Moving up past 85 APS -- Not Easy
So I've put a lot of time in on XanderCat, what is supposed to be version 12. Wow is it hard to improve at this point. I had roughly 10 new ideas for improving and all of them had no positive effect. The only improvement so far has been in some refactoring I have done, which has simplified a few things, but not changed my score any. There are a few instances of robots that some other top competitors score nearly perfect against, but XanderCat only gets maybe 85 or 90. I think I may figure out how to make some improvements by closely scrutinizing those instances. I also want to take a closer look at each of my special case scenarios to ensure none of them are counter productive -- this includes my ram escape scenario, anti-mirror scenario, my targeting detector drive engagement scenario, and my bullet shielding protected gun wrapper. I don't expect to find anything wrong though; probably will just be verifying that they work as intended. Ho hum...what next...it seems like there should be some big gains to be had, given XanderCat is still 5 APS behind the leader, but darned if I can find them. I'll keep trying when I have the time though...
How you account distance danger? I get with Tomcat 0.7 APS when take in account min danger to enemy on path instead of distance on last point. Looks like it can help you.
I took a look at some comparisons vs DrussGT and Diamond. One thing that strikes me is that in a bunch of them, your survival is much lower, and DrussGT/Diamond get very high survival. This could just be a side effect of not being strong vs those bots, but it could also be a deficiency in your bullet power selection. Actually one of my most recent improvements to Diamond was in my bullet power formula.
It definitely seems like there are a bunch of matchups that should give some clues. And I agree distancing is important. I gained a bunch just from changing the shape of the curve I use when factoring distance into my danger calculation.
Alright -- nobody run any more Rumble battles, because I currently have the top score against DrussGT. :-P I still don't know what I did that made me lose about 0.3 APS, but even if I figure it out, I won't be putting up a new version for a little while. I have some new changes I'm working on and I intend to run a gazillion test battles before the next time I release anything, so for awhile I'll be out of the top 10.
I just compared the versions 11.15 and 11.11 with eachother, and the new version is much worse against 'simple' bots. 97% dropping to 88% is very very bad, the other quadrupled its score! My tip: run battles against f.e. Supersample.SuperCorners and watch what is happening. Loosing rounds against such an opponent should not happen and the reason is hopefully easy to see (radarlock, gundirection).
GrubbmGait -- good advice. Definitey something wrong there. There is also one battle against Timbot where XanderCat only scored a 7 APS. I would guess an error slipped through where an exception can occur causing XanderCat to freeze and lose a round occasionally (or in the case of the Timbot instance, get shutdown completely). Whatever it is, it definitely needs to be fixed.
I'm pretty certain I found and fixed the primary bug (there may be an underlying secondary bug to fix, but the exception should be gone). I know I said I wasn't going to release another version pre version 12, but since it is an exception that is occurring, I feel I should go ahead and fix it. So I'm putting out 11.16 today.
