User talk:Nat/Targeting Research Codename M79

Angular Targeting

I will dropped the VCS-GF gun for a while. Now with more unique gun, Angular Targeting/Factored! Current gun code is only 196 lines long! --Nat Pavasant 13:33, 2 October 2009 (UTC)

VCS-GF Gun

Inter-bin precise angle smoothing

Skilgannon inspire me on this. He said once that when using very precise calculation, we can dodge some GF gun taht have jsut 21 or 11 bins by moving to inter-bin. Here is Pros and Cons of low bins:

Pros.

1. Less memory
2. No need for bot width (should this be Cons. instead?)

Cons.

1. Not very precise

With hugh amout of buffers and segmentation, using low bins will reduce memory using a lot! But it will reduce its effective. Thinking of inter-bin precise angle smoothing, I don't know is Rednaxela implemented it already, but this is my ideas. Normal VCS-GF gun will find the greatest spike in stats. Instead of finding one greatest spike, or multiple spikes on bot width, this technique will find two great spikes.Then calculate firing bin like first index + 0.5 + normalize(2nd bin - 1st bin). This formular may be change if I can think of more precise calculation. Anyone have ideas? » Nat | Talk » 16:01, 10 April 2009 (UTC)

Edges smoothing

This was really old and discussion is hard to find (I think it on some bot page in old wiki that I was once read). The 'old' method is when wave hit enemy near GF1, mirror hit bin to the bin outside MEA. Also, when wave hit near GF0, mirror hit, too. Anyone have more ideas? » Nat | Talk » 16:01, 10 April 2009 (UTC)

Automate Segmentation

Just read that Fractal version later than 0.32 does this already. Reading (again, decompiled) source of Fractal. It's really complex code with a lot of deep recursion. (start thinking that DrussGT source is very easy to read, compare to other real OO bot) Even though I don't understand really of them (including the main part, chosing segment), I'll not copy-past because it close-source. Actually, I have difference algorithm on this. I think Fractal start recursive from most segment, if data fall in that segment is less than 50 hits, then disquilified that segment (for current iteraion). It stop when there are one or more segments with more than 50 visit, weight each segments with 'reliability' (don't really understand what it is). My algorithm is to chose the less flat. Say, if diff of min and max (normalize) of each segment is less than, say, 0.2, then disquilified that segment If more than one segment are quilified, then choose the least flat. Question or Ideas? (Not to bleme me about decompile ;)) » Nat | Talk » 16:01, 10 April 2009 (UTC)