makeDiiF() can now be given coefficients of inbreeding. Facilitates either:
f in a new generation of the pedigree if f has already been calculated for previous generationsproLik() (and proLik4()) REMOVED from the package.
asreml R package (which requires purchasing an expensive license) and due to the unstable behavior of asreml discovered when revising proLik() it was decided that nadiv should no longer support this other package in this way.gremlin R package for a replacement to asreml.makeA() was affected by a bug in Matrix <1.6-0
Matrix::chol2inv() bug highlighted a messy order of operations in nadivtcrossprod() and solve() which are just what chol2inv() uses
nadiv and take 1 step away from future bugs like thischol2inv() and tcrossprod(solve()) allocated same amount of memory according to profmem packagemakeAinv(), makeGGAinv(), and makeDiiF() caused coefficients of inbreeding (f) and Mendelian sampling variances (dii) to be ordered incorrectly
dii to be incorrect - namely the A matrix from makeA() and so consequently anything built from A (i.e., dominance and epistatic relatedness matrices through makeD() etc.)nadiv version 2.17.2 caused CRAN to archive due to error induced by Matrix updates
Matrix author) helpfully provided excellent explanations and patchesnadiv code caused by new methods for all.equal() and rbind2()is.numPed() and is.proLik()
inherits()pin() does not work with asreml 4 (should still work with asreml 3 model objects)
nadiv will not support this in the future as asreml 4 has vpredict()proLik4(), essentially the same as proLik(), but works on asreml v4
proLik() is kept to retain compatibility with asreml v3 model objectsmakeM() creates mutational effects relatedness matrix M directly
brms/JAGS etc. that require relatedness matrices (not their inverse) in mixed modelsMatrix non-virtual subclasses
Matrix 1.4-2 and specifically >=1.5-0prepPed() as often encountered/reported from use in optiSel
makeMinv() creates the inverse of the (additive) mutational effects relatedness matrix.makeT() creates the lower triangle of the cholesky factor of the additive genetic numerator relatedness matrix.sparseMatrix() instead of Matrix()
ggcontrib() to internally use makeT() to directly create a subset of the T matrix (which the subset is Q for the special setup for genetic groups).
geneDrop() conducts a gene dropping simulation down a user-supplied pedigree.simGG() now simulates phenotypes with heterogeneous additive genetic variances among the genetic groups (i.e., immigrant and resident groups)
makeGGAinv() added as a new function to construct genetic group-specific inverse relatedness matrices (Ainv).
?makeGGAinv), but below is a basic example:ggPed <- Q1988[-c(3:7), c("id", "damGG", "sireGG")]
AinvOut <- makeGGAinv(ggPed, ggroups = 2)$Ainv #<-- list with 2 Ainv matricesmakeTinv() and makeDiiF() functions
fmakeGGAinv().makeDiiF() creates the D matrix of the Cholesky factorization of the relatedness matrix below (i.e., A and the coefficients of inbreeding (diagonals-1 of A)makeTinv() creates Tinv of the Cholesky factorization of the inverse relatedness matrix below (i.e., Ainv)
1 / d_ii
Dinv <- D followed by Dinv@x <- 1 / D@xmakeTinv(Mrode2)
makeDiiF(Mrode2)simPedDFC() to allow more flexibility in designing pedigreesroxygen2 documentationmakeAinv() and makeS()
makeSd() and makeSdsim()
makeD() and makeDsim() accomplish for autosomesproLik() improved/bug fixed to find confidence limits
optimize() quitting too early with default tol argumentNA if confidence limits are not, in fact, found (e.g., for boundary parameters, variances that are not significantly greater than zero)plot.proLik() now includes vertical lines to better visualize CIsmakeAinv() and makeD()genAssign() and prunePed()
genAssign.numPed() code (and to some extent prunePed.numped())genAssign() and prunePed() are frequently called in many nadiv functions which operate on class ‘numPed’, this will have modest, but significant performance increasesprofvis for bringing my attention to this!fuzz argument.makeAinv.Rd help file or by running the following commands in R:?makeAinv # launches the help documentation
example(makeAinv) # runs the examples in the help documentation* Notably, fuzzy classification can be set to 'null', where each phantom parent is assigned to one genetic group with probability=1. This produces the same **Astar** matrix as regular genetic group methods (without fuzzy classification). See this demonstrated in the examples of the help documentation.makeAstarMult() function to create the inverse numerator relationship matrix with genetic groups (and possibly also fuzzy classification of genetic groups) through matrix multiplication instead of using direct algorithms to set this up.
ggcontrib() and makeAinv() to create Q and A^-1 directly, then multiplies these in such a way as to obtain Astar.makeAstarMult.Rd help file or run them in R with the command:F2009 dataset
makeAinv())F2009.Rd or in R type:simGG() function to simulate pedigree and phenotype when immigration occurs in a focal population
ggTutorial dataset (below)ggTutorial dataset
ggTutorial.Rd or in R type:LRTest() is now an exported function to do log-likelihood ratio testsmakeAinv().
fuzz argument
fuzz == NULL then dispatch the method makeAinv.default()fuzz == "matrix" | fuzz == "Matrix" then dispatch makeAinv.fuzzy()proLik() and the confidence interval estimation
LRTest() as basis of constrainFun() within proLik() so consistently define log-likelihood ratio test statisticsggcontrib() can now incorporate fuzzy classification of genetic groups
ggcontrib() have been changed. For more information and examples, read the help documentation ggcontrib.Rd or in R type:makeAinv()makeAinv() now can construct the augmented A-inverse matrix for genetic groups
makeAinv(), however, the defaults are set to produce the normal A-inverse. For more information and examples, read the help documentation makeAinv.Rd or in R type:makeAinv() - significant speed-up!numPed for pedigrees constructed by numPed().
is.numPed()) and re-ordering rows (ronPed()) currently availablenumPed, use ronPed() instead of typical subsetting operators (e.g., '[') to retain the class attribute numPed. For example:nPed <- numPed(Mrode2)
is.numPed(nPed) # TRUE
# re-order using typical R functions
nPed_sub <- nPed[order(nPed[, 2], nPed[, 3]), ]
is.numPed(nPed_sub) # FALSE; see help via ?'[' about dropping attributes
class(nPed_sub) # matrix
nPed_subnadiv <- ronPed(nPed, order(nPed[, 2], nPed[, 3]))
is.numPed(nPed_subnadiv) # TRUE
class(nPed_subnadiv) # numPedwarcolak dataset.
sex are now "M" and "F" instead of 0 & 1.warcolak.Rd.Q1988 from Quaas 1988 and (2) Mrode3 from Mrode (2005) chapter 3. See their descriptions in Q1988.Rd and Mrode3.Rd or in R type:heterogametic argument in makeS().
won’t work! Must now type:
as a minimum
TDtT(), a function to take the TDT’ Cholesky decomposition of a matrix (not currently exported).founderLine() which traces all individuals back to either the paternal or maternal foundergrfx() now has a new argument to allow user to supply the standard normal deviates instead of generating them within the function.
incidence = NULL... argument to drfx() so that arguments for the internal use of grfx() can be supplied to drfx().simPedHS() or simPedDFC().ggcontrib(), default is “matrix”pcc() return FALSE if the object (asreml) shows the log-likelihood did not convergesilent = FALSE argument to pcc() so that the default can be changed to not show messages
proLik() so that profile likelihoods should be “valleys” (instead of “hills”, as they were in versions previous to 2.13.3)makeAinv()
makeA() to use inverse of cholesky factorization of the A-inverse matrix
base::chol2inv() to obtain A
prepPed() to prepare pedigrees for use in other functionsmakeAinv()ggcontrib() so that genetic group contributions can be calculated
pin() and pcc() functions for the delta method and parameter value convergence checking, respectively, for asreml type REML models. Also added the pin.Rd and pcc.Rd help/documentation files.
pin() is not exported in this version (need nadiv:::pin() to use it)makeDufam(), but did not export it.
makeDomEpi() argument “Dinverse” to “invertD” to be similar to makeD()pcc() checks to constrainFun() so that only likelihood ratio test statistics of the constrained model returned if both the loglikelihood & parameter estimates have convergedfindDFC(), makeD(), makeDsim()FindDFC() to findDFC()