From Robowiki
Jump to navigation Jump to search

The return of Corbos!

Oh my! Corbos returns with an update to Chalk! Salutations to you Corbos! Chalk has been a good testbed bot for me and will be interested to see it evolve more :) --Rednaxela 05:58, 26 May 2009 (UTC)

No, Corbos returns with a massive RoboRumble uploads =) » Nat | Talk » 10:59, 26 May 2009 (UTC)

Good to be back. We'll see if anything productive comes of it - Chalk's been stuck for a while. I've got a few ideas but they may simply ruin my performance enchancing bugs versus make Chalk more competitive. See you in the Rumble! Corbos 14:01, 26 May 2009 (UTC)


Hey, which license do you use? "Open Source" is not a license. Or you publish this robot under Public Domain? » Nat | Talk » 13:35, 26 May 2009 (UTC)

Nice catch. I changed the license to public domain. Corbos 14:01, 26 May 2009 (UTC)

Thrashed by PolishedRuby

Hey Corbos! I just looked here and it seems that Chalk gets thrashed by PolishedRuby but has a close battle and wins against RougeDC. Given the size of the difference I wouldn't write it off as random noise either. Since PolishedRuby is just a simple mirror movement with RougeDC's fancy gun ontop, that would indicate to me that your gun probably neeeds more work than your movement does, at least against top-range bots... Maybe this will help lead Chalk to a rise? :) --Rednaxela 12:43, 27 May 2009 (UTC)

If PolishedRuby keeps beating up on top bots, I might demand it get its own page. :-P --Voidious 15:39, 27 May 2009 (UTC)
Well, it's not like it quite beats Diamond... though Diamond does worse than the 'benchmark' of RougeDC vs PolishedRuby... (To explain why I call that a 'benchmark': since both use the same gun, and one is mirror, it gives a picture of how much disadvantage the mirror has due to poor movement when first getting to the mirror point and one-tick-offsets) --Rednaxela 15:51, 27 May 2009 (UTC)

Yeah, Chalk doesn't fair well against other Mirrors as well. I have a few theories:

  1. Chalk's gun kinda sucks
  2. Chalk's movement is tuned to Chalk-like guns so it imparts the benefit to the mirroring bot
  3. There's something strange going on

I'm not sure which I believe most. #1 doesn't feel right because his gun seems to hit other tough movements (he holds his own against some tough bots). His competitiveness may be mostly from movement; but to be honest, I've never thought of his movement as particularly strong. #2 seems unlikely because I don't even consider Chalk's gun when I work on movement. He's certainly not in his own testbed. Who knows though, maybe it reflects the way I think, so the movement is unintentionally tuned.

That leaves #3. I've noticed Chalk's performance is more erratic than the top bots. He'll take 70% of the points on a battle and then turn around and lose the next. Maybe I'm imagining it. I guess every bot scores better than expected against some bots and worse against others. Still, it feels like there's a bad assumption or bug hidden away. Time for some objective testing. Thanks for the feedback. --Corbos 17:37, 27 May 2009 (UTC)

Hm, in addition to your 'Chookious/Dookalk' plot, why not try out Targeting Challenge 2K7#Fast_Learning? That would be a pretty good way to prove/disprove #1 I think. --Rednaxela 17:59, 27 May 2009 (UTC)

Hmm, the ChalkiCape/ChalkiLightning results tell me the following:

  • The obvious: The gap between Dookious's Movement and your movement, creates a larger rumble score change, than the gap between Dookious's Gun and your gun.
  • Against your own movement, guns rank as follows Dookious > RougeDC > Chalk (compare PolishedRuby results of Chalk/ChalkiCape/ChalkiLightning to see why), with a sizable gap between each of them.
  • Unlike the above two it's not definitively proven, but I get the impression that while your movement may be stronger than your gun in rumble points, your gun looses more points against a strong adaptive movement than your movement loses against a strong adaptive shooter. In other words, if you're going for 'normal' rumble ranking, work on your movement, but if you're going for PL rumble ranking it may be your gun that needs the most work really.
  • It also tells me on the side, that RougeDC's movement sucks for it's ranking, but I already knew that :)

That's my analysis of the situation anyway :) --Rednaxela 15:17, 28 May 2009 (UTC)

I think you're right-on regarding Chalk. My TC2K7 Fast learning tests show the latest gun is significantly worse than his best. When I added my kd-ish-tree, my performance on the 500 round TC2K7 jumped. I proceeded to tweak without seeing a problem in 500 rounds but it obviously messed with my rumble-length bouts. His movement, though it could be improved, hasn't changed. Thanks. --Corbos 00:30, 29 May 2009 (UTC)
There are no threads on this page yet.