Using ANN

Replies to This Discussion

Permalink Reply by Lorenzo Greco on May 24, 2017 at 2:34am

Hi Ralph,

You need to feed the NN a data tree of inputs and another one with results. Then you train it and afterwards you will be able to query it.

Hope the attached file makes more clarity! 

Attachments:

• HVOpt_LG.gh, 32 KB

Permalink Reply by Ralph Spencer Steenblik on May 26, 2017 at 8:31am

Thank you Lorenzo,

It really helped to see it applied to my problem, cheers. I am wondering if you have considered integrating a multiobjective evolutionary solver similar to Octopus into Dodo? Maybe I am missing the point but it seems that one must spell out all of the possible solutions to feed into the NN? Who is integrating a solver into their workflow with Dodo (The included solvers seem to be great but with a single target it seems that one would be limited to a linear solution set, because of the limitation one fitness)? I love the multiobject abilities of Octopus, and could see it being useful. Thoughts??, use cases??

Permalink Reply by Lorenzo Greco on May 29, 2017 at 8:17am

ANN are part of Machine Learning algorithms. Their use is to get the "fittest" solution to a given set of inputs after having being trained with already solved problems. It might seem silly, but this avoids recalculating new solutions every time but learning from the past.

Large optimization problems might take ages to solve and most of all, some are hard to solve because are hard to formulate. For instance, what would be the or goal for optimizing structure, architecture, mep, etc?

Consider you have to do it every time you are faced with a new project..

With ML you store the results, train the algo and in a fraction, you get the best fit.

ML and optimization are part of the AI field, they should not be seen as competitors, but two different tools that can be used together :)

Hope this was useful!