feedback

I was able to make some first tests with biomorpher and I have to say that it works already quite well and i really like it! Nice that it works also with galapagos and embryo!

I think i didn't understand everything so far but I will nevertheless try to give some feedback and then extend this list once oi gained more knowledge about it.

1) It would be nice if you could set and save a view vector for all models at the same time. that would make the comparison a lot easier because every model has different interesting views

2) Color could be helpful i think. Would it be possible to read the mesh vertex color values? If that not straight forward maybe the component could have another input which takes a list of colors for each mesh you input in the geometry input.

3) Is the closeness of two outcomes only computed by the closeness of the genes? Sometimes you have very ruff parameter spaces where a little shift in the genes can make quite big changes in the outcome. Would it be possible to compute closeness or similarity of two outcomes based on the outcome itself? For example compare the resulting meshes to each other?

4) It would be also great if you could replace a certain outcome with a solution that you manually generated by altering the sliders. That would be also a way to influence the direction.

5) I see that you are still working on the history. It would be great if one could use biomorpher as well to store you favorite versions of a definition. Something like a more advanced version of the state manager where you can also see the states and crossbreed them easily.

Best, Chris

Replies to This Discussion

Permalink Reply by John Harding on April 11, 2017 at 2:31

Hi Chris,

Many thanks for your feedback, much appreciated. Pleased that you like things so far... early days.

On your points:

1) Absolutely. Just to clarify, do you mean that whilst you are navigating a single design and you wish the others to be updated to that view also? There's a trade off here as we would need an extra button on the viewport (that we're trying to keep uncluttered as best we can)... I think it wouldn't be too bad. I'll put it on the enhancement list on git.

2) One of the best things about meshes in wpf is the simplicity... one of the worst things is lack of vertex shading which is a real pain. Getting into OpenGL, DirectX, etc. is not going to happen with the spare time I currently have, but what you suggest is quite a nice compromise (i.e. one colour per mesh). I'm still hoping I can figure a method to get vertex shading with wpf (stackoverflow shows others with similar struggles!), although it looks very tricky. We are looking into it!

3) It is, and yes this is a big problem when the mapping between genotype and phenotype is indirect (for example, the gh definition contains a random component or something). You can compare on the phenotype as you suggest, but this takes computational effort, such as some form of shape analysis. One hope is to compare genotype and phenotype 'solution' spaces (and thus give a measure of directness). On the TODO list!

4) Hmmm, not thought of that one. Good idea, a kind of import design into population. Thanks! That could work nicely.

5) Current idea is to maybe save a population history file, which would as you suggest act like a state manager of sorts. The data structure is quite complex in order to future proof the history approach, but we think we have a nice way of doing it and will be in the next release.

Thanks again for your feedback Chris, its much appreciated.

Best wishes,

John.

PS: What is that mesh shape you have there? Curious.

Permalink Reply by Christian Schmidts on April 11, 2017 at 8:19

Hi John,

thanks for you reply, glad to hear that the feedback can be useful in your development process.

Some supplementary notes:

to 3) Yes indeed sometimes slightly changes in some parameter ranges can have much stronger impact on the final design.

The problem is that usually the more "nonlinear" the mapping is the more interesting the result usually are because a definition with a very "linear" mapping doesn't have so much potential for surprise and unforeseen solutions.

It can be a random number as you stated, but also some like for example a slightly different point coordinate leads to different typology in a delaunay triangulation, now the strength of the impact also depends on the total sum of the delaunay points as well on how early the triangulation happens in the definition.

I seen that a shape analysis is a not easy at all not only technical but especially by defining the criteria. Looking forward to see your approach compare genotype and phenotype 'solution' spaces. Maybe an additional approach could be to have something like a gene manager where you can narrow down ranges of certain genes, weight them or freeze them.

to 4) and 5) looking forward to see the history once its ready. I think it could be beneficial to also be able to insert solution "by hand" for further crossbreeding and saving.

What i found myself doing quite often was taking solutions from the biomorpher and then tweaking one or more parameters "by hand" because then you can really see the impact and then you would like to have the possibly to bring that solution back into the biomorpher process.

I will go on testing and get back to you guys in some weeks! I attached you the my definition in case you want to have a look. Its needs kangaroo1, lunchbox, heteroptera and wb. Its more a graphical formal exercise:

best, chris

Attachments:

biomorpher_test01_chris.gh, 91 KB

Permalink Reply by John Harding on May 31, 2017 at 8:53

"What i found myself doing quite often was taking solutions from the biomorpher and then tweaking one or more parameters "by hand" because then you can really see the impact and then you would like to have the possibly to bring that solution back into the biomorpher process."

Just to mention Chris, we've just released 0.2.0 as an initial history test, but this feature will be included in 0.2.1. Thanks again for your help with this!

John.

RSS