**  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 (1.9-0, July 2023)

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

1.9-4

*** scasm function provided for shape constrained additive smooth models, via
 linear and quadratic programming, with smoothness by extended Fellner-Schall. 

* Censored Box-Cox Gaussian, bcg, family introduced. 

* plot.gam now allows plotting of derivatives of 1-D smooth terms, to give 
  direct analogue of GLM regression coefficients, also some re-arrangement to
	allow use of bootstrap samples for uncertainty bands.

* predict.gam now allows a 'bs' field of simulated replicate parameter vectors,
  and direct computation of CIs at different levels. 	

* New scheme 2 for default 1-D plots. 

* bam now returns approx REML/ML for the model itself, rather than for the 
  working linearized model (other smoothness criteria still reported for 
  working model only). 

* Improved scale estimation initialization for extended families in bam: can
  make a noticable difference with 'tw' family in close to Poisson situation.  

* 4-D plotting code axis error fix (y axis labels were being recycled from 
  x axis labels).

* Long ignored argument 'shade' removed from 'plot.gam'.  

* cnorm/clog single vector response fix. 

* cnorm and cpois improved numerical stability.

* Fix of one letter typo in gam.outer call to efsud that caused offsets to be 
ignored. Thanks to Brian Montgomery for reporting the problem. 

* nat.param now warns is model matrix not full rank.

* sf and sw soap films given new classes to avoid clas=h with sf package 
	- thanks Gavin Simpson.

* Slight change to gammals scale link to make interpretation more
  straightforward.	

* gumbls link change and derivative correction.

* Added option to inSide to allow nmes for x and y to be provided.

* gam.vcomp modified to have a print method, rather than print itself - thanks
 to Gavin Simpson. 

* long deprecated function full.score removed. 

1.9-3

* fixes a read-one-beyond-allocated memory error (in a redundant check with no
  consequences)

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.

* added basic linear programming routine, lp, mainly to find initial values 
  under linear constraint. 

* default margins changed for plot.gam(...,pages=1)

* 'clog' censored logistic family added for AFT models. Written by Chris Shen.

* 'cpois' censored Poisson family added. 

* 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. Also adjusted to allow correct behaviour if model
  frame has been deleted.

* 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.