You are here

OpenMx 1.3 beta released

The beta release of the OpenMx 1.3 series is available for download. Use the alternate installation command in an R terminal source(""). The new series of OpenMx is the result of nearly 3 months of development and much feedback from the user community.

In our development cycle a beta-release indicates that each new feature has been tested with some type of validation script in our test repository. However we need to broader community to shake out any final bugs before moving to the OpenMx 1.3.0 release. We always recommend verifying the output against the latest stable version of OpenMx before rushing to publication. Below is the full change log since the OpenMx 1.2.4 release. We have made numerous small improvements and bug fixes.

Beta Release 999.0.0-2126 (August 03, 2012)

  • added mxOption() for "Analytic Gradients" with possible values "Yes"/"No"
  • added 'cache' and 'cacheBack' arguments to mxEval()
  • added omxLocateParameters() function (see ?omxLocateParameters)
  • type='RAM' allowing 'manifestVars' argument to appear in a different order
    than in the observed covariance matrix.
  • bug fix in the identification of NA definition variables
  • the configuration mxRun(model, checkpoint = TRUE) writes a line
    in the checkpoint file at the conclusion of model optimization.
  • added "Always Checkpoint" to mxOptions() with values "Yes" or "No"
  • header for the checkpoint file will identify anonymous
    free parameters with the string modelName.matrixName[row,col]
  • omxGetParameters() and omxSetParameters() support anonymous
    free parameters
  • bugfix to omxRAMtoML() when input model has covariance data
  • implemented cov2cor in the OpenMx backend
  • mxOption "Major iterations" accepts either a value or a function
  • now tracking the MxAlgebra and MxMatrix objects that need to be updated
    when populating free parameters or updating definition variables.


This beta release contains two primary changes to the OpenMx backend.

  • In RAM models with observed covariance data, analytical gradients are calculated and supplied to the numerical optimizer. We are in the process of implementing analytical hessian calculations for these types of models. A future goal is to implement these changes for RAM models with raw data.
  • Each free parameter keeps track of the matrices and algebras that must be recomputed when the free parameter changes values. Each FIML objective function keeps track of the matrices and algebras that must be recomputed when the definition variables change values. In models with several hundred algebra expressions, we have observed a x15 speedup in runtime.

We accidentally created a dependency of the beta release to R 2.14.0 or greater. The dependency will be removed in the next beta release. Let us know if you encounter this problem and cannot upgrade R on your machine.