########## Version 0.2-1 (July 2012) ###################
A makevar.win file was added to the src directory in order to proper
link required Lapack routines under Windows
########## Version 0.2-0 (June 2012) ###################
The following major changes were made in Version 0.2-0 of R package HiDimDA:
1) In adition to the original RFlda (linear discriminant analysis based
on a factor-model estimator of the correlation matrix), three new
classifiers were included in the package:
(i) Dlda which peforms Diagonal linear discriminant analysis.
(ii) Slda which performs Shrunken linear discriminant analysis
using the shrunken covariance estimators of Ledoit-Wolf/Fisher-Sun.
(iii) Mlda which performs Maximum uncertainty linear discriminant
analysis as proposed by Thomaz, Kitani and Gillies.
2) RFlda and the three new (Dlda, Slda and Mlda) discrimination routines,
now return by default canonical discriminant functions, while
direct-classification functions (the main output of RFlda in version 0.1-1
are returned only the the argument ldafun is set to "classification"
3) Two S3 classes named canldaRes and clldaRes (both a predict method) were created
in order to store in a uniform format the results of a canonical (canldaRes)
or classification (canldaRes) analysis for the four classifiers now available.
Four extensions of these classes named RFcanlda, Scanlda, RFcllda and Scllda
were also created in order to add the estimated covariance and precision (in
appropriate objects using compact representations) matrices to the results
returned by RFlda and Slda.
4) The examples and documentation were updated.
########## Version 0.1-1 (June 2011) ###################
The following changes were made in Version 0.1-0 of R package HiDimDA:
1) A bug was corrected in the print method for class SigFqInv: in
version 0.1.0 the print.SigFqInv method assumed incorrectly that
the 'x' argument was an object describing a covariance and not
a precision matrix.
2) The names of functions ForbSigap and ForbSigap1 (approximation of
covariance matrices by minimization of error Frobenius norms) were
changed to FrobSigAp and FrobSigAp1, and additional arguments were introduced,
given the user more control of optimization procedure.
Alias to the previous names were created in order to maintain backward compability.
A bug was corrected in the function FrobSigAp1: this version of the approximation
function that takes as input the matrix square root of the covariance to be
approximated (instead of the covariance itself, as function FrobSigAp does),
was not working properly in version 0.1.0.
3) The method RFlda for data frames and the methods LeftMult and
RightMult for classes SigFq and SigFqInv (specialized matrix products
for objects of classes SigFq and SigFqInv), that previously were internal
to the package, are exported in version 0.1.1.
4) The help files were improved and updated.