We are pleased to announce the official release of OpenMx version 2.20.6. Click here for instructions on how to install the package from our repository. As usual, our repository has package binaries for Windows and macOS, and source tarballs for Linux/GNU and other non-Mac Unix-likes, all of which come with the proprietary NPSOL optimizer. Alternately, users may install the fully open-source build of the new version from CRAN.
New Features and Performance Improvements Since v2.19.8:
- The memory footprint of GREML models has been significantly reduced, thanks to the final elimination of the incompletely implemented
runstate
slot from the MxModel object class. - New vignettes: factor analysis, debugging analytic derivatives (access vignettes with
browseVignettes("OpenMx")
). - By default, the MxComputeNumericDeriv compute step now skips calculating numeric derivatives if the fitfunction is able to provide analytic derivatives. Therefore, it is now possible to run a GREML model with the default compute plan, without OpenMx ever numerically differentiating the fitfunction, even if the user has not provided analytic derivatives of the covariance matrix.
dependencyModels
, a new argument toimxRobustSE()
, makes that function compatible with a broader subset of multigroup models.
Bug-fixes and Tweaks Since v2.19.8:
- Previously, when OpenMx would compute the WLS summary statistics from raw data, a bug could cause those statistics to be inaccurate. That bug has been repaired. Inaccuracy was particularly bad for large sample sizes. All WLS models should be re-run with this release and compared with prior results.
- There was formerly a bug that could make the GREML fitfunction return numerically incorrect fitfunction derivatives. This bug was only in effect when the number of available threads was large relative to the number of free parameters, and when argument
infoMatType="expected"
(which is not the default) was passed tomxFitFunctionGREML()
. This bug has been repaired. - The GREML fitfunction backend now no longer unnecessarily stores copies of the covariance matrix's derivatives during runtime.
- Regularization no longer adds the same parameter to the penalty more than once.
omxAssignFirstParameters()
now respects upper and lower bounds.mxAutoStart()
now works correctly with models that have only 1 endogenous variable.- Several memory leaks in the OpenMx backend have been patched.
- A number of small improvements have been made to the Newton-Raphson optimizer.
Known Issues:
summary()
does not work correctly when used on a WLS model that is using the interface for user-provided weight matrices that was new in version 2.19.8. This bug has been repaired in the OpenMx source repository, and will be repaired in the next OpenMx release.- The OpenMx Project's issue tracker can be found here.