abrooks / wallframe-alpha

Find pleasing arrangement of pictures frames on dining room wall

WallFrame

I apologize for this code. It is not good code. Perhaps I shouldn’t be sharing it at all.

On the other hand, it helped me solve a real world problem, and I learned some things along the way. Maybe if you want to read it, and it doesn’t cause you to do anything that you would also have to apologize for, well then maybe sharing it is helpful.

The problem

We want to cover a dining room wall with a bunch of picture frames that were already obtained at yard sales, etc. The frames must not overlap, should have some gaps between them (but not too much or too little) and the resulting layout should be esthetically pleasing. Yes, that’s all rather vague. Except for the not overlapping thing, that’s pretty precise, but all the rest …​ yeah, vague.

Some "solutions"

The chouser.wallframe namespace is the big pile of code that contains practially no comments, poor variable names, naive algorithms, etc. There are a functions named go1 through go4 that take no arguments and are meant to be run at a REPL. They apply various algorithms and approaches, including a kind of force-directed graph as well as a sort of genetic algorithm approach.

The results are generally written to wall.html

Successful solution

After spending the evenings of about one week on this, we were still unhappy with the resulting layouts and so tried a different approach. We used a few of the functions to produce an SVG file, used inkscape to partition it into a few letter-sized pages, and printed them. Then we cut out the frame rectangles and manually arranged them on the page. Within a couple hours we had generated several satisfactory layouts, each of them better than any of the automated layouts up to that point.

Future work

Our faith in computers is unshaken, and I’m confident that with sufficient effort an automated solution could produce good layouts faster than we can manually. We believe the genetic approach has the most promise, but would require significant reworking. However, having solved our actual problem, we don’t have any current plans to improve the code.