goat 2.1 - an optimization solver component

We are happy to announce version 2.1 of our fast and versatile optimization component goat. Download is available as usual from

http://www.rechenraum.com/goat/download.html

This release adds a feature for the upcoming AA Rome Visiting School. If the objective dock is not connected, the user is asked to manually enter the value of the objective for the current variables, e.g. to optimize the response of a complex, real-world system to given input variables.

Connect the number sliders to the goat component, just as before, but do not connect anything to the "objective" dock. Disable the time stopping criteria (you may stop the optimization process anytime by pressing the "Cancel" button). Relax the other stopping criteria if used. Manually enter the score for the current variable values when asked to do so.

Happy Coding! simon

Replies to This Discussion

Permalink Reply by Rémy Maurcot on March 11, 2013 at 11:08am

Simon,

An image to represent it?

We can go a goal value now?

Permalink Reply by Simon Flöry on March 14, 2013 at 3:12am

Hi,

here is a screenshot, just give it a try, it is straight-forward to use.

Optimization towards a goal value is described in goat's FAQ in the meanwhile.

best, simon

Permalink Reply by Rémy Maurcot on March 14, 2013 at 4:36am

Simon,

I tested it is not at all clear. What are the operations performed to reach the goal in your case?

Maybe I do not understand.

Permalink Reply by Simon Flöry on March 15, 2013 at 2:49am

The standard approach is to define a parametric model: The input to the parametric model are the number sliders. The output is a single value, that is fed to goat. goat computes that configuration of input number sliders, that minimize / maximize the single output value.

Now, the new feature replaces the definition of the parametric model with user input. Whenever goat wants to know the objective value for a specific configuration of input variables, it pops up a dialog and asks the user to enter that single value rating the current input number sliders.

As an example (a trivial though), consider a number slider "x" ranging from -1 to 1. If you want to minimize x^2, you are supposed to enter the square of the current number slider value whenever goat asks you to do so. You will end up with the optimal solution x = 0. Of course you could define a simple a parametric model for that. But there are cases, where casting a real-world system into a paramatric model is not feasible. This is what the new feature was introduced for.

best, simon

RSS