next up previous contents
Next: IESAFE Up: Decisions with Locally Weighted Previous: Forward models

Maximizing or Minimizing a Learned Model

We have seen how to search a forward model to get a target tracking controller, but that is just one of many things Optex can do with a memory based model. Suppose that instead of tracking a target, we'd like to maximize the output of a certain system. This would be the case if the action was a temperature control on a chemical fermentation process, and the behavior was the process yield. To do this, we modify the Taskspec to describe the new problem.

Optex -> Edit -> Taskspec -> Maximize expression
                             OUTPUT Variable weight 1.0
                             OUTPUT Target weight   0.0
                             INPUT Low              0.0

Optex suggests action 0.4, and predicts that it will yield behavior 0.93. You can ask Vizier for a prediction, or check fig. 28 to verify the Optex selection.

Jeff Schneider
Fri Feb 7 18:00:08 EST 1997