OpenMx version 2.7.10 is now available through CRAN.
OpenMx is now able to check the first-order optimality conditions for solutions obtained by SLSQP when MxConstraints are involved. Note that SLSQP therefore may now warn about status code 6 in cases where formerly it did not.
Expectation-maximization has been made more generic.
A new function usable in MxAlgebras, Another new function usable in MxAlgebras, The way the WLS fitfunction handles MxConstraints has been improved.
When using the WLS fitfunction, OpenMx will not attempt to calculate a chi-square statistic when no standard errors are available.
Previously, The Nelder-Mead optimizer's default method for handling equality MxConstraints is now the newly implemented "GDsearch" method, which uses SLSQP as a subsidiary optimizer.
A bug involving
New features, performance improvements, and bug-fixes in OpenMx v2.7.10 include the following:
mxRobustLog(), has been implemented. mxRobustLog() returns the natural logarithm of its argument, unless the argument is zero, in which case it returns -745 instead of -Inf (see documentation for details). It is recommended for use with mixture models.
mxEvaluateOnGrid(), has been implemented. It is useful for quadrature integration.
omxParallelCI() now respects locally set mxOptions, and no longer unnecessarily redoes the primary optimization (for point estimates).
mxGetExpected() could fail in cases where one MxModel references a component of another MxModel, when both are submodels inside a "container" MxModel. This bug has been repaired.
mxGetExpected() now works with GREML expectation.
mxGetExpected() can now provide asymptotic ("infinite-time") covariance matrices for state-space expectations.
mxSE() now safely handles MxConstraints.
mxTryHard() now works with MxModels that use the Nelder-Mead optimizer.
mxCompare() now behaves more sensibly when it is comparing MxModels that have the same degrees-of-freedom.
mxRefModels() when using WLS with a multigroup MxModel has been repaired.
mxAutoStart() now works better with multigroup models.
Mac users of R v3.4 can download a "bleeding edge" developers' build of OpenMx from here, but note that the bleeding edge build is not a stable package release and is only minimally tested. Furthermore, it does not work on Mac OS X 10.11 "El Capitan."