor-smooth interactions.
* mgcv function deprected, since magic and gam are much better (let me know
if this is really a problem).
1.7-8
* gamm.setup fix. Bug introduced in 1.7-7 whereby gamm with no smooths would
fail.
* gamm gives returned object a class "gamm"
1.7-7
* "fs" smooth factor interaction class introduced, for smooth factor
interactions where smoothing parameters are same at each factor level.
Very efficient with gamm, so good for e.g. individual subject smooths.
* qq.gam default method modified for increased power.
* "re" terms now allowed as tensor product marginals.
* log saturated likelihoods modified w.r.t. weight handling, so that weights
are treated as modifying the scale parameter, when scale parameter is free.
i.e. obs specific scale parameter is overall scale parameter divided by
obs weight. This ensures that when the scale parameter is free, RE/ML based
inference is invariant to multiplicative rescaling of weights.
* te and t2 now accept lists for 'm'. This allows more flexibility with
marginals that can have vector 'm' arguments (Duchon splines, P splines).
* minor mroot fix/gam.reparam fix. Could declare symmetric matrix
not symmetric and halt gam fit.
* argument sparse added to bam to allow exploitation of sparsity in fitting,
but results disappointing.
* "mrf" now evaluates rank of penalty null space numerically (previously
assumed it was always one, which it need not be with e.g. a supplied
penalty).
* gam.side now corrects the penalty rank in smooth objects that have
been constrained, to account for the constraint. Avoids some nested
model failures.
* gamm and gamm.setup code restructured to allow smooths nested in factors
and for cleaner object oriented converion of smooths to random effects.
* gam.fit3 bug. Could fail on immediate divergence as null.eta was matrix.
* slanczos bug fixes --- could segfault if k negative. Could also fail to
return correct values when k small and kl < 0 (due to a convergence
testing bug, now fixed)
* gamm bug --- could fail if only smooth was a fixed one, by looking for
non-existent sp vector. fixed.
* 'cc' Predict.matrix bug fix - prediction failed for single points.
* summary.gam failed for single coefficient random effects. fixed.
* gam returns rV, where t(rV)%*%rV*scale is Bayesian cov matrix.
1.7-6
** factor `by' variable handling extended: if a by variable is an
ordered factor then the first level is treated as a reference level
and smooths are only generated for the other levels. This is useful
for avoiding identifiability issues in complex models with factor by
variables.
* bam bug fix. aic was reported incorrectly (too low).
1.7-5
* gam.fit3 modified to converge more reliably with links that don't guarantee
feasible mu (e.g poisson(link="identity")). One vulnerability removed + a
new approach taken, which restarts the iteration from null model
coefficients if the original start values lead to an infinite deviance.
* Duchon spline bug fix (could fail to create model matrix if
number of data was one greater than number of unique data).
* fix so that 'main' is not ignored by plot.gam (got broken in 1.7-0
object orientation of smooth plotting)
* Duchon spline constructor now catches k > number of data errors.
* fix of a gamm bug whereby a model with no smooths would fail after
fitting because of a missing smoothing parameter vector.
* fix to bug introduced to gam/bam in 1.7-3, whereby '...' were passed to
gam.control, instead of passing on to fitting routines.
* fix of some compiler warnings in matrix.c
* fix to indexing bug in monotonic additive model example in ?pcls.
1.7-4
* Fix for single letter typo bug in C code called by slanczos, could
actually segfault on matrices of less than 10 by 10.
* matrix.c:Rlanczos memory error fix in convergence testing of -ve
eigenvalues.
* Catch for min.sp vector all zeroes, which could cause an ungraceful
failure.
1.7-3
** "ds" (Duchon splines) smooth class added. See ?Duchon.spline
** "sos" (spline on the sphere) smooth class added. See ?Spherical.Spline.
* Extended quasi-likelihood used with RE/ML smoothness selection and
quasi families.
* random subsampling code in bam, sos and tp smooths modified a little, so
that .Random.seed is set if it doesn't exist.
* `control' argument changed for gam/bam/gamm to a simple list, which is
then passed to gam.control (or lmeControl), to match `glm'.
* Efficiency of Lanczos iteration code improved, by restructuring, and
calling LAPACK for the eigen decompostion of the working tri-diagonal
matrix.
* Slight modification to `t2' marginal reparameterization, so that `main
effects' can be extracted more easily, if required.
1.7-2
* `polys.plot' now exported, to facilitate plotting of results for
models involving mrf terms.
* bug fix in plot.gam --- too.far had stopped working in 1.7-0.
1.7-1
* post fitting constraint modification would fail if model matrix was
rank deficient until penalized. This was an issue when mixing new t2
terms with "re" type random effects. Fixed.
* plot.mrf.smooth bug fix. There was an implicit assumption that the
`polys' list was ordered in the same way as the levels of the covariate
of the smooth. fixed.
* gam.side intercept detection could occasionally fail. Improved.
* concurvity would fail if model matrix contained NA's. Fixed.
1.7-0
** `t2' alternative tensor product smooths added. These can be used with
gamm4.
** "mrf" smooth class added (at the suggestion of Thomas Kneib).
Implements smoothing over discrete geographic districts using a
Markov random field penalty. See ?mrf
* qq.gam added to allow better checking of distribution of residuals.
* gam.check modified to use qq.gam for QQ plots of deviance residuals.
Also, it now works with gam(*, na.action = "na.replace") and NAs.
* `concurvity' function added to provide simple concurvity measures.
* plot.gam automatic layout modified to be a bit more sensible (i.e.
to recognise that most screens are landscape, and that usually
squarish plots are wanted).
* Plot method added for mrf smooths.
* in.out function added to test whether points are interior to
a region defined by a set of polygons. Useful when working with
MRFs.
* `plot.gam' restructured so that smooths are plotted by smooth specific
plot methods.
* Plot method added for "random.effect" smooth class.
* `pen.edf' function added to extract EDF associated with each penalty.
Useful with t2 smooths.
* Facilty provided to allow different identifiability constraints to be
used for fitting and prediction. This allows t2 smooths to be fitted
with a constraint that allows fitting by gamm4, but still perform
inference with the componentwise optimal sum to zero constraints.
* mgcv-FAQ.Rd added.
* paraPen works properly with `gam.vcomp' and full.sp names returned
correctly.
* bam (and bam.update) can now employ an AR1 error model in the
guassian-identity case.
* bam.update modified for faster updates (initial scale parameter
estimate now supplied in RE/ML case)
* Absorption of identifiability constraints modified to allow
constraints that only affect some parameters to leave rest of
parameters completely unchanged.
* rTweedie added for quick simulation of Tweedie random deviates
when 1<p<2.
* smooth.terms help file fixed so cyclic p spline identifies as "cp"
and not "cs"!
* bug fix in `gamm' so that binomial response can be provided as 2 column
matrix, in standard `glm' way.
1.6-2
** Random effect support for `gam' improved with the addition of
a random effect "smooth" class, and a function for extracting
variance components. See ?random.effects and links for details.
* smooths now contain extra elements: S.scale records the scale factor
used in any linear rescaling of a penalty matrix; plot.me indicates
whether `plot.gam' should attempt to plot the term; te.ok indicates
whether the smooth is a suitable marginal for a tensor product.
* Fix in `gamm.setup' -- models with no fixed effects (except smooths)
could fail to fit properly, because of an indexing error (caused odd
results with indicator by variables)
* help files have had various misuses of `itemize' fixed.
* initialization could fail if response was actually a 1D array. fixed.
* New function `bam.update' allows efficient updating of very large
strictly additive models fitted by `bam' when additional data become
available.
* gam now warns if RE/ML used with quasi families.
* gam.check now accepts graphics parameters.
* fixed problem in welcome message that messed up ESS.
1.6-1
* Bug in cSplineDes caused bases to be non-cyclic unless first
knot was at 0. This also affected the "cp" smoother class. Fixed.
* null.deviance calculation was wrong for case with offset and
weights. Fixed.
* Built in strictly 1D smoothers now give an informative error message if
an attempt is made to use them for multidimensional smoothing.
* gam.check generated a spurious error when applied to a model with no
estimated smoothing parameters. Fixed.
1.6-0
*** Routine `bam' added for fitting GAMs to very large datasets.
See ?bam
** p-value method tweaked again. Reference DoF for testing now
defaults to the alternative EDF estimate (based on 2F - FF where
F = (X'WX+S)^{-1}X'WX). `magic.post.proc' and `gam.fit3.post.proc'
changed to provide this. p-values still a bit too small, but only
slightly so, if `method="ML"' is the smoothness selector.
* bad bug in `get.null.coef' could cause fit failure as a result of
initial null coefs predicting infinite deviance.
* REML/ML convergence could be response scale sensitive, because of
innapropriate convergence testing scaling in newton and bfgs -
fixed.
* Slight fix to REML (not ML) score calculation in gam.fit3 -
Mp/2*log(2*pi*scale) was missing from REML score, where Mp is
total null space dimension for model.
* `summary.gam' bug fix: REML/ML models were always treated as if
scale parameter had been estimated. gamObject should now contain
`scale.estimated' indicating whether or not scale estimated
* some modifications to smoothCon and gam.setup to allow smooth
constructors to return Matrix style sparse model matrices and
penalty matrices.
* fixed misplaced bracket in print.mgcv.version, called on attachment.
* added utility function `ls.list' to give memory usage for elements
of a list.
* added function `rig' to generate inverse Gaussian random deviates.
1.5-6
* "ts" and "cs" modified so that zero eigen values of penalty
matrix are reset to 10% of smallest strictly positive eigen
value, rather than 1%. This seems to lead to more reliable
performance.
* `bfgs' simplified and improved so that it now checks the Wolfe
conditions are met at each step. No longer uses any Newton steps,
so if it's used with gam.control(outerPIsteps=0) then it's
first derivative only for smoothing parameter optimization.
* `outerPIsteps' now defaults to zero in `gam.control'.
* New routine `initial.spg' gets jth initial sp to equalize
Frobenious norm of S_j and cols of sqrt(W)X which it penalizes,
where W are initial fisher weights. This removes the need for a
performance iteration step to get starting values (so
outerPIsteps=0 in gam.control can now bypass PI completely).
* fscale set from get.null.coef (facilitates cleaner initialization).
* large data set rare event logistic regression example added to
?gam.
* For p-value calculation for smooths, summary.gam subsamples rows of
the model matrix if it has more than 3000 rows. This speeds things
up for large datasets.
* minor bug fix in `gamm' so that intercept gets correct name, if
it's the only non-smooth fixed effect.
* .pot files updated, German translation added, thanks to Detlef Steuer.
* `in.out' was not working from 1.5 --- fixed.
* loglik.gam now ups parameter count for Tweedie by one to account for
scale estimation.
* There was a bug in the calculation of the Bayesian cov matrix, when the
scale parameter was known: it was always using an estimated scale
parameter. Makes no statistically meaningful difference for a model
that fits properly, of course.
* Some junk removed from gam object.
* summary.gam pseudoinversion made slightly more efficient.
* adaptive smooth constructor is a bit more careful about the ranks
of the penalties.
* 2d adaptive smoother bug fix --- part of penalty was missing due
to complete line error.
* `smoothCon' and `PredictMat' modified so that sparse smooths can
optionally have sparse centering constraints applied.
* `gamm' fix: prediction and visualization from `x$gam' where x is a
fitted `gamm' object should not require the random effects to be
provided. Now it doesn't.
* minor bug fix: a model with no penalties except a fixed one would fail
with an index error.
* `te' terms are now only subject to centering constraints if all
marginals would usually have a centering constraint.
* `te' no longer resets multi-dimensional marginals to "tp", unless
they have been set to "cr", "cs", "ps" or "cp". This allows tensor
products with user supplied smooths.
* Example of obtaining derivatives of a smooth (with CIs) added to
`predict.gam' help file.
* `newdata.guaranteed' argument to predict.gam didn't work. fixed.
1.5-5
* `gamm.setup' made an assumption about basis dimensions which was not
true for tensor products involving the "cc" basis. This is now fixed.
1.5-4
* smooth.construct.tensor.smooth.spec modified, so that
re-parameterization in terms of function values is only if it's
stable, and by default the parameters are function values with
even spacing. Otherwise it was possible for tensor products of
p-splines to fail.
1.5-3
* `gam' now attempts to coerce `data' to a data frame, if it is not
already a list or a data frame, provided that it is already an object
that model.frame can deal with. This is to support an undocumented
feature of versions prior to 1.5-2 that `data' could actually be
something other than a list or data frame.
* An offset of type "array" could cause gam.fit3 to fail. fixed.
* `variable.summary' bug fixed, (it caused gam(y~1) to fail).
1.5-2
* Several exported functions had no usage entries in the help files.
Everything exported does now.
* `vis.gam' had a bunch of bugs (which could make it fail altogether)
as a result of trying to set default conditioning values from the gam
object model frame. `gam' and `gamm' now obtain summary statistics of
the predictor variables, stored in `var.summary' in the gam object,
which `vis.gam' now uses. As a result `vis.gam' `view' and `cond'
arguments should now contain original variable names, not model frame
term names.
* `data' argument of `gam' no longer stored in the `gam' object, by
default to save memory (can restore this --- see `gam.control').
* `summary.gam' failed under na.exclude. Fixed.
* `mroot' failed on 1 by 1 matrices, Fixed.
1.5-1
* The stability of the fitting methods had become substantially greater
than the stability of the edf calculations after fitting. So it was
possible to fit very poor models, and then have non-sensical effective
degrees of freedom reported. This has been fixed by using much more stable
expressions for edf calculation, when outer iteration smoothness
selection is employed. (Changes are to gam.fit3, gam.outer and a new
routine gam.fit3.post.proc).
* edfs in version 1.5-0 were calculated using newton, rather than fisher
weights, in the matrix F=(X'WX+S)^{-1}X'WX, the diagonal of which gives
the edf array. The problem with this is that it is possible for X'WX
not to be +ve definite, and then degrees of freedom can be non-sensical.
Fisher weights are always used now (although the original problem is
exceedingly hard to generate an example of).
* The summation convention code could be *very* memory intensive for cases
in which the matrix arguments of a smooth feature many repeated values.
Code now fixed to make much more efficient use of any repeated rows in
matrix arguments. This enables much larger signal regression problems to
be tackled.
* Some help file fixes.
1.5-0
*** Efficient and general REML/ML smoothing selection implemented.
Smoothness selection criterion and numerical optimizer are now
selected using arguments `method' and `optimizer' of `gam', and
`gam.method' has been removed.
*** To further enhance stability and efficiency, Fisher scoring is now
only used for canonical links, when it corresponds to full Newton.
With non-canonical links PIRLS is based on full Newton.
** Derivative iteration as in Wood (2008) has been replaced by a direct
implicit function method (which costs no more given Newton based
PIRLS).
** An option `select' has been added to `gam' to allow terms to be
completely removed from a model by smoothness selection.
** The shrinkage smoothers "cs" and "ts" have been modified
substantially. The Wood (2006, 4.1.6) proposal of adding a
small multiple of the identity matrix to the penalty matrix
is flawed in that it tends to corrupt small eigen values
of the penalty matrix for large (dimension) penalty matrices.
It is much better to set the zero eigenvalues of the penalty matrix
to a small proportion of the smallest +ve eigenvalue, and to
use the matrix with the resulting eigen-decomposition as the
penalty. This is now done. Thanks to Roman Torgovitsky for
reporting the original problem.
** Tweedie family added (including `ldTweedie' function to evaluate
log Tweedie densities for powers in (1,2]).
* "ps" "cp" and "cc" smooths can now be supplied with 2 knots to be
treated as `endpoints' of the smooths (full set of knots can still
be supplied as before).
* The `newton' optimizer was dropping terms when their gradient was
below the convergence threshold (and allowing re-entry). This
promotes zig-zagging unless the terms are independent. Now only
drops terms if gradient and second derivative are very small
(so obective is really flat).
* The adaptive smoothing "ad" class has been greatly simplified and 2D
penalty improved. Much faster as a result, and 2D adaptive actually
quite good.
* gam.fit3 now checks that the initial PIRLS step produces an
improvement in penalized deviance relative to a null model. If not
then step halving towards the null model parameter is employed.
The null model is as close to constant predicted values as the model
structure allows (it is estimated up front in estimate.gam, to save
computation).
* `gam.side' now takes account of whether the model has an intercept
(or the model model matrix column corresponding to an intercept is
in the column space of the model matrix of the parametric model
components).
* The smoothing parameter array returned by `gam' now includes names
for the smoothing parameters.
* s and te check that `id' is a single element.
* By default, partial residuals are no longer plotted for smooths with
`by' variables since they are usually meaningless here (they can
be re-instated by argument `by.resids').
* `min.sp' processing modified to work with `paraPen' argument to
`gam'.
* vcov.gam defaults to Bayesian covariance matrix.
* indexing error in `parametricPenalty' corrected.
* plot.gam modified so that page change behaviour is like plot.glm
* negbin family upgraded to work with (RE)ML.
* `sp.vcov' function added to extract covariance matrix of log
smoothing parameters from (RE)ML based fits.
* `power' links now handled by default fitting methods (i.e. gam.fit3)
* `magic.post.proc' now expects weights, not sqrt(weights) as the
`w' argument (unless `w' is a matrix).
* p-values tweaked again, for slightly better performance with smooths of
several variables. Still not quite right.
* record of intial sp's is now carried in `smooth' objects in field
`sp'.
* ?linear.functional.term error fix.
* memory leak in magic.c:magic fixed --- all fixed smoothing
parameters lead to 2 arrays being left unfreed.
* various .Rd file fixes.
1.4-2
* Some minor .Rd file fixes
1.4-1
* `Predict.matrix2' was not in NAMESPACE: fixed.
* term specific offsets handled properly w.r.t. `by' variables in
`smoothCon' (a rather specialized topic!)
* minor doc bug fix for `smooth.construct'.
1.4-0
*** Model terms can now include linear functionals of smooths, by
supplying matrix arguments and matrix `by' variables to model
smooth terms. This allows, for example, a model to depend on the
integral of a smooth function, or its derivative, or for models to
depend on functional predictors. See ?linear.functional.terms. Main
code changes are in `smoothCon' and `predMat'.
** Smooth terms can now be linked in order that they have the same
smoothing parameters (and, by default, bases). Linkage is specified
using the `id' argument to `s' or `te'. Terms with the same `id' value
will have the same smoothing parameter(s).
** `by' variables can now be factor variables. Also smooth terms with a
`by' variable are only subject to a sum-to-zero constraint if it
is needed for identifiability.
** Argument `paraPen' of `gam' allows (multiple) penalization of
parametric model terms. This allows `gam' to fit any model that
can be expressed as a penalized GLM.
** p-values returned by `summary.gam' now default to a Bayesian
approximation which gives (substantially) better frequentist behaviour
than the old method.
** The 2 standard error bands for smooths shown by `plot.gam' can now
include the uncertainty about the overall mean, by default. Such
intervals have better coverage probability (of their target of
inference) than intervals for centred smooths. Argument
`type="iterms"' to `predict.gam' will return such standard errors.
** An adaptive smoother class has been added, for smoothing with respect
to one or two variables: invoked with `s(...,bs="ad",...).
** `gamm' now supports nested smooth terms, and uses the same, constraint
absorbed, parameterization as `gam'.
** `s' and `te' terms accept an `sp' argument setting the term specific
smoothing parameters (and over-riding argument `sp' of `gam'). Ignored
by `gamm'.
** Negative binomial handling changed. `negbin' family added: adapted from
MASS to work with gam outer iteration fitting. `gam.negbin' fitting
routine added in order to enable use of `negbin' with outer iteration.
See ?negbin for details. MASS families no longer supported.
`nb.theta.mult' removed from `gam.control'.
** The Eilers and Marx style p-spline class is now one of the default
smoothing classes, rather than just being an example of how to set up
a class in the help file. cyclic versions are also available.
** `smoothCon' now handles `by' variables and centering constraints
automatically, removing the need for smooth constructors to do so.
`PredMat' handles `by' variables automatically. Users can over-ride
this behaviour when adding smooth classes, if needed - see
documentation.
** The interface for adding user defined smooths has been simplified,
but this may mean that some user defined classes which worked before
no longer work: see ?user.defined.smooths
* `smooth.construct' methods are now expected to set default values
for the penalty order `p.order' and the basis dimension `bs.dim'
if none are supplied. They should also sanity check supplied
values. Previously this was done by `s', but this put unhelpful
restrictions on new smooth classes.
* `smooth.construct' now expects to recieve `data' and `knots' arguments
with names corresponding exactly to `object$term'. In addition `data'
should contain only what is required by `object$term' + a final column
containing a `by' variable, if present. Predict.mat expects the same of
its `data' argument. wrapper functions smooth.construct2 and
Predict.mat2 will accept a data frame containing any number of variables
-- all that is required is that `object$terms' can be evaluated using
it. These functions handle repeat rows in matrix arguments efficiently.
* bug fix in `plot.gam' -- no longer requires to hit return if `select'
used (ever).
* bug fix in `fixDependence' --- a completely dependent `X2' would not
be detected, since the first element of R2 would be zero: used first
element of R1 to set scale instead.
* `gam.fit' passes corrected n to `magic' so that `n' used in gcv/ubre
does not include obs with zero prior weight. `gam.fit3' already doing
this...
* `magic' and `gam.fit3' now allow log smoothing parameters to be a
linear transformation of a smaller set of underlying smoothing parameters.
* `mgcv' based fitting has been removed as an option in `gam', as has
Pearson based GCV. In consequence `am' argument removed from
`gam.method' and `globit' removed from `gam.control'.
* `get.var' now coerces matrix values to numeric vectors, to facilitate
the handling of linear functionals of smooths.
* `gam.fit2' has been removed, since gam.fit3 is simply better.
* The default optimizer for the generalized case has been made slightly
more efficient (derivative free evaluation of GCV/AIC has been
improved). The upshot is that the default is now faster than performance
iteration in almost all cases (while still being more reliable).
* the `absorb.cons' option has been removed from `gam.control'.
* `fix.family.link' and `fix.family.var' bug fix --- only return
family unmodified if all required derivative functions are present.
* `smoothCon' now returns a list of smooth objects to facilitate factor
`by' variables.
* `smoothCon' makes smooth object labels more informative, if there are
`by' variables... this also makes default plots more informative.
* `plot.gam' indicates `by' variables in labels
* `gamm.setup' modified to call `gam.setup' for most of the setup, leaving
just the re-parameterization step to do.
* `gamm' modified to allow constraint absorption (same as `gam')
* `gamm' bug fixed whereby "cc" smooths would get the wrong null space
dimension (effect was small, but noticeable, in practice e.g. Cairo
temperature example from chapter 6 of Wood, 2006, book).
* print methods now return first argument invisibly as they should.
* code for (very) old style summary removed.
* `gam.fit3' now traps derivative iteration divergence, and suggests
tightening the convergence tolerance `epsilon' in `gam.control'.
Divergence can happen for ill-conditioned models if the PIRLS has
not converged sufficiently.
* gamm.Rd updated to reflect change to gammPQL in 1.3-28.
1.3-31
* There was a most annoying warning generated by R 2.7.0 every time `gam'
was used. Now there isn't.
1.3-30
* change to DESCRIPTION file.
1.3-29
* `magic' could segfault if supplied with many constraints and relatively
high rank penalties, so that after constriant the penalty matrix
square roots had more columns than rows (never happened in additive
model case, but can happen in more general settings). Fixed.
* `gamm' now silently drops grouping factors within the correlation
structure formulae that duplicate random effects grouping factors
(which automatically act as grouping factors on the correlation
structures anyway).
* Some replacement of dubious `as.matrix' calls with use of `,drop=FALSE]'
in gamm.r
1.3-28
** `gamm' modified to call a routine `gammPQL' in place of MASS::glmmPQL.
This avoids some duplication, and facilitates maintainance.
* Bug fix in `formXtViX' where matrix dimensions got dropped when
subsetting thereby messing up variance calculations for gamm fits in
which some group sizes were 1.
1.3-27
** Fix of nasty bug in large dataset handling with "tp" basis (introduced
in 1.3-26). Subsampling code was re-seeding RNG instead of intended
behaviour of saving RNG state and restoring it. Fixed and tested.
1.3-26
* modification to `gam' so that GCV/UBRE scores reported with all fixed
smoothing parameters are consistent with equivalent under s.p.
estimation.
* gam.fit3 modified to test for convergence of coefficients as well
as penalized deviance, otherwise in extreme cases the derivative
iterations can diverge.
* modifications of gam.setup, predict.gam and plot.gam to allow smooths
to contribute an offset term to the model (offset is returned from
smooth.construct or Predict.matrix as an "offset" attribute of
model/prediction matrix). This is useful for smooths which have known
boundary conditions of some sort.
* PredictMat can now handle NAs in a returned prediction matrix.
* vis.gam can handle NA's in predictions.
** Modification of large dataset handling for "tp" and "ts" bases. If
there are more that 3000 unique covariate combinations for a tprs then
3000 combinations are randomly sub-sampled, and used as the initial
knots for tprs basis construction. The same random number seed is used
every time, (R's RNG state is unaltered by this). Control of this is
usually via the `max.knots' (default 3000) and `seed' (default 1)
elements of the `xt' argument of `s'. In consequence, `max.tprs.knots'
has been removed from `gam.control'.
* Modification of `s' and `te' to allow an extra argument `xt' which can
contain extra information to pass to the basis constructors for smooths.
* removal of `full.call' from smooth.spec objects - it wasn't used
anywhere any more, and is a pain to maintain.
* removal of `full.formula' from the `gam' object - it is no longer used
anywhere and requires alot of code to construct.
1.3-25
* A bug in `null.space.dimension' caused prediction to fail for `s' terms
of 4 or more variables, unless the `m' argument was supplied explicitly
(and was large enough for the number of variables). Fixed.
1.3-24
* summary.gam modified so that it behaves correctly if fitting routines
detect and deal with rank deficiency in parameteric part of a model.
* spring cleaning of help files.
* gam.check modified to report more useful convergence diagnostics.
** `model.matrix.gam' added.
** "cr" basis constructor modified to use the same centering conditions
as other bases (sum to zero over covariates, rather than parameters
sum to zero). This makes centred confidence intervals for smooths, of
the sort used in plot.gam, behave in a similar way for all bases. With
the old "cr" centering constraint there could be high negative
correlation between coefficients of a centered smooth and the intercept:
this could make centred "cr" smooth CIs wider than CIs for other bases
(not really wrong, but disconcerting).
1.3-23
* step size correction bug fixed in gam.fit3. `Perfect' convergence could
cause the divergence control loop to fail: the divergence control loop
was asking for near strict decrease in the penalized deviance, which
could be numerically impossible to achieve if the algorithm had actually
converged completely.... fixed.
* minor doc bug fixes.
1.3-22
* Cheap but unneccesary code added to gdi.c and magic.c to stop
inappropriate uninitialized variable warnings from some compilers.
** Bad bug in gam.fit3 fixed. Prior weights of zero were not handled
correctly - prior weight vector should have been subsetted before
gdi call, but this didn't happen. Result was infinite derivatives
and fit failure. fixed.
* Related bug in gam.fit3: dropped observations not handled correctly
in deviance calculation, which can result in inappropriate step
halving. fixed.
* inner loop 3 in gam.fit2 and gam.fit3 modified so that step halving
continues until penalized deviance is at worst non-increasing.
* stupid bug in summary.gam, p-value calc. fixed.
1.3-21
* minor bug in gam.fit() - edf array not passed to `mgcv.find.theta'
if method "perf.magic" used - so wrong EDF used for theta estimation
with neagative binomial. fixed.
* Theta estimate added to family object of fitted gam if negative binomial
used...
* extract.lme.cov(2) modified to allow use with single level grouping
factors (not really sure when this is useful)
* bug in gam4objective called when using gam.method(outer="nlm") - never
used GCV.
* fixed bug in `newton' whereby immediate convergence actually caused
routine to fail.
* modified `smoothCon' and `predictMat' so that `qrc' attribute always
created if constraint absorption used, even if there are no constraints.
This attribute can then be used to test that there are no unabsorbed
constraints (e.g. in `gam.outer').
1.3-20
* Bad bug in `newton' - step halving set up so that step *never*
accepted (it still beat all previous methods in simulations)
* Minor bug in `newton' step limiting of Newton steps reduced step
to max component 1, rather than `maxNstep'.
* Some documentation fixes
1.3-19
*** SUBSTANTIAL CHANGE: Improved outer iteration added via gdi.c coupled
with gam.fit3. Exact first and second derivatives of GACV/GCV/UBRE/AIC
are now available via new iteration methods. These improve the
speed and reliability of fitting in the *generalized* additive model
case.
* numerous changes to NAMESPACE and gamm related functions to pass
codetools checks.
** gam.method() modified to allow GACV as an option for outer GCV
model selection.
* magic.c::mgcv_mmult modified so that all inner loop calculations are
optimal (i.e. inner loop pointers increments are all 1).
* `smooth.construct' functions for "cc" and "cr" smooths now increase `k'
to the minimum possible value (and warn), if it's too low.
** `gam' modified to allow passing of `mustart' etc to gam.fit and
gam.fit2, properly
* `gam' modified to fix a bug whereby fitting in two steps using argument
`G' could fail when some sp's are to be estimated and some fixed.
** an argument `in.out' added to `gam' to allow user initialization of
smoothing parameters when using `outer' iteration in the generalized
case. This can speed up analyses that rely on several refits of the same
model.
1.3-18
* gamm modifed so that weights dealt with properly if lme type varFunc
used. This is only possible in the non-generalized case, as gamm.Rd
now clarifies.
* slight modification to s() to add `width.cutoff=500' to `deparse'
* by variables not handled properly in p-spline example in
smooth.construct.Rd - fixed.
* bug fix in summary.gam.Rd example (pmax -> pmin)
* color example added to plot.gam.Rd
* bug fix in `smooth.construct.tensor.smooth.spec' - class "cyclic.smooth"
marginals no longer re-parameterized.
* `te' documentation modified to mention that marginal reparameterization
can destabilize tensor products.
1.3-17
* print.summary.gam prints estimated ranks more prettily (thanks Martin
Maechler)
** `fix.family.link' can now handle the `cauchit' link, and also appends a
third derivative of link function to the family (not yet used).
* `fix.family.var' now adds a second derivative of the link function to
the family (not yet used).
** `magic' modified to (i) accept an argument `rss.extra' which is added
to the RSS(squared norm) term in the GCV/UBRE or scale calculation; (ii)
accept argument `n.score' (defaults to number of data), the number to
use in place of the number of data in the GCV/UBRE calculation.
These are useful for dealing with very large data sets using
pseudo-model approaches.
* `trans' and `shift' arguments added to `plot.gam': allows, e.g. single
smooth models to be easily plotted on uncentred response scale.
* Some .Rd bug fixes.
** Addition of choose.k.Rd helpfile, including example code for diagnosing
overly restrictive choice of smoothing basis dimension `k'.
1.3-16
* bug fix in predict.gam documentation + example of how to predict from a
`gam' outside `R'.
1.3-15
* chol(A,pivot=TRUE) now (R 2.3.0) generates a warning if `A' is not +ve
definite. `mroot' modified to supress this (since it only calls
`chol(A,pivot=TRUE)' because `A' is usually +ve semi-definite).
1.3-14
* mat.c:mgcv_symeig modified to allow selection of the LAPACK routine
actually used: dsyevd is the routine used previously, and seems very
reliable. dsyevr is the faster, smaller more modern version, which it
seems possible to break... rest of code still calls dsyevd.
* Symbol registration added (thanks largely to Brian Ripley). Version
depends on R >= 2.3.0
1.3-13
* some doc changes
** The p-values for smooth terms had too low power sometimes. Modified
testing procedure so that testing rank is at most
ceiling(2*edf.for.term). This gives quite close to uniform p-value
distributions when the null is true, in simulations, without excessive
inflation of the p-values, relative to parametetric equivalents when
it is not. Still not really satisfactory.
1.3-12
* vis.gam could fail if the original model formula contained functions of
covariates, since vis.gam calls predict.gam with a newdata argument
based on the *model frame* of the model object. predict.gam now
recognises that this has happened and doesn't fail if newdata is a model
frame which contains, e.g. log(x) rather than x itself. offset handling
simplified as a result.
* prediction from te smooths could fail because of a bug in handling the
list of re-parameterization matrices for 1-D terms in
Predict.matrix.tensor.smooth. Fixed. (tensor product docs also updated)
* gamm did not handle s(...,fx=TRUE) terms properly, due to several
failures to count s(...,fx=FALSE) terms properly if there were fixed
terms present. Now fixed.
* In the gaussian additive mixed model case `gamm' now allows "ML" or
"REML" to be selected (and is slightly more self consistent in
handling the results of the two alternatives).
1.3-11
* added package doc file
* added French error message support (thanks to Philippe Grosjean), and
error message quotation characters (thanks to Brian Ripley.)
1.3-10
* a `constant' attribute has been added to the object returned by
predict.gam(...,type="terms"), although what is returned is still not an
exact match to what `predict.lm' would do.
** na.action handling made closer to glm/lm functions. In particular,
default for predict.gam is now to pad predictions with NA's as opposed
to dropping rows of newdata containing NA's.
* interpret.gam had a bug caused by a glitch in the terms.object
documentation (R <=2.2.0). Formulae such as y ~ a + b:a + s(x) could
cause failure. This was because attr(tf,"specials") is documented as
returning indices of specials in `terms'. It doesn't, it indexes
specials in the variables dimension of the attr(tf,"factors") table:
latter now used to translate.
* `by' variable use could fail unreasonably if a `by' variable was not of
mode `numeric': now coerced to numeric at appropriate times in smooth
constructors.
1.3-9
* constants multiplying TPRS basis functions were `unconventional' for d
odd in function eta() in tprs.c. The constants are immaterial if you are
using gam, gamm etc, but matter if you are trying to get out the
explicit representation of a TPRS term yourself (e.g. to differentiate
a smooth exactly).
1.3-8
* get.var() now checks that result is numeric or factor (avoids
occasional problems with variable names that are functions - e.g `t')
* fix.family.var and fix.family.link now pass through unaltered any family
already containing the extra derivative functions. Usually, to make a
family work with gam.fit2 it is only necessary to add a dvar function.
* defaults modified so that when using outer iteration, several performance
iteration steps are now used for initialization of smoothing parameters
etc. The number is controlled by gam.control(outerPIsteps). This tends
to lead to better starting values, especially with binary data. gam,
gam.fit and gam.control are modified.
* initial.sp modified to allow a more expensive intialization method, but
this is not currently used by gam.
* minor documentation changes (e.g. removal of full stops from titles)
1.3-7
* change to `pcls' example to account for model matrix rescaling changing
smoothing parameter sizes.
* `gamm' `control' argument set to use "L-BFGS-B" method if `lme' is using
`optim' (only does this if `nlminb' not present). Consequently `mgcv' now
depends on nlme_3.1-64 or above.
* improvement of the algorithm in `initial.sp'. Previously it was possible
for very low rank smoothers (e.g. k=3) to cause the initialization to
fail, because of poor handling of unpenalized parameters.
1.3-6
* pdIdnot class changed so that parameters are variances not standard
deviations - this makes for greater consistency with pdTens class, and
means that limits on notLog2 parameterization should mean the same thing
for both classes.
** niterEM set to 0 in lme calls. This is because EM steps in lme are not
set up to deal properly with user defined pdMat classes (latter
confirmed by DB).
1.3-5
** Improvements to anova and summary functions by Henric Nilsson
incorporated. Functions are now closer to glm equivalents, and
printing is more informative. See ?anova.gam and ?summary.gam.
* nlme 3.1-62 changed the optimizer underlying lme, so that indefintie
likelihoods cause problems. See ?logExp2 for the workaround.
- niterEM now reset to 25, since parameterization prevents parameters
wandering to +/- infinity (this is important as starting values for
Newton steps are now more critical, since reparameterization
introduces new local minima).
** smoothCon modified to rescale penalty coefficient matrices to have
similar `size' to X'X for each term. This is to try and ensure that
gamm is reasonably scale invariant in its behaviour, given the
logExp2 re-parameterization.
* magic dropped dimensions of an array inapproporiately - fixed.
* gam now checks that model does not have more coefficients than data.
1.3-4
* inst/CITATION file added. Some .Rd fixes
30/6/2005 1.3-3
* te() smooths were not always estimated correctly by gamm(): invariance
lost and different results to equivalent s() smooths. The problem seems
to lie in a sensitivity of lme() estimation to the absolute size of the
`S' attribute matrices of a pdTens class pdMat object: the problem did
not occur at the last revision of the pdTens class, and there are no
changes logged for nlme that could have caused it, so I guess it's down
to a change in something that lme calls in the base distribution.
To avoid the problem, smooth.construct.tensor.smooth.spec has been
modified to scale all marginal penalty matrices so that they have
largest singular value 1.
* Changes to GLMs in R 2.1.1 mean that if the response is an array, gam
could fail, due to failure of terms like w * X when w is and array
rather than a vector. Code modified accordingly.
* Outer iteration now suppresses some warnings, until the final fitted
model is obtained, in order to avoid printing warnings that actually
don't apply to the final fit.
* Version number reporting made (hopefully) more robust.
* pdconstruct.pdTens removed absolute lower limit on coef - replaced with
relative lower limit.
* moved tensor product constraint construction to BEFORE by variable
stuff in smooth.construct.tensor.smooth.spec.
1.3-1
* vcov had been left out of namespace - fixed.
* cr and cc smooths now trap the case in which the incorrect number of
knots are supplied to them.
* `s(.)' in a formula could cause a segfault, it get's trapped now,
hopefully it will be handled nicely at some point in the future. Thanks
Martin Maechler.
* wrong n reported in summary.gam() in the generalized case - fixed.
Thanks YK Chau.
1.3-0
*** The GCV/UBRE score used in the generalized case when fitting by
outer iteration (the default) in version 1.2 was based on the Pearson
statistic. It is prone to serious undersmoothing, particularly of binary
data. The default is now to use a GCV/UBRE score based on the deviance:
this performs much better, while still maintaining the enhanced
numerical convergence performance of outer iteration.
* The Pearson based scores are still available as an option (see
?gam.method)
* For the known scale parameter case the default UBRE score is now
just a linearly rescaled AIC criterion.
1.2-6
* Two bugs in smooth.sconstruct.tensor.smooth.spec: (i) incorrect
testing of class of smooth before re-parameterizing, so that cr smooths
were re-parameterized, when there is no need to; (ii) knots used in
re-parameterization were based on quantiles of the relevant marginal
covariate, which meant that repeated knots could be generated: now uses
quantiles of unique covariate values.
* Thanks to Henric Nilsson a bug in the documentation of magic.post.proc has
been fixed.
1.2-5
** Bug fix in gam.fit2: prior weights not subsetted for non-informative
data in GCV/UBRE calculation. Also plot.gam modified to allow for
consequent NA working residuals. Thanks to B. Stollenwerk for reporting
this bug.
** vcov.gam written by Henric Nilsson included... see ?vcov.gam
* Some minor documentation fixes.
* Some tweaking of tolerances for outer iteration (was too lax).
** Modification of the way predict.gam picks up variables.
(complication is that it should behave like other predict functions, but
warn if an incomplete prediction data frame is supplied -since latter
violates what white book says).
1.2-2
*** An alternative approach to GCV/UBRE optimization in the
*generalized* additive model case has been implemented. It leads to more
reliable convergence for models with concurvity problems, but is slower
than the old default `performance iteration'. Basically the GAM IRLS
process is iterated to convergence for each trial set of smoothing
parameters, and the derivatives of the GCV/UBRE score w.r.t. smoothing
parameters are calculated explicitly as part of the IRLS iteration. This
means that the GCV/UBRE optimization is now `outer' to the IRLS
iteration, rather than being performed on each working model of the IRLS
iteration. The faster `performance iteration' is still available as an
option. As a side effect, when using outer iteration, it is not possible
to find smoothing parameters that marginally improve on the GCV/UBRE
scores of the estimated ones by hand tuning: this improves the logical
self consistency of using GCV/UBRE scores for model selection purposes.
* To facilitate the expanded list of fitting methods, `gam' now has a
`method' argument requiring a 3 item list, specifying which method to
use for additive models, which for generalized additive models and if using
outer iteration, which optimization routine to use. See ?gam.method for
details. `gam.control' has also been modified accordingly.
*** By default all smoothing bases are now automatically
re-parameterized to absorb centering constraints on smooths into the
basis. This makes everything more modular, and is usually user
transparent. See ?gam.control to get the old behaviour.
** Tensor product smooths (te) now use a reparameterization of the
marginal smoothing bases, which ensures that the penalties of a tensor
product smooth retain the interpretation, in terms of function shape, of
the marginal penalties from which they are induced. In practice this
almost always improves MSE performance (at least for smooth underlying
functions.) See ?te to turn this off.
*** P-values reported by anova.gam and summary.gam are now based on
strictly frequentist calculations. This means that they are much better
justified theoretically, and are interpretable as ordinary frequentist
p-values. They are still conditional on smoothing parameters, however,
and are hence underestimates when smoothing parameters have been
estimated.
** Identifiability side conditions modified to work with all smooths
(including user defined). Now works by identifying possible dependencies
symbolically, but dealing with the resulting degeneracies numerically.
This allows full ANOVA decompositions of functions using tensor product
smooths, for example.
* summary.gam modified to deal with prior weights in adjusted r^2
calculation.
** `gam' object now contains `Ve' the frequentist covariance matrix of
the paremeter estimators, which is useful for p-value calculation. see
?gamObject and ?magic.post.proc for details.
* Now depends on R >=2.0.0
* Default residual plots modified in `gam.check'
** Added `cooks.distance.gam' function.
* Bug whereby te smooths ignored `by' variables is now fixed.
1.1-6
* Smoothing parameter initialization method changed in magic, to allow
better initialization of te() terms. This affects default gam fits.
* gamm and extract.lme.cov2 modified to work correctly when the
correlation structure applies to a finer grouping than the random
effects. (Example of this added to gamm help file)
* modifications of pdTens class. pdFactor.pdTens now returns a vector,
not a matrix in accordance with documentation (in nlme 3.1-52). Factors
are now always of form A=B'B (previously, could be A=BB') in accordance
with documentation (nlme 3.1-52). pdConstruct.pdTens now tests whether
initializing matrix is proportional to r.e. cov matrix or its inverse
and initializes appropriately. gamm fitting with te() class tested
extensively with modifications and nlme 3.1-52, and lme fits with pdTens
class tested against equivalent fits made using re-parameterization and
pdIdent class. In particular for gamm testing : model fits with single
argument te() terms now match their equivalent models using s() terms;
models fitted using gam() and gamm() match if gam() is called with the
gamm() estimated smoothing parameters.
* modifications of gamm() for compatibility with nlme 3.1-52: in
particular a work around to allow everything to work correctly with a
constructed formula object in lme call.
* some modifications of plot.gam to allow greater control of
appearance of plots of smooths of 2 variables.
* added argument `offset' to gam for further compatibility with
glm/lm.
* change to safe prediction for parameteric terms had a bug in offset
handling (offset not picked up if no newdata supplied, since model frame
not created in this case). Fixed. (thanks to Jim Young for this) 1.1-5
* predict.gam had a further bug introduced with parametric safe
prediction. Fixed by using a formula only containing the actual variable
names when collecting data for prediction (i.e. no terms like
`offset(x)')
1.1-5
* partial argument matching made col.shade be matched by col passed in
..in plot.gam, taking away user control of colors. 1.1-5
* 2d smooth plotting in plot.gam modified.
* plot.gam could fail with residuals=TRUE due to incorrect counting in
the code allowing use of termplot. plot.gam failed to prompt before a
newpage if there was only one smooth. gam and gamm .Rd files updated
slightly.
1.1-3
* extract.lme.cov2 could fail for random effect group sizes of 1
because submatrices with only a row or column lose their dimensions, and
because single number calls to diag() result in an identity matrix.
1.1-2
* Some model formulae constructed in interpret.gam and used in
facilitating safe prediction for parametric terms had the wrong
environment - this could cause gam to fail to find data when e.g. lm,
would find it. (thanks Thomas Maiwald)
* Some items were missing from the NAMESPACE file. (thanks Kurt
Hornik)
* A very simple formula.gam function added, purely to facilitate
better printing of anova method results under R 2.0.0.
1.1-1
* Due, no doubt, to gross moral turpitude on the part of the author,
gamm() calculated the complete estimated covariance matrix of the
response data explicitly, despite the fact that this matrix is usually rather
sparse. For large datasets this could easily require more memory than
was available, and huge computational expense to find the choleski
decomposition of the matrix. This has now been rectified: when the
covariance matrix has diagonal or block diagonal structure, then this is
exploited.
* Better examples have been added to gamm().
* Some documentation bugs were fixed.
1.1-0
Main changes are as follows. Note that `gam' object has been modified, so
old objects will not always work with version 1.1 functions.
** Two new smooth classes "cs" and "ts": these are like "cr" and "tp"
but can be penalized all the way down to zero degrees of freedom to
allow fully automatic model selection (more self consistent than having a
step.gam function).
* The gam object expanded to allow inheritance from type lm and type
glm, although QR related components of glm and lm are not available
because of the difference in fitting method between glm/lm and gam.
** An anova method for gam objects has been added, for *approximate*
hypothesis testing with GAMs.
** logLik.gam added (logLik.glm with df's fixed): enables AIC() to be
used with gam objects.
** plot.gam modified to allow plotting of order 1 parametric terms via
call to termplot.
* Thanks to Henric Nilsson option `shade' added to plot.gam
* predict.gam modified to allow safe prediction of parametric model
components (such as poly() terms).
* predict.gam type="terms" now works like predict.glm for parametric
components. (also some enhancements to facilitate calling from
termplot())
* Range of smoothing parameter estimation iteration methods expanded
to help with non-convergent cases --- see ?gam.convergence
* monotonic smoothing examples modified in light of above changes.
* gamm modified to allow offset terms.
* gamm bug fixed whereby terms in a model formula could get lost if
there were too many of them.
* gamm object modified in light of changes to gam object.
1.0-7
* Allows a model frame to be passed as `newdata' to predict.gam: it
must contain all the terms in the gam objects model frame, `model'.
* vis.gam() now passes a model frame to predict.gam and should be more
robust as a result. `view' and `cond' must contain names from
`names(x$model)' where x is the gam object.
1.0-6/5/4
* partial residuals modified to be IRLS residuals, weighted by IRLS
weights. This is a much better reflecton of the influence of residuals
than the raw IRLS residuals used before.
* gamm summary sorted out by using NextMethod to get around fact that
summary.pdMat can't be called directly (not in nlme namespace exports).
* niterPQL and verbosePQL arguments added to gamm to allow more
control of PQL iteration.
* backquote=TRUE added when deparsing to allow non-standard names.
(thanks: Brian Ripley)
* bug in gam corrected: now gives correct null deviance when an offset
is present. (thanks: Louise Burt)
* bug in smooth.construct.tp.smooth.spec corrected: k=2 caused a
segfault as the C code was reseting k to 3 (actually null space
dimension +1), and not enough space was being allocated in R to handle
the resultng returned objects. k reset in R code, with warning. (Thanks:
Jari Oksanen)
* predict.gam() now has "standard" data searching using a model frame
based on a fake formula produced from full.formula in the fitted object.
However it also warns if newdata is present but incomplete. This means
that if newdata does not meet White book specifications, you get a
warning, but the function behaves like predict.lm etc. predict.gam had
been segfaulting if variables were missing from newdata (Thanks: Andy
Liaw and BR)
* contour option added to vis.gam
* te smooths can be forced to use only a single penalty (theoretical
interest only - not recommended for practical use)
1.0-3
* Fixes bugs in handling graphics parameters in plot.gam()
* Adds option of partial residuals to plot.gam()
1.0-2/1
* Fixes a bug in evaluating variables of smooths, knots and by-variables.
1.0-0
*** Tensor product smooths - any bases available via s() terms in a gam
formula can be used as the basis for tensor product smooths of multiple
covariates. A separate wiggliness penalty and smoothing parameter is
associated with each `marginal' basis.
** Cyclic smoothers: penalized cubic regression splines which have the
same value and first two derivatives at their first and last knots.
*** An object oriented approach to handling smooth terms which allows
the user to add their own smooths. Smooth terms are constructed using
smooth.construct method functions, while predictions from individual
smooth terms are handled by predict.matrix method functions.
** p-splines implemented as the illustrative example for the above in
the help files.
*** A generalized additive mixed model function gamm() with estimation
via lme() in the normal-identity case and glmmPQL() otherwise. The main
aim of the function is to allow a defensible way of modelling correlated
error structures while using a GAM.
* The gam object itself has changed to facilitate the above. Most
information pertaining to smooth terms is now stored in a list of smooth
objects, whose classes depend on the bases used. The objects are not
back compatible, and neither are the new method functions. This has been done
in an attempt to minimize the scope for bugs, given the amount of time
available for maintenance.
** s() no longer supports old stlye (version <0.6) specification of
smooths (e.g. s(x,10|f)). This is in order to reduce the scope for
problems with user defined smooth classes.
* The mgcv() function now has an argument list more similar to magic().
* Function GAMsetup() has been removed.
* I've made a general attempt to make the R code a bit less like a
simultaneous translation from C.
0.9-5/4/3/2/1
* Mixtures of fixed degree of freedom and estimated degree of freedom
smooths did not work correctly with the perf.iter=FALSE option. Fixed.
* fx=TRUE not handled correctly by fit.method="magic": fixed.
* some fixes to GAMsetup and gam documentation.
* call re-instated to the fitted gam object to allow updating
* -Wall and -pedantic removed from Makevars as they are gcc specific.
* isolated call to Stop() replaced by call to stop()!
0.9-0
*** There is a new underlying smoothing parameter selection method,
based on pivoted QR decomposition and SVD methods implemented in LAPACK.
The method is more stable than the Wood (2000) method and allows the
user to fix some smoothing parameters while estimating others,
regularize the GAM fit in non-convergent cases and put lower bounds on
smoothing parameters. The new method can deal with rank deficient
problems, for example if there is a lack of identifiability between the
parametric and smooth parts of the model. See ?magic for fuller details.
The old method is still available, but gam() defaults to the new method.
* Note that the new method calls LAPACK routines directly, which means
that the package now depends on external linear algebra libraries,
rather than relying entirely on my linear algebra routines. This is a
good thing in terms of numerical robustness and speed, but does mean
that to install the package from source you need a BLAS library installed
and accesible to the linker. If you sucessfully installed R by building
from source then you should have no problem: you have everything already
installed, but occasionally users may have to install ATLAS in order to
install from source.
* Negative binomial GAMs now use the families supplied by the MASS library
and employ a fast integrated GCV based method for estiamting the
negative binomial parameter. See ?gam.neg.bin for details. The new
method seems to converge slightly more often than the old method, and
does so more quickly.
* persp.gam() has been replaced by a new routine vis.gam() which is
prettier, simpler and deals better with factor covariates and at all
with `by' variables.
* NA's can now be handled properly in a manner consistent with lm()
and glm() [thanks to Brian Ripley for pointing me in the right direction
here] and there is some internal tidying of GAM so that it's behavious
is more similar to glm() and lm().
* Users can now choose to `polish' gam model fits by adding an nlm()
based optimization after the usual Gu (2002) style `power iteration' to
find smoothing parameters. This second stage will typically result in a
slightly lower final GCV/UBRE score than the defualt method, but is much
slower. See ?gam.control for more information.
* The option to add a ridge penalty to the GAM fitting objective has been
added to help deal with some convergence issues that occur when the
linear predictor is essentially un-identifiable. see ?gam.control.
0.8-7
* There was a bug in the calculation of identifiability side conditions
that could lead to over constraint of smooths using `by' variables in
models with mixtures of smooths of different numbers of variables. This
has been fixed.
0.8-6
* Fixes a bug which occured with user supplied smoothing parameters, in
which the weight vector was omitted from part of the influence (hat)
matrix calculation. This could result in non-sensical variance
estimates.
* Stronger consistency checks introduced on estimated degrees of freedom.
0.8-5
* mgcv was using Machine() which is deprecated from R 1.6.0, this
version uses .Machine instead.
0.8-4
* There was a memory bug which could occur with the "cr" basis, in
which un-allocated memory was written to in the tps_g() routine in the
compiled C code - this occured when that routine was asked to clean up
its memory, when there was nothing to clean up. Thanks to Luke Tierney for
finding this problem and locating it to tps_g()!
* A very minor memory leak which occured when knots are used to start
a tps basis was fixed.
0.8-3
* Elements on leading diagonal of Hat/Influence matrix are now
returned in gam object.
* Over-zealous error trap introduced at 0.8-2, caused failure with
smoothless models.
0.8-2
* User can now supply smoothing parameters for all smooth terms (can't
have a mixture of supplied and estimated smoothing parameters). Feature
is useful if e.g. GCV/UBRE fails to produce sensible estimates.
* svd() replaced by La.svd() in summary.gam().
* a bug in the Lanczos iteration code meant that smooths behaved
poorly if the smooth had exactly one less degree of freedom than the
number of data (the wrong eigenvectors were retained in this case) -
this was a rather rare bug in practice!
* pcls() was not using sensible tolerances and svdroot() was using
tolerances incorrectly, leading to problems with pcls(), now fixed.
* prior weights were missing from the pearson residuals.
* Faulty by variable documentation fixed (have lost name of person who
let me know this, but thanks!)
* Scale factor removed from Pearson residual calculation for
consistancy with a higher proportion of authors.
* The proportion deviance explained has been added to summary.gam() as
a better measure than r-squared in most cases.
* Routine SANtest() has been removed (obsolete).
* A bug in the select option of plot.gam has been fixed.
0.8-1
* The GCV/UBRE score can develop phantom minima for some models: these
are minima in the score for the IRLS problem which suggest large
parameter changes, but which disappear if those large changes are
actually made. This problem occurs in some logistic regression models.
To aid convergence in such cases, gam.fit now switches to a cautious
mgcv optimization method if convergence has not been obtained in a user
defined number of iterations. The cautious mode selects the local
minimum of the GCV/UBRE closest to the previous minimum if multiple
minima are present. See gam.control for details about controlling
iterations.
* Option trace in gam.control now prints and plots more useful
information for diagnosing convergence problems.
* The one explicit formation of an inverse in the underlying multiple
GCV optimization has been replaced with something more stable (and
quicker).
* A bug in the calculation of side conditions has been fixed - this
caused a failure with models having parametric terms and terms like:
s(x)+s(z)+s(z,x).
* A bug whereby predict.gam simply failed to pick up offset terms has
been fixed.
* gam() now drops unused levels in factors.
* A bug in the conversion of svd convergence criteria between version
0.7-2 and 0.8-0 has been fixed.
* Memory leaks have been removed from the C code (thanks to the superb
dmalloc library).
* A bug that caused an undignified exit when 1-d smoothing with full
splines in 0.8-0 has been fixed.
0.8-0
* There was a problem on some platforms resulting from the default
compiler optimizations used by R. Specifically: floating point registers
can be used to store local variables. If the register is larger than a
double (as is the case for Intel 486 and up), this means that:
double a,b;
a=b;
if (a==b)
can evaluate as FALSE. The mgcv source code assumed that this could
never happen (it wouldn't under strict ieee fp compliance, for example).
As a result, for some models using the package compiled using some
compiler versions, the one dimensional "overall" smoothing parameter
search could fail, resulting in convergence failure, or undersmoothing.
The Windows version from CRAN was OK, but versions installed under Linux
could have problems. Version 0.8 does not make the problematic
assumption.
* The search for the optimal overall smoothing parameter has been
improved, providing better protection against local minima in the
GCV/UBRE score.
* Extra GCV/UBRE diagnostics are provided, along with a function
gam.check() for checking them.
* It is now possible for the user to supply "knots" to be used when
producing the t.p.r.s. basis, or for the cubic regression spline basis.
This makes it feasible to work with very large datasets using the
of the data. It also provides a mechanism for obtaining purely "knot
based" thin plate regression splines.
* A new mechanism is provided for allowing a smooth term to be
multiplied by a covariate within the model. Such "by" variables allow
smooths to be conditional on factors, for example.
* Formulae such as y~s(x)+s(z)+s(x,z) can now be used.
* The package now reports the UBRE score of a fitted model if UBRE was
used for smoothing parameter selection, and the GCV score otherwise.
* A new help page gam.models has been added.
* A bug whereby offsets in model formulae only worked if they were at
the end of the formulae has been fixed.
* A bug whereby weights could not be supplied in the model data frame
has been fixed.
* gam.fit has been upgraded using the R 1.5.0 version of glm.fit
* An error in the documentaion of xp in the gam object has been fixed,
in addition to numerous other changes to the documentation.
* The scoping rules employed by gam() have been brought into line with
lm() and glm by searching for variables in the environment of the model
formula rather than in the environment from which gam() was called -
usually these are the same, but not always.
* A bug in persp.gam() has been fixed, whereby slice information had
to be supplied in a particular order.
* All compiled code calls now specify package mgcv to avoid any
possibility of calling the wrong function.
* All examples now set the random number generator seed to facilitate
cross platform comparisons.
0.7-2
* T and F changed to TRUE and FALSE in code and examples.
* Minor predict.gam error fixed (didn't get correct fitted values if
called without new data and model contained multi-dimensional smooths).
0.7-1
* There was a somewhat over-zealous warning message in the single
smoothing parameter selection code - gave a warning everytime that GCV
suggested a smoothing parameter at the boundary of the search interval -
even if this GCV function was also flat. Fixed.
* The search range for 1-d smoothing parameter selection was too wide
- it was possible to give so little weight to the data that numerical
problems caused all parameters to be estimates as zero (along with the
edf for the term!). The range has been narrowed to something more sensible
[above warning should still be triggered if it is ever too narrow - but
this should not be possible].
* summary.gam() documentation extended a bit. p-values for smooths are
slightly improved, and an example included that shows the user how to
check them!
0.7-0
* The underlying multiple GCV/UBRE optimization method has been
considereably strengthened, as follows:
o First and second guess starting values for the relative
smoothing parameters have been improved.
o Steepest descent is used if either: i) the Hessian of the
objective is not positive definite, or (ii) Steps in the Newton direction
fails to improve the GCV/UBRE score after 4 step halvings (since in
this case the quadratic model is clearly poor).
o Newton steps are rescaled so that the largest step component
(in log relative smoothing parameters) is of size 5 if any step
components are >5. This avoids very large Newton steps that can occur
in flat regions of the objective.
o All steepest descent steps are initially scaled so that their
longest component is 1, this avoids long steps into flat regions of
the objective.
o MGCV Convergence diagnostics are returned from routines mgcv
and gam.
o In gam.fit() smoothing parameters are re-auto-initialized
during IRLS if they have become so far apart that some are likely to
be in flat parts of the GCV/UBRE score.
o A bug whereby poor second guesses at relative smoothing
parameters could lead to acceptance of the first guess at these
parameters has been removed.
o The user is warned if the initial smoothing parameter guesses
are not improved upon (can happen legitmately if all s.p.s should be
very high or very low.)
The end result of these changes is to make fits from gam much more
reliable (particularly when using the tprs basis available from version
0.6).
* A summary.gam and associated print function are provided. These
provide approximate p-values for all model terms.
* plot.gam now provides a mechanism for selecting single plots, and
allows jittering of rug plots.
* A bug that prevented models with no smooth terms from being fitted
has been removed.
* A scoping bug in gam.setup has been fixed.
* A bug preventing certain mixtures of the bases to be used has been
fixed.
* The neg.bin family has been renamed neg.binom to avoid masking a
function in the MASS library.
0.6-2
revisions from 0.6.1
* Relatively important fix in low level numerics. Under some circumstances
the Lanczos routines used to find the thin plate regression spline basis
could fail to converge or give wrong answers (many thanks to Charles
Paxton for spotting this). The problem was with an insufficiently stable
inverse iteration scheme used to find eigenvectors as part of the
Lanczos scheme. The scheme had been used because it was very fast:
unfortuantely stabilizing it is as computationally costly as simply
accumulating eigen-vectors with the eigen-values - hence the latter has
now been done. Some further examples also added.
0.6-1
* Junk files removed from src directory.
* 3 C++ style comments removed from tprs.c.
0.6-0
* Multi-dimesional smoothing is now available, using "thin plate
regression splines" (MS submitted). These are based on optimal
approximations to the thin-plate splines.
* gam formula syntax upgraded (see ?s ). Old syntax still works, with
the exception that if no df specified then the tprs basis is always used
by default.
* plot.gam can now deal with two dimensional smooth terms as well as
one dimensional smooths.
* persp.gam added to allow user to visualize slices through a gam
[Mike Lonergan]
* negative binomial family added [Mike Lonergan] - not quite as robust
as rest of families though [can have convergence problems].
* predict.gam now has an option to return the matrix mapping the
parameters to the linear predictor at the supplied covariate values.
* Variance calculation has been made more robust.
* Routine pcls added, for penalized, linearly constrained optimization
(e.g. monotonic splines).
* Residual method provided (there was a bug in the default - Thanks
Carmen Fernandez).
* The cubic regression spline basis behaved wrongly when extrapolating
[thanks Sharon Hedley]. This is now fixed.
* Tests included to check that there are enough unique covariate
combinations to support the users choise of smoothing basis dimension.
* Internal storage improved so that large numbers of zeroes are no
longer stored in arrays of matrices.
* Some method argument lists brought into line with the R default
versions.
0.5
* There was a bug in gam.fit(). The square roots of the correct iterative
weights were being used in place of the weights: the bug was
apparent because the sum of fitted values didn't always equal the sum of
the response data when using the canonical link (which it should as a
result of X'f=X'y when canonical link used and unpenalized). The bug has
been corrected, and the correction tested. This problem did not affect
(unweighted) additive models, only generalized additive models.
* There was a bug that caused a crash in the compiled code when there were
more than 8000 datapoints to fit. This has been fixed.
* The package now reports its version number when loaded into R.
* predict.gam() now returns predictions for the original covariate values
(used to fit the model) when called without new data.
* predict.gam() now allows type="response" as an argument - returning
predictions on the scale of the response variable.
* plot.gam() no-longer defaults to automatic page layout, use argument
pages=1 to get the old default behaviour.
* A bug that could cause a crash with the model formula y~s(x)-1 has been
fixed.
* Yet more sloppy practices are now allowed for naming variables in model
formulae. e.g. d$y ~ s(d$x) now works, although its not recommended.
* The GCV score is now reported by print.gam() (whether or not GCV was
actually used - it isn't the default for Poisson or binomial).
* plot.gam() modified to avoid prompting for input when not used
interactively.
0.4
* Transformations allowed on lhs of gam formulae .
* Argument order same as Splus gam.
* Search for data now designed to be like lm() , so you can now be quite
sloppy about where your data are.
* The above mean that Venables and Ripley examples can be run without
having to read the documentation for gam() so carefully!
* A bug in the standard error calculations for parametric terms in
predict.gam() is fixed.
* A serious bug in the handling of factors was fixed - it was previously
possible to obtain a rank deficient design matrix when using factors,
despite having specified an identifiable model.
* Some glitches when dealing with formulae containing offset() and/or I()
have been fixed.
* Fitting defaults can now be altered using gam.control when calling gam()
0.3-3
* Documentation updated, including removal of wrong information about
constraints and mgcv . Also some readability changes in code and no
smooths are now allowed.
0.3-2/1
* Allows all ways of specifying a family that glm() allows (previously
family=poisson or family="poisson" would fail). Some more documentation
fixes.
* 0.2 lost the end of long formulae (because of a difference in the way
that R and Splus deal with formulae). This is now fixed.
* A minor error that meant that QT() failed under some versions of Windows
is now fixed.
* All package functions now have help(). Also the help files have been
more carefully checked - version 0.2 actually contained no information
on how to write a GAM formula as a result of a single missing '}' in the
help file!
0.2
* Fixed d.f. regression splines allowed as part of gam() model
specification.
* Bug in knot placement algorithm fixed (caused crash with df close to
number of data).
* Replicate covariate values dealt with properly in gam()!
* Data search method in gam() revised - now looks in frame from which
gam() called.
* plot.gam() can now deal with missing variance estimates gracefully.
* Low (1,2) d.f. smooths dealt with gracefully by gam() - no longer cause
freeze or crash.
* Confidence intervals simulation tested for normal(identity),
poisson(log), binomial(logit) and gamma(log) cases. Average coverage
probabilities from 0.89 to 0.97 term by term, 0.93 to 0.96 "across the
model", for nominal 0.95.
* R documentation updated and tidied.