At the Developers Meeting on 5/18/12 we discussed the following:
- Michael Spiegel has found and patched several memory leaks in the latest release of OpenMx. A new release of OpenMx with fixes for these will be released shortly. The group also discussed automating Valgrind checks for memory leaks into the test process.
- The group discussed the progress on implementing WLS. Ryne is looking into constructing a weight matrix for WLS and identifying statistical properties of the matrix. The group may look to several undergraduates and/or graduates students to help with this project.
- Steve and Michael Neal will be teaching an OpenMx workshop at the upcoming APS Conference.
- The group discussed Tim Brick's progress on implementing analytical gradients. The analytical gradients are not faster than gradients currently computed, but the two methods currently almost require the same amount of time. The overhead of the analytical gradients could be further reduced by turning down the verification level of npsol. Currently it rigorously double checks the gradients, which is not necessary with the analytical gradients.
- The group discussed updating the summary information when the RMSEA is negative. If the RMSEA is negative it will be set to 0. A field will also be added to the summary that will indicate if the RMSEA was set to zero because it was negative, or if when computed it was exactly 0.