Tomcat's future

Jump to navigation Jump to search

Tomcat's future

Ok now i can say, that Tomcat reach 15th place. And it's means that easy level is completed and now starts really hard challenge. And there is future plans of Tomcat's development:

  • Research kNN vs RS - i think it will first thing, which i will do. But my bycicle inventor's soul encourage me to implement my own vantage tree to NNS
  • use in targeting same data source choose algorithm, like in movement. I think this will be second thing
  • NN or GA approximator for simple (HOT, Linear and Circular) guns
  • return to my new single-tick PIF gun research, which give me 66% hit rate against Crazy and 99% against Walls
  • make real good movement (100% against HOT, >99% againt DevilFish and DoctorBob)
  • try to implement RS/kNN-GF gun
  • Bullet shielding. I leave it at the end, because this points i can get with gurantee. For Tomcat it's more significant improvement, because he successfully exploit information about places, where enemy is not fire
  • Multiple wave surfing. I leave it lasty with same reason. Please, who done this, say how much APS you get with this improvement
Jdev12:27, 5 October 2011

And main: it's time to sleep a little:)

Jdev12:54, 5 October 2011
 

Good job, and don't worry, if you get too close i'll write a gun and just jump up into the top 10 (Nene uses Raiko's gun, who ranks #85).

Though I don't want to, as I have so much more work to do in the movement.

Chase-san15:10, 5 October 2011
 

Chase, kNN rulez! So you may start write yours' gun:p

Jdev12:11, 6 October 2011
 

Wow, that's one hell of a gun update. Nice work, and congrats on top 10!

Voidious15:22, 6 October 2011
 

Thanks:) It was easier, than i expecting:)

Jdev15:25, 6 October 2011
 

Oh nice job.

Chase-san19:53, 6 October 2011
 

Interesting, but change RS to kNN in movement give me -1 APS... So getting to club 2100 postponed a little:)

Jdev10:23, 7 October 2011
 

Congrats! I think you are the first one to ever push Shadow out of the top-10! And you are making it harder for me to reach top-5 (if I ever get the time).

GrubbmGait17:16, 7 October 2011
 

Another idea to try: every time, when info about enemy fire angle collected, calculate intersection between profiles produced by hit & visit logs with same segmentations. Theoretically, intersection area must correlate with quality of segmentation. Or, if it can be executed in bounds of turn limit, generate every tick random segmentation, build visit and hit logs with this segmentation and see, how they intersects. In both cases use visit logs with segmentation with best intersection

Jdev08:41, 17 October 2011
 

As for the gains from multiple wave surfing. I don't know the APS figures, but I know this: When I about ideas for how to improve my bot (on Google Plus), Voidious mentioned this as his first item. I'm sure that must mean something. =) That said, CassiusClay does surf multiple waves, in a very simple manner (weighting each wave based on proximity).

PEZ19:13, 3 November 2011
 

Next week my girlfriend going to business trip for a week, so... it's all ok and no reason to spend time on robocode (looks to Voidious and Skilgannon :))

But Tomcat begins its hunt. So...

List to closest todos with high probability of improvements:

  1. If it's possible, go on current wave to position, starts from which allows pass safest point on current wave in best direction on second wave
  2. Devide current gun to virtual guns array with main and AS guns
  3. implement multiply wave surfing

List of future plans for investigation:

  1. reimplement with new vision my single tick pif idea and try it for targeting and precise enemy movement prediction in movement
  2. play with idea of guess factor pattern matching or use gf pattern as context for kNN-PIF
Jdev12:23, 17 November 2011