| NEWS | R Documentation |
PcaHubert will crash if X is 1-dimensional and mcd=FALSE: fixed.
Fixed 'noLD' issues in tlda.R and okg4.R
Fixed a problem with wrong scores in PcaProj() (reported by Matthieu Lesnoff <matthieu.lesnoff@gmail.com>)
Fixed a problem with nsamp="exact" or nsamp="best" in CovMve(), CovSest() (reported by Claudio Agostinelli) - these functions, differently from CovMcd, should not take non-numeric 'nsamp'
Added parameter 'control' to Linda - to select the robust location and covariance estimator to use in LDA. Now any estimator derived from class 'Cov' can be used, even such that are not in 'rrcov'. Return this parameter in the returned S4 object.
Linda returns now the location and covariance estimator used as 'covobj'. This is useful for controlling cross-validation, for example.
Linda and LdaClassic use generalized inverse if the common covariance matrix is singular.
Fixed an issue the 'predict' function.
Removed the dependence on packages 'cluster' and 'ellipse'.
Added data set diabetes; data set soil from package rrcovHD moved here.
Linear and quadratic discriminant analysis can use the MRCD estimates.
Fixed an issue with CovControlMcd(nsamp="deterministic") - this would not work, because nsamp was defined in the class definition as "numeric". Now it is "Cnumeric" - union of "character" and "numeric'.
Corrected the code for Minimum Regularized Covariance Determinant estimator (MRCD) -
CovMrcd() - the step of adjusting the eignevalues in r6pack() is excluded
now because it has no effect when n > p.
Added Minimum Regularized Covariance Determinant estimator (MRCD)
(Boudt et al. 2018) -
functions CovMrcd() and CovControlMrcd()
Added data set octane; data set olitos from package rrcovHD moved here.
The 'pairs' plot is now available for classical covariance matrix