keep your great job,

I depends on what you’re doing, but I’d probably just start a new project in eclipse, then copy GT.java from the source tar.gz file (http://www.psychicorigami.com/source/5KTuner.tar.gz) into the project’s directory. That should be enough really.

Just a warning though – I’m not sure this code will work on Android as is. It’s also not “good” code – as it was written to optimise for code size. It’s just a curiosity that you made find useful to look at.

cheers,

John

i’m a student and learning android application.
your video really interesting, and i want to know how to open your source code in eclipse environment.

Thanks & be my teacher.

Like a charm:

class AdminObjectForm(forms.ModelForm):
def __init__(self, *args, **kwargs):
super(AdminObjectForm, self).__init__(*args,**kwargs)
self.fields['graphic'].widget = AdminImageWidget()

Thank you!

well it sounds like you’re no longer doing the Travelling Salesman Problem then. That means the representation of your solution, objective function and operators need to be different. They might end up being similar to the TSP, but the key thing with this sort of stuff is to get your representation right – otherwise you’ll never get any good results.

You should probably store the solution as a list of line segments (equivalent of pairs of cities), then modify the objective function to calculate the total distance the plotter pen would travel to draw that solution (as that’s what you are trying to optimize).

The reverse sections operator would still then work. It’d just re-order which lines get drawn when. You might also want to consider adding an extra operator that will reverse the direction of a line too or building that in to the reverse sections operator.

Unfortunately their is a whole topic of research over choosing annealing schedules – it’s not very clear cut at all. It’s also been a while since I’ve had to pick any myself, but in the past I’d basically start of with a few values, then do multiple runs to see what seemed to work ok and didn’t get stuck too quickly. In short though the longer the annealing schedule (higher initial temperature and larger values of alpha) the better the final solution will be, but it will take longer…

cheers,

John

Firstly, thanks for this great write-up! It’s incredibly helpful. I’m trying to implement this for line segments rather than points (to draw more efficiently with a pen plotter) but my results don’t seem so optimal. I’m doing it such that lines are one-way roads and have approached things very simply:

I modified the cartesian matrix function so that for each start point of a line, the distances are calculated from the end point of that line to the start point of all the other lines. Should that be enough, or is the fact that I’m using “reversed sections” likely problematic considering that when you reverse the route between line a and line b you get a very different result? What strategy would you take instead?
The current strategy does something at least, and in a few places it seems to be doing alright…which leads me to a second question:
What is your general strategy for choosing start_anneal_temp and anneal_alpha? Is there some kind of intuitive ratio you use based on the number of points you are feeding in? I find that I sometimes get a divide-by-zero error by guessing the wrong ratio…

Many thanks!
Ryan