Application -- fisheries management --
The International Pacific Halibut Commission

Dr. Pat Sullivan is a population dynamicist with the International Pacific Halibut Commission in Seattle WA. (Pat has recently moved to Cornell University) Part of his job consists of constructing nonlinear statistical models to get information about the state of the Halibut fisheries for management and other scientific purposes. He is a long time Splus user who has been interviewed about Splus for PC Week (6/20/97) magazine

AD Model Builder is not intended to be a replacement for products like Splus. However there are certain areas, specifically the construction of computer code for nonlinear statistical models, where AD Model Builder is far superior to Splus. We are always happy to find Splus users who have discovered the advantages of AD Model Builder. To begin we asked Pat to give his general impressions of AD Model Builder.


Nothing can beat AD Model Builder for nonlinear estimation. I use it on everything from estimating two parameter growth curves to age and size structured population assessments containing hundreds of parameters. It is relatively simple to program in, allowing a variety of models to be quickly examined, while still using a sophisticated approach to optimization. The automatic differentiation, by which AD Model gets its name, increases both the speed and precision of the algorithm. But, best of all is the philosophy behind its development, which, when recognized, can be used to create quite versatile approaches to estimation that are on the cutting edge of statistics.

Otter Research:

Could you describe how AD Model Builder has helped you with a specific modeling task?


As a population dynamicist with the International Pacific Halibut Commission I was recently faced with the task of rapidly restructuring our stock assessment procedure (a multiparameter nonlinear optimization program). The original code was written in Fortran77 and developed by several authors over a twenty year period. But recent changes in halibut population dynamics forced us to rethink our modeling approach. Fortunately, ADMB was available. It allowed us to quickly develop a series of models that we were able to use, modify, test, and re-use. The ADBM language was easy to program in and was devoid of the matrix and array dimension declaration issues common to programming Fortran subroutines. This made modification of the program and program components straightforward, rapid, and accurate. The ADMB optimization algorithm performed superbly even for highly parameterized models. No other nonlinear optimization program I know of has these features or performs so well.