At developers meeting on 2/24 we discussed the following:
- The group discussed Tim Brick's progress with implementing analytic/smart gradients. The gradients save estimations but not iterations in the optimization. Michael Spiegel is going to pursue further performance testing to attempt to determine the possible speedup for various models and scripts.
- The group discussed the status of various grant proposals to fund OpenMx.
- The group discussed the name of several different coding styles within OpenMx. These styles have been discussed in previous meetings. The group decided on the name piecewise to refer to coding style which stores each piece of the model in a separate R variable and the constructs the mxModel using each piece/variable with a single concise call. The group also decided on the name stepwise to refer to the coding style which makes lots of calls to the mxModel constructor.
- Ross discussed his progress on translating the R implementation of solnp to the C programming language. He will be meeting with Tim Brick this week to go over the intricacies of the algorithm.
- Michael Spiegel discussed his progress with implementing multilevel modeling for covariance data. He is continuing to make progress.
- The group discussed how data will be handled for multilevel models. Users will need some mechanism to make sql-like queries about the data. At least three types of mechanisms will be needed:
- An expanded table based solution with a very small learning curve, that is wasteful of resource but effective for fairly small data sets.
- An SQL-based solution which allows users to access data in a database like manner.
- A solution that can handle "big data". The "big data" community has no decided on the best mechanism to handle this problem. We will be watching to see what they decide on an adopt their solution.