At developers meeting on 9/16 we discussed the following:
The group discussed the efforts to parallelize the OpenMx optimizer. Specifically, the following was discussed:
The backend now has the capability to duplicate a state object. With this capability and using OpenMP the parallelization efforts will begin at the largest grain of granularity - speeding up the Hessian calculation. OpenMP supports loop parallelization as opposed to task-based parallelization. The parallel FIML will need to be written in a loop-based style.
The group has been working on an alternative parallelization library to Rmpi called PiebaldMPI. PieBaldMPI has by Michael Spiegel. Based on some very rough benchmarks and it appears the alternative is 32x faster than Rmpi for small datasets and 1.5x faster for large datasets.
The group discussed installing Piwki analytics to get google-independent statistics about the website and OpenMx's popularity. This installation is underway.
The group discussed attending the useR! conference in the future. This could give OpenMx more exposure to the R-developers.
- The group discussed the updates made to the OpenMx documentation. Specifically, the following was discussed:
- The use of the 'you' in the documentation. While using 'you' can be seen as informal, the group decided it was appropriate for the user guide.
- Balancing the font size of the html documentation with what font size will fit on the page in the pdf documentation.
- Possible coloring issues for color blind readers.