Fix issue with building for CRAN.
Fixed a bug in pickLSDstatistic that occurs when there is only one prediction for a level of a factor.combination
specified using the argument LSDby
.
Fix a bug that meant that the settings of the arguments allow.unconverged
, allow.fixedcorrelation
and update
for addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests were not being passed on to functions that they called. Set the defaults for the arguments allow.unconverged
, allow.fixedcorrelation
and update
to TRUE
for addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests to match the defaults that had in practice been the settings of these arguments for the functions that fit the spatial models.
Fix bug in fitting correlations models in addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests where correlations on both grid dimensions become bound.
Add code to remove a correlation function from a term when one of more of its correlation parameters are bound and the argument allow.fixedcorrelation
is FALSE
in fitting the function using addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests.
Add checks and fixes for ill-fitting correlations models in addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests, such as for when (i) bound residual term after fitting a correlation term, (ii) all random and residual terms in one dimesion are bound, and (iii) some correlations on both grid dimensions become bound.
Add check that spatial variance is not bound whilst the nugget variance is positive when fitting correlation models in addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests. If it is, fix the nugget variance.
Add the argument nugget.variance
to addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests to control the fitting of a nuggetvariance when fitting correlation models.
Add trace
argument addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests, primarily for tracking the fitting of correlation models.
Add traps for errors resulting from singular information matrices when fitting correlation models in addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests.
Fix bug in addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests when fitting metric correlation functions.
Fix bug in newfit.asreml that occurred when change to random model results in no random terms.
Modify the fitting of corb correlation models in addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests to make it more robust.
Fix a bug in infoCriteria.asreml and infoCriteria.list that caused an incorrect-dimensions
error when computing the full likelihood.
Make sure that, when false.pos.wt
is used with pickLSDstatistics and there are several values with the same weighted sum, LSDstatisics with less false positives are favoured.
Fix faulty links in manual.
Add EGLS.linTransform
to linTranform.alldiffs, predictPlus.asreml, predictPresent.asreml and pairdiffsTransform.alldiffs.
Add S3 method function isCompoundSymmetric.matrix.
Modify print.alldiffs so that the diagonal elements of pairwise differences, sed and p-values are printed as blank spaces.
Fix bug in checking loaded versions that occurs with R 4.4.0.
Update for CRAN submission.
Revise warning message in R2adj.asreml for when generalized inverse fails.
Add fix for bug in as.asrtests that caused a C stack overflow error when trying to update the wald.tab
component of an asrtests.obj
.
For chooseSpatialModelOnIC.asrtests, change the abbreviations for the two P-spline models from TPPCS
and TPP1LS
to TPPSC2
and TPPSL1
. If the call to chooseSpatialModelOnIC.asrtests includes the former abbreviations, they will be converted to the new abbreviatons and all output will be labelled with the new abbreviations.
Revise the deocumentation of the P-spline details for addSpatialModel.asrtests, addSpatialModelOnIC.asrtests or chooseSpatialModelOnIC.asrtests.
For R2adj.asreml, correct the check that terms to be included in the random model for calculating R2.adj
are in the fitted model so that a warning is no longer issued when a term for inclusion differs from a fitted term only in the order of the factors/variables comprising the terms.
Trap errors issued by svd
function from base R
in computing generalized inverses for R2adj.asreml. Instead, issue a warning and return NA
.
Fix a bug in estimateV that meant that terms involving the asreml at
function on other than the first variable in the term caused an error.
Instead of producing an error when the number of columns in the design matrix does not equal the number of coefficients for the term, now add columns of zeros for missing design-matrix columns. An error may still occur when this does not work.
Add checks in estimateV.asreml and R2adj.asreml for when the design matrix for a term is not available. Rather than issue an error, NA
is returned and an attribute missing.termmatrix
added that contains a list of the problematic terms. Also added an error to variofaces.asreml and bootREMLRT for when this problem occurs.
Extend the the include.which.fixed
and include.which.random
arguments of R2adj.asreml to process correctly formulae to update the existing fixed and random formula using a formula that includes the '.' and terms, typically a list of bracketed terms preceded by a '-' so that terms are removed from a formula.
Spell check the manual.
Revise the detection of variance parameters involving sections
in addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests.
Implement bespoke strategy for fitting corb
correlation models that first fits a random units length term and constrains the residual variance to one. Then, the corb models are fitted and, when corr.orders
is set to 0, up to 10 correlation bands are fitted depending on whether additonal bands improve an information criterion.
Fix a bug in estimateV when a corb
function is involved in a variance term and asreml version 4.2 is being used.
Minor corrections to documentation for makeTPPSplineMats.data.frame, addSpatialModel.asrtests, addSpatialModelOnIC.asrtests or chooseSpatialModelOnIC.asrtests.
Fixed bugs in passing the arguments for setvarianceterms.call via ... from addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests to changeTerms and changeModelOnIC.asrtests.
Minor revisions to documentation for R2adj.asreml.
Add an error for a NULL
vcov
component in the alldiffs.obj
supplied to the ratioTransform.alldiffs function.
Add printPlot
argument to plotPredictions.data.frame and change the function to return the plot.
Add function R2adj.asreml to calculate an adjusted coefficient of determination for linear mixed models.
Fixed bugs in estimateV for asr_str
terms with non-diagonal variance structures.
Rectify a problem in extracting the correct columns of the design matrix for some terms that affected estimateV.
Update the vignettes.
Add link to GitHub issues page for BugReports in DESCRIPTION file.
Fix a bug in plotLSDerrors so that the correct errors are now plotted.
Add argument false.pos.wt
to pickLSDstatistics.alldiffs, which when it is not NULL
results in mininzing the weighted sum of the false.pos and false.neg comparisons of pairwise differences.
Add the use of the R optimizer nloptr::bobyqa
to search for the optimal angles for the rotation of the eigenvlues of the penalty matrix when fitting cubic P-splines using addSpatialModel.asrtests, addSpatialModelOnIC.asrtests or chooseSpatialModelOnIC.asrtests. This provides an alternative to searching over a two-dimensional grid of angles.
Fix a bug in the fits used in searching for optimal rotation angles by doing the search before any spatial P-spline models are fitted.
There has been several changes to the arguments of makeTPPSplineMats.data.frame, addSpatialModel.asrtests, addSpatialModelOnIC.asrtests or chooseSpatialModelOnIC.asrtests as follows:
The default for ngridangles
has changed to NULL
to provide for using an optimizer to search for the optimal rotation.
The arguments dropRowterm
and dropColterm
have been replaced with the more general arguments dropFixed
and dropRandom
.
The default for asreml.option
has changed from mbf
to grp
. While grp
requires more storage that mbf
, it is easier to apply asreml
and asrtests
methods to fitted models when grp
is used.
Add the S3-method function convEffectNames2DataFrame.asreml.
Increment version for resubmission to CRAN after fixing an S3 method declaration error.
Revise manual pages to meet cran syntax requirements.
Add the argument usRandLinCoeffs
to addSpatialModel.asrtests, addSpatialModelOnIC.asrtests or chooseSpatialModelOnIC.asrtests to control whether to attempt the fit of the unstructured variance model for the constant and linear terms in random interactions with smooth terms. Also, added code to reject the model if any of the variances or covariances are bound.
Fix bug that prevented the fitting of unstructured variance model between the random constant and linear parameters in the interaction of these terms with the random smooth row and random column terms in a TPPCS
model.
For fitting in P-spline models using addSpatialModel.asrtests, addSpatialModelOnIC.asrtests or chooseSpatialModelOnIC.asrtests, change the function for fitting the covariance of the constant and the linear term from us
to corh
or corgh
.
Fix bug in investigating nugget variance for correlation models when there are sections
.
Further improve the handling of situations in which the residual variance or the spatial variance associated with a correlation term becomes singular or bound in fitting a spatial correlation model using addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests.
Add allow.corrsJointFit
argument as a device for dealing with fits that hang when attempting to fit simultaneously both correlations for the two grid dimesnions.
Internal recoding.
Added the maxit
option for asreml
to addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests, setting its default to 30 for these functions
to increase the chances that the models will converge.
Extensively revise the code to improve the handling of situations in which the residual variance or the spatial variance associated with a correlation term becomes singular or bound in fitting a spatial correlation model using addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests.
Revise checking for nugget variance when there are sections and the residual has a separate terms for each section.
Added a basic check for separate random and residual terms for sections and, if there are none, issue a warning that better fitting can result if they are included.
Make the default number of cores (ncores
) for parallel processing in bootREMLRT.asreml, simulate.asreml and variofaces.asreml.
Add the function convAsremlobj.asreml that recreates an existing asreml
object using the currently loaded asreml
version so that the object is compatible with this version.
Made the following modifications to addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests:
Tweaked the code underlying these functions for fitting local spatial variation models to make it more robust.
Added reporting the optimal rotation angles to the test.summary when rotateX
is TRUE
. Also, added theta(s) as an attribute to the asreml.object
.
Added the which.rotacriterion
so that one could choose between the deviance
, likelihood
, AIC
and BIC
as the criterion to use in determining the optimal roation angle for rotating the eigenvectors of the penalty matrix.
Add the capacity to parallel process, for each column angle, the search of the row angles. This option is only available for the asreml.option
set to grp
.
Removed the need to precede a call to these functions with a call makeTPPSplineMats.data.frame when asreml.option
is set to mbf
.
Fix a bug in fitting natural cubic smoothing spines (TPNCSS
) using these functions where only the first of the sections
had splines fitted to it.
Change from ordinal numbers to levels in labelling the fitting of correlation corr
spatial models when using these functions.
Modify makeTPPSplineMats.data.frame so that the default enviroment of the data.frames for the mbf
option of asreml.opt
is that of the function that calls makeTPPSplineMats.data.frame. Also, for mbf.env
set to NULL
, do not assign the mbf
data.frames; just include them in the list
that is returned.
Make the data.plus
component of the list
produced by makeTPPSplineMats.data.frame when asreml.option
is set to mbf
equivalent to that produced when asreml.option
is set to grp
.
Corrected bug in detecting Fixed (F)
variance components in rmboundary.asrtests.
Added to newfit.asreml, a test for bound and singular terms in a new model when update
is TRUE
and, if there are any, added code to refit the model as if update
was FALSE
, to see if any bound terms become unbound terms.
Clarified the dcoumentation of the update
argument in all functions that include it.
Minor internal code changes.
Add arguments rotateX
and theta
to makeTPPSplineMats.data.frame so that the eigenvectors of the penalty matrix used in generating the design matrices for the random terms could be rotated by the angles in theta
.
Add code to carry out a grid search to find the optimal angle by which to rotate the eigenvectors of the penalty matric when fitting P-splines. Add arguments rotateX
and ngridlines
to addSpatialModel.asrtests, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests to allow the use to request that grid seach be made for the optimal ritation angle and to control the size of the grid searched.
Fix a bug in fitting a correlation model that occurs when addSpatialModel is used to fit the model.
There remains an svd bug with the oneMKL libraries version 2023.1.0 that is causing problems for IClikelhood
set to "full"
.
Modifications to deal with classhes between multithreading and external linear algebra packages under Windows. The environmental variable OMP_NUM_THREADS needs to be set to 1 if parallel processing is to be undertaken on a Windows machine. This can be done for individual tasks by using Sys.setenv("OMP_NUM_THREADS" = 1)
prior to running the task; otherwise, the environmental variable can be set for all tasks in the Windows system via Advanced system settings
> Envirnomental Variables
.
There remains an svd bug with the oneMKL libraries version 2023.1.0 that is causing problems for IClikelhood
set to "full"
.
This version has known issues when the Intel MKL libraries are installed in the R installation directories.
Modified makeTPPSplineMats.data.frame to implement the use of mbf in fitting P-spline models for local spatial variation.
Modifications for ASReml-R version 4.2. Those that may impact upon users include the following:
The names of the columns of the design matrix has changed, as has the order of the columns.
character codes are now stored in both vparameters.type
and vparameters.con
,
making vpt.char and vpc.char obsolete and so no longer available.
For predict, the intercept-only classify and the name of the column in the
predictions.freame
, has changed from (Intercept)
to Intercept
.
The designation of estmates on the boder of the parameter space has changed from
fixed (F
) to (B
) e.g. correlations.
When the levels of a factor are being specified in terms with an at
, as
opposed to an ordinal number for the position of the level in the levels list, then
the levels must be enclosed in single or double quotes. Further, the names of the terms
will nclude single or double quotes.
Correct issue with alldiffs attributes not being retained when subsetting using subset.alldiffs.
Remove bloated system-generated testthat-problems.rds file so that the tarball to be submitted is less than 5 Mb.
Fix problem with deprecated function setvarianceterms.asreml when submitted to CRAN.
Add new vignette WheatSpatialModels
that illustrates the fitting of models for local spatial variation.
Generalize correlation spatial models to allow the independent specification of any asreml variance model for rows and columns. Improve the fitting of spatial correlation models, including changing the default correlation model to ar1
and postpone removing bundary terms until after all components of the correlarion model have been fitted.
Revised the spatial modelling to resolve several bugs associated with them when the residual specifies heterogeneous variance using eith idh
or dsum
.
Repurpose row.factor
and col.factor
arguments of addSpatialModel.asrtests
, addSpatialModelOnIC.asrtests
and chooseSpatialModelOnIC.asrtests
to specify the row and column factors for correlation model where required and add the arguments dropRowterm
and dropColterm
to specify the main effect terms to be dropped in fitting spline models.
Modified changeTerms.asrtests and changeModelOnIC.asrtests so that the bounds
and initial values
of terms specified using set.terms
without changing any terms in the model.
Fix problem detected during CRAN submission that found setvarianceterms.asreml was apparently an unregistered exported generic.
Fix bug in identifying residual terms for ASReml-R v. 4.x.
Fix bug in rmboundary.asrtests when removing a bound term results in no random terms.
Fix bug in fitting P-splines when difforder is one.
Change infoCriteria to use the deviance in calculating information criteria for family
set to asr_binomial or asr_poisson.
Add the argument transform.function
to addBacktransforms.alldiffs, allDifferences.data.frame, predictPlus.asreml and predictPresent.asreml so that predictions
and error.intervals
on the scale of the transform.function
can be backtransformed.
Add classify
argument to as.predictions.frame, checking that variables in the classify
are in the predictions.frame
and conversion of any variables in the classify
that are not factor to factors.
Update ggplot2 functions to use tidy evaluation.
Fix internal anomaly causing a Note on CRAN.
Add the argument useIntervals
to plotLSDerror.alldiffs.
Fix a bug in plotLSDerrors that resulted in not being able to determine which LSDs in the LSD component apply to the pairwise comparisons. It resulted in all comparisons being shown as n.a.
.
Rename makeTPSPlineXZMats.data.frame to makeTPPSplineMats.data.frame.
Tidy up code for trySpatial = "none".
Revise makeTPSPlineXZMats.data.frame to cover asreml.opt
set to grp
.
Fix bug in plotLSDerrors when LSDs have beeen computed for LSDtype
set to factor.combinations
.
Fix bug in testing whether an ar1 fit has been achieved in addSpatialModel.asrtests and chooseSpatialModelOnIC.asrtests.
Modify system tests to cater for for M1Mac numerical accuracy.
Fix bug in which the setting of IClikelihood was not being passed to infoCriteria calls in chooseSpatialModelOnIC.asrtests, resulting in always reporting the REML
likelihood.
Change that spatial.IC
component of the list
returned by chooseSpatialModelOnIC.asrtests to contains the information criteria for each spatial model that is fitted. Also, rename the components best
and best.AIC
to best.spatial.mod
and best.spatial.IC
.
Fix bug in reporting the likelihood when a TPNCSS model does not fit.
Add function addSpatialModel.asrtests that adds a spatial model to a supplied model without asessing the fit of the model.
Add arguments nestorder
, degree
and difforder
to functions addSpatialModelOnIC.asrtests and makeTPSPlineXZMats.data.frame, as well as argument nestorder
to chooseSpatialModelOnIC.asrtests. Add the option TPLV
to the argument trySpatial
from chooseSpatialModelOnIC.asrtests to allow a linear spatial model with first-order penalty to be fitted.
Fix bug in testing for fixed correlations that meant they were not always detected.
Fix error in changeModelsonIC.asrtests in the action when the model is Unswapped
- it was being reported as Changed random
.
Stop crashing of changeTerms.asrtests and ChangeModelOnIC.asrtests when rmboundary.asrtests unable to remove singular terms.
Fix bugs in choosing the best spatial model in chooseSpatialModelOnIC.asrtests. Add components best.spatial
and best.spatial.AIC
to the value returned by chooseSpatialModelOnIC.asrtests.
Add linear-row-by-linear-column-covariates interaction to the model fitted for TPNCSS option of addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests.
Add functions makeTPSPlineZMats.data.frame, addSpatialModelOnIC.asrtests and chooseSpatialModelOnIC.asrtests.
Revise test for fixed correlation so that allow.fixedcorrelation recognises any correlation term, rather than being restricted to ar1.
Compress pdf files and resubmit to CRAN.
Rebuild for R 4.2.2 and submission to CRAN.
Added function pickLSDstatistics.alldiffs.Rd.
Revise documentation for the arguments LSDtype
, LSDsupplied
, LSDby
and LSDstatistic
.
Fix bugs in function newfit.asreml that involves making changes to models that involve the at
function. This bug affected the functions changeTerms.asrtests, changeModelOnIC.asrtests, rmboundary.asrtests, testswapran.asrtests and testranfix.asrtests. In removing a term that involve a single lewvel of a factor in an at function, it is now possible to specify either the level or the level index.
Modify rmboundary.asrtests to allow the removal of random term for a single level of an at
function when the levels are specified singly in the random model.
Add function addto.test.summary.
Minor correction for resubmitting to CRAN.
Rebuild under R 4.2.1 for new submission.
Add functions plotLSDerrors.alldiffs and plotLSDerrors.data.frame.
Replace the use of the package reshape
with the use of reshape2
.
Replace X1
and X2
with Rows
and Columns
as the default names for the the first two columns of the data.frame
produced by plotPvalues.alldiffs, plotLSDs.alldiffs and plotLSDerrors.alldiffs and provided as input to plotPvalues.data.frame, plotLSDs.data.frame and plotLSDerrors.data.frame.
Change plotPvalues.alldiffs and plotLSDs.alldiffs to return a two component list, one component of which is a data.frame
that contains the statistics to be plotted and the other of which is a list
containing the plots that have be produced as ggplot
objects.
Add an error if all residual variances are bound.
Fix bug in chooseModelOnIC result from a change in as.vector for data.frames and lists in R 4.2.0.
Rebuild under R 4.1.2 for new submission.
Put in a trap for fixed-effects variances being inestimable when calculating the full likelihood.
Modify linTransform.alldiffs so that an intercept-only submodel can be fitted.
Revise the manual entry for exploreLSDs.alldiffs and for pairdiffsTransform.alldiffs.
Fix bug in newfit.asreml that re-orders the fixed model because of R limitations.
Fix problem in linTransform.alldiffs when the product of the numbers of levels in the classiify factors is large and not all combinations of the levels are observed.
Relax the requirement for LSDby
variable to be in the classify
to beng that they must be in the predictions
component of the alldiffs.obj
.
Fix bug in false.pos and false.neg for LSDtype
set to supplied
when LSDby
is not NULL
.
Add quantiles 25 (q25
) and 75 (q75
) to the values that LSDstatistic
can take and the exploreLSDs.alldiffs investigates.
Allow for multiple LSDstatistic
values when LSDtype
is factor.combination
and LSDby
is set.
Add sig.size
, sig.face
and sig.colour
arguments to plotPvalues.data.frame and plotPvalues.alldiffs.
Make sure that when predicted.values
and the associated vcov
and sed
in an alldiffs.object
have values that are close to zero as a result of the limitations of real number calculations, they are set to zero so as to avoid further calculation problems.
Add zero.tolerance
argument to linTransform.alldiffs.
Modify the test for fixed correlation from a test for Fixed
(F
) to cover also correlations that are bound
(B
) or singular
(S
).
Add tests on incoming fits for fixed correlation.
Support is no longer provided in asremlPlus for versions of ASReml-R prior to version 4.0.
Add a new argument allow.fixedcorrelation
to changeModelOnIC.alldiffs, newfit.asreml, chooseModel.asrtests, changeTerms.asrtests, testranfix.asrtests, testresidual.asrtests, testswapran.asrtests and reparamSigDevn.asrtests so that a model with a fixed correlation parameter may be rejected.
Fix a bug in pairdiffsTransform.alldiffs where the factors indexing the predictions were not generated correctly.
In changeModelOnIC.alldiffs, add detection of absent terms to be dropped and do not change the model if one or more are detected.
Fix bug in the calculation of false positives and negatives in exploreLSDs.alldiffs for LSDtype
set to overall
.
Change default for plotHistogram
in exploreLSDs.alldiffs to FALSE
.
Rebuild under R 4.1.1 for new submission.
Change number of comparisons symbol in an LSD.frame
from n
to c
.
Add False positives and negatives counts to LSD statistics.
Fix bug in redoErrorIntervals when only one factor in LSDby
.
Add function print.LSDdata.
Fixed bug in halfLeastSignificant.limit
when LSDtype
is factor.combination
and the assignedLSD
is zero to now ensure that the limit is NA
.
Reinstated passing of parameters from linTransform.alldiffs to redoErrorIntervals.alldiffs.
Fix newly introduced bug for LSDtype = "halfLeastSignificant"
when there is not a numeric covariate involved in the LSDby
argument.
Stop extra frequency plot produced by exploreLSDs.
Minor internal modification.
Fix bug for LSDtype = "halfLeastSignificant"
when there is a numeric covariate involved in the LSDby
argument.
Minor internal modification.
Fix bug in pairdiffsTransform when only two factors in classify
.
Add function subset.list.
Add functions exploreLSDs, plotLSDs.alldiffs, plotLSDs.matrix and plotLSDs.data.frame.
Incorporate arguments retain.zerLSDs
and zero.tolerance
into AllDifferences.data.frame, redoErrotIntervals.alldiffs amd exploreLSDs.alldiffs to allow control of the retention of the zero LSD values when calculating LSD summaries.
Deprecate the function asrtests, which has been replaced by the function as.asrtests.
There has been a substantial revision of the calculation and saving of the LSD component of an alldiffs
object.
The argument meanLSD.type
has been changed to LSDtype
. All functions that are S3 methods for alldiffs
objects will check for the meanLSD.type
attribute and, if found, change it to LSDtype
.
The option supplied
has been added to the argument LSDtype
so that the user can supply LSD values to replace those in the assignedLSD
column of the LSD
component.
The two new arguments LSDstatistic
and LSDaccuracy
have been added. The argument LSDstatistic
allows a choice of minimum
, mean
, median
or maximum
. The values of the nominated LSD statistics are stored in a column of the LSD component named assignedLSD
. The argument LSDaccuracy
allows the specification of a statistic summarizing the differences between the the LSDs and the assigned LSD value for each row of the LSD component. The values of the chosen measure are stored in a column of the LSD
component named accuracyLSD
. These two arguments have been added to the functions allDifferences.data.frame, as.alldiffs.object, as.alldiffs, linTransform.alldiffs, pairdiffsTransform, predictPlus.asreml, predictPresent.asreml, ratioTransform.alldiffs, recalcLSD.alldiffs and redoErrorIntervals.alldiffs.
The LSD
component is always calculated and stored as a component when pairwise
is TRUE
.
A new argument accuracy.threshold
has been introduced to control the addition of a column names LSDwarning
to the predictions
of an alldiffs
object. THis column indicates whether or not the an halfLeastSignificant
error.interval
is reliable as determined using the supplied value of accuracy.threshold
.
The attributes of an alldiffs object and of the predictions
component have been modified. The arguments LSDtype
, LSDstatistic
and LSDby
, if set, are attributes of an alldiffs
object. They are also attributes of a prediction.frame
and the backtransforms
component of an alldiffs
object, as is LSDvalues
, when halfLeastSignificant
error.intervals
are specified. The LSDvalues
attribute has the values used in calculating half-LSD intervals. The meanLSD
attribute has been removed.
A separate entry for LSD.frame
has been introduced into the manual to dedscribe the data.frame
that is stored in the LSD
component of an alldiffs
object.
Fix bug in facRecast that meant that the argument levels.order
was being ignored.
Fix a bug in facRecast where only the predictions component was appropriately re-ordered.
Add new functions ratioTransform.alldiffs and pairdiffsTransform.
Add new method to sort for a predictions.frame
: function sort.predictions.frame.
Change the argument sortWithinVals
to sortParallelToCombo
in sort.alldiffs.
Add the new argument sortNestinFactor
to sort.alldiffs.
Add new function getTestEntry.asrtests.
Fix bug in infoCriteria.asreml that resulted in values of Inf
for the full
likelihood.
Deprecate facRecode.alldiffs because it only make sense to redo the current levels or labels.
Tidy up function facRecast.alldiffs.
Add new function facRecast.alldiffs that supercedes facRecode.alldiffs.
Revise sort.alldiffs to use facRecast.alldiffs to reorder the alldiffs components. This fixed a bug when not all levels of the saortFactor
are present in the subset on which the sort is based.
Add argument alpha
to plotPvalues.data.frame and plotPvalues.alldiffs.
Add argument axis.text.size
to plotPvalues.data.frame and plotPvalues.alldiffs.
Modify testresidual to not change the variance model when there is no difference in the variance DF between the models being compared.
In as.asrtests, add variance DF to test.summary
when forming an asrtests
object.
Rebuild vignerttes.
Modify more links to asreml.
Modify link to asreml.
Revise doi specification in a link to a reference in the manual.
Increment version number for resubmission.
Increment version number for resubmission.
Recompile for CRAN submission.
Fix bugs in redoErrorIntervals.alldiff concerning whether or not meanLSD.type and LSDby settings result in the LSD component being recalculated.
Add rowname overall
to the LSD component when meanLSD.type
is set to overall
.
Modify URLs for CRAN submission.
Try to neutralize problem that occurs with colourise
in print.predictions.frame.
Rebuild for R 4.0.3 and recompile vignettes for CRAN submission.
Include a key
option in the which
that produces vparametersummary
rather than the full asremlsummary
.
Remove attributes from formuale when executing printFormulae.
Add the ChickpeaEnd data set.
Modify code so that error intervals are set to NA when the LSDs are zero.
Modify code for calculating LSDs so that all zero LSDs are allowed; this is needed when a linear transformation has resulted in predictions that are equal.
Rebuild the vignettes prior to CRAN submission.
Add envir
argument to getFormulae.asreml and printFormulae.asreml to make them more robust to where the objects in the asreml.obj
have been defined.
Fix bug in getFormulae.asreml and printFormulae.asreml that occurs when formula(e) have been supplied to asreml in an object.
Change the Maintainer email address.
Add ellipsis to as.asrtests call within an rmboundary.asrtests call so that arguments in the calling routine are passed through.
Add a trap in recalcWaldTab.asrtests for a NULL
wald.tab
before dealing with denDF
so that recalcWaldTab.asrtests throws a warning rather than an error.
Modify recalcWaldTab.asrtests to deal with a Wald table that has Chi-square instead of F statistics (this happens when denDF = "none"
.)
Increment version number for resubmission to CRAN.
Fix bug in changeTerms when both addFixed
and dropFixed
are NULL
.
Revisions to the manual.
Fix bug in forming test.summary.
Added the IClikelihood
argument to chooseModel.asrtests, reparamSigDevn.asrtests, rmboundary.asrtests, testranfix.asrtests, testresidual.asrtests, and testswapran.asrtests.
Fix bug in printFormulae when the formula is long.
Various bug fixes associated with asreml-R version 3.
Fix bug in processing a formula that includes an at
function.
Removed the material.diff
argument and the both
option from the which.IC
argument of changeModelOnIC.
Change likelihood
in infoCriteria to IClikelihood
to make it consistent with other functions.
Replace tol.diff
with material.diff
and set the default to 0.5 in changeModelOnIC.
Revise all vignettes.
Add a new vignette for the use of information criteria with the wheat experiment.
Fixed bugs in changeModelOnIC associated with addFixed
and dropFixed
and with dealing with unconverged models.
This version represents a major revision in that the facilities for employing information criteria (AIC and BIC) have been considerabley expanded. The infoCriteria method has been modified to add the possibility of calculating not only the the information criteria based on the REML likelihood, but those based on the full likelihood instead. There are now two infoCriteria methods, one for asreml
objects and the other for list
s of asreml
objects. An alternative is to use changeTerms.asrtests wiith IClikelihood
set to REML
or full
and the information criteria will be included in the The test.summary
of the asrtest
object. A new function changeModelOnIC has been included to allow model selection to be based on the information criteria, rather than on hypothesis tests. However, the two methods of selecting models can be interweaved.
Add argument omit.columns
to print.test.summary that allows the specification of columns to be omitted from the printout.
Various modifications to the internal code for the handling of the test.summary
component of an asrtest.object
to allow more flexibility in the make up of test.summary
.
Add function iterate.asrtests to execute additional iterations of the fit stored in an asreml.obj
within an asrtests
object; the model itself cannot be changed.
Add functions getFormulae.asreml and printFormulae.asreml to extract and print the formula
e stored in the call
of an asreml
object.
Add trace
argument to newfit.asreml and activated it for asreml-R version 4.
Made R >= 3.5.0 a requirement for asremlPlus.