At developers meeting on 10/7 we discussed the following:
- The group is close to getting OpenMx thread-level parallelism working for calculating the hessian. Michael Spiegel, Tim Brick and Ross are working on addressing the final loose ends and testing the implementation to ensure it calculating the hessian values correctly.
- We discussed possible funding opportunities at the Census Board for OpenMx. OpenMx would serve as an improved data management and collaborative analysis environment for census data analysis.
- A bug has been identified in the implementation of $IndependenceLikelihood value from @output for a diagonal model diagonal model with freely estimated variances and means. The value should be the -2 log likelihood of a diagonal model however, we have found at least one case where it the calculation is wrong by several orders of magnitude. The group is working on addressing this. More information about the bug is available here.
- Next week will continue discuss implementing multi-level modeling in OpenMx. The group discussed implementing a small three-level model using covariance data as an initial step of determining workflow and specifying interfaces for multi-level modeling.