TE111206

Attendees

Jeremy Beaulieu (jeremy.beaulieu@yale.edu)
Naim Matasci (nmatasci@iplantcollaborative.org)
Brian O'Meara (omeara.brian@gmail.com)
Barb Banbury (bbanbury@utk.edu)
Kurt Michels

Agenda

Progress with the optimization project

The optimization project is on track and Kurt has produced the code to generate the plots of those regions where optimizers fail. He also reports that the initial value has an effect of the optimization convergence and therefore it's worth exploring that parameter as well. However given the computational needs, the code will need to be parallelized. Naim to describe the strategy on a wiki page. Furthermore, we should avoid focusing exclusively on geiger, as we want to provide a community-wide perspective: Ape functions should also be tested. Also, Brian and Jeremy suggested that daily builds of the code are committed to the R-forge repo.
Kurt could show that the problem with optimization is pervasive and affect multiple optimizers. He will now confirm the problem with datasets of different sizes and produce figures to illustrate the problem. The figures will be he base for the manuscript. Brian has a method to produce variable denisity contour plots that could be useful.
Naim will test the Geospiza dataset on Mac OSX and Windows to ensure that the problem is not platform specific.