Bug in galapagos in mutating binary gene ?

I tried some experiments with galapagos.

http://www.grasshopper3d.com/photo/galapagos-structual-1?context=user

http://www.grasshopper3d.com/photo/galapagos-structual-2?context=user

http://www.grasshopper3d.com/photo/galapagos-structual-3?context=la...

It is really excellent tool and very easy to use.

However, when I tried to perform structural topology optimization with the galapagos, and made many binary number sliders (with only integer 0-1 range), the galapagos did not mutated the binary values.

when mutation process was performed, the result was always 0 -> 0 , 1->1 ...

So, I had to made 1-4 slider, with 1~2 for non-existence and 3~4 for existence.

and, one more thing...

it was too hard for me to assign so many sliders to galapagos...

Maybe I need to develope a custom component for assigning a group of items for their on/off condition. Before starting, do you(david) have any idea about it?

Anyway, galapagos is great!!!

Thanks to david!

Replies to This Discussion

Permalink Reply by Mårten Nettelbladt on June 21, 2010 at 12:02pm

> it was too hard for me to assign so many sliders to galapagos...

Yeah, it would be nice to be able to select a bunch of sliders and connect them to Galapagos all at once.

Permalink Reply by David Rutten on June 21, 2010 at 12:26pm

Yeah, I'm also entertaining the ability to hook up a single number parameter containing any amount of numbers. That way you could quite easily define thousands of genes in one go. I'm not too sure about the performance though. Galapagos does seem to run significantly slower with more genes. But this is something that is potentially fixable.

--
David Rutten
david@mcneel.com
Turku, Finland

Permalink Reply by Mårten Nettelbladt on June 21, 2010 at 3:38pm

What about several inputs for the fitness value in order to define more than one goal?

Permalink Reply by David Rutten on June 22, 2010 at 12:45am

I don't understand how this would work. Can you elaborate?

--
David Rutten
david@mcneel.com
Turku, Finland

Permalink Reply by Mårten Nettelbladt on June 22, 2010 at 3:01pm

Please have a look at:
http://www.grasshopper3d.com/photo/mngalapagoscircles-1

In the example I had two goals that I wanted Galapagos to reach (at the same time). I tried to combine them into a single fitness-value, but found it a bit hard to get the balance/priority right. I'm not sure how it would work.
Perhaps Galapagos could find an array of solutions where the first one solved goal 1 perfectly and the last one goal 2. In between there would be a lot of solutions taking both goals into account. Perhaps you could "flicker through" them with a slider? (Just brainstorming...)

(edit) Oh, there is a video too:
http://www.grasshopper3d.com/video/mngalapagos01-1

Permalink Reply by David Rutten on June 23, 2010 at 6:11am

Ok, I sort of understand. But even with just two (conflicting?) fitness values, there might not be any single path between the two peaks. Finding the balance between two constraints is very much part of the fitness computation.

And then of course you have cases where there are 3 or 4 or N conflicting constraints pulling the solution in all directions.

I think it might be useful to have some sort of 'Fitness combination' component, which allows you to visually merge different fitness values with weighting etc. on screen, before the one remaining fitness target goes off into Galapagos.

The interface needs a lot of work in any case, I'll put this stuff on the list.

--
David Rutten
david@mcneel.com
Turku, Finland

Permalink Reply by Mårten Nettelbladt on June 25, 2010 at 3:24pm

What if Galapagos was set to work for say 60 minutes and then you could pull some fitness value sliders afterwards to see different results instantly? A bit like how "Multilight" works in Maxwell Render, see video here. (After the render is finished you can adjust the values of the different light sources in a scene and see the changes directly).

Permalink Reply by Tuan N. Tran on June 22, 2010 at 1:18am

David, thousands of genes in one go would be great! I've been dreaming about a "gene replicator". This component would have the following inputs: Interval, Float/Integer Toggle, Count. The output is a list of numbers. Thoughts?

Permalink Reply by David Rutten on June 23, 2010 at 6:12am

What would this component produce? A list of ascending numbers? Random numbers?

--
David Rutten
david@mcneel.com
Turku, Finland

Permalink Reply by Tuan N. Tran on June 26, 2010 at 11:54pm

Random numbers. What i'm thinking of is the ability to use the random component with galapagos.

Permalink Reply by David Rutten on June 28, 2010 at 6:30am

I still don't get it. You want a component that produces X random numbers in range R. But clearly you want these random numbers to be variable, otherwise you could just use the Random Component. But what controls the variability?

--
David Rutten
david@mcneel.com
Turku, Finland

Permalink Reply by Jacek Jaskólski on August 30, 2010 at 9:35am