‘../DESCRIPTION’ now uses Authors@R
.
‘../tests/SP500-ex.R’ tweaks to pass on 'M1mac' (Mac "Big Sur" M1 proc. arm64)
‘../tests/SP500-ex.R’: new main tweak: set all weights < 1e-10 to zero .
Should finally work on Solaris 32-bit
CLA(mu, covar, ..)
gets new option check.cov
which by default
is TRUE
and hence checks if the covariance matrix covar
is positive definite.
CLA()
for small cases may have "empty" new lambda even
when there free weights left. This is now caught, with a warning unless
new option check.f
is set to false. See new example mc3
.
the platform dependent tests in ‘../tests/SP500-ex.R’ now should also work on Solaris 32-bit.
modularized partly platform dependent testing into
moreSessionInfo()
in ‘./xtraR/platform-sessionInfo.R’;
updated to newest version from robustbase (so to detect KH's debian
OpenBLAS setup).
Not back-compatible changed result: The very
first “turning point” from the algorithm with (lambda
= NA
etc) is no longer stored as part of the result, compatibly
with the litterature and other implementations.
Added data set muS.10ex
orginally from Markowitz and Todd.
Added ‘NEWS’ file.
needed , drop=FALSE
in internal getMatrices()
,
e.g., for the *10ex
example.
Test works with old R 3.3.x.
In some cases such as muS.10ex
, the last turning
point with lambda=0
was not computed previously.
adapt ‘man/findSig.Rd’ example to R 3.6.0's new default
sample()
kind.
Run the ‘tests/*.R’ with R versions that use one OpenBLAS, ATLAS, or MKL (and skip some tests on these).
Add nice plot()
method base on Shi's MS_plot()
Literature references in ?CLA
Vectorized findSig()
and findMu()
.
findSig.1()
now works.
test also works in R configured with --no-long-double
.
First release to CRAN