Talk:RoboRumble/Participants/GigaRumble
- [View source↑]
- [History↑]
Contents
Thread title | Replies | Last modified |
---|---|---|
Nominations 2017-11-25 | 15 | 15:27, 30 November 2017 |
Going official | 7 | 22:53, 7 May 2013 |
Nominations | 4 | 14:56, 26 March 2013 |
Impact of bullet shielding | 2 | 23:12, 25 March 2013 |
GigaRumble results | 3 | 23:14, 4 August 2012 |
First page |
Previous page |
Next page |
Last page |
Seems that an update is appropriate, as rsalesc.Knight(#4) and cb.Firestarter(#5) would enter. Next to that, updated versions of aaa.ScalarBot, gh.GresSuffurd and jk.Neuromancer and the re-entrance of ags.Midboss. Alas ags.RougeDC, kc.Hydra and pulsar.pulsarMax will (temporarily ?) disappear from the list. I will update the list tonight and try to start running a client.
Updated to Java 8, but I can't run battles for rsalesc.Roborio and xander.XanderCat. Seems that they are made with Java 9 specific things. So if anyone else with Java 9 can also run Gigarumble, the ranking can become complete.
Team rumble and Twin Duel are broken in Java 9. I won't upgrade until the Robocode engine is fixed.
The battle bugs (Bug-389, Bug-390) have been fixed, but only with version 1.9.3.0. The rumble is still stuck on version 1.9.2.6, because the server won't be upgraded until a regression analysis is made, and no one has yet volunteered to do that.
Both works fine here in Java 8, so maybe restarting the rumble/cleaning the caches helps? btw I would like to run Gigarumble as well, may you post a working config file for it? Hope that is included in the next release of robocode.
I removed the caches from all installations (RoboRumble and GigaRumble are separate installs), and now XanderCat also works! I can't run battles till I get home though, as the firewall at work prevents download/upload of necessary files and results.
My config file is listed in the discussion below. The gigarumble.bat file contains:
java -Xmx512M -cp libs/robocode.jar;libs/roborumble.jar;libs/codesize-1.1.jar; roborumble.RoboRumbleAtHome ./roborumble/gigarumble.txt
It will be cool to see DrussGT #1 in GigaRumble be threatened as we kick BulletShielding-susceptible bots off from the top30 :P
There are 4 bots vulnerable to BulletShielding present: Phoenix (#16), Chalk (#21), CassiusClay (#22) and Seraphim (#27). They earn their presence as they kick a lot of butts. But you are right, DrussGT only clings on the first place due to its Bullet Shielding. Diamond on the other hand does not use it.
Yeah, sure they do, but these are bots we can probably assume that won't receive new updates, so they will be BulletShielding-susceptible forever and, let's be honest, they will eventually be kicked off from top30. Maybe it will take very long, but they will. And of course, they will be kicked from top30 in part because of the legacy they left behind :P
And just to clarify, I would love to be wrong about those bots not receiving updates :P
To effectively BS, you have to shoot the other bullet out of the air. So you have to exactly know where the other shoots. If the enemy moves, it will aim directly at you, but fires the next tick. Or maybe they compensate, but still they will not fire from the exact place they interpolate. Some bots just spin their radar and shoot whenever they see someone. Only good bots will fire really exact, and only they will be vulnerable to BS. When BulletCatcher came around in 2009, the actively developed bots took their countermeasures. Only a slight deviation from the 'ideal' angle is enough to let a BS miss. DrussGT tries to shield against a few different firingmethods, before it decides to fall back to its native gun and movement.
Thanks for the detailed explanations! I thought the firing control part should be mostly the same among a lot of bots (e.g. via copying the sample bots or OS bots), but it seems that this is also the part a lot of people made mistakes in, and there are also so many variants. Some bot aims precisely but fires whenever gun-turn-remaining is less than bot width, some forgot to call setAdjustXXX, some aims from the next tick but uses velocity from the last tick to predict... Such mistakes are so subtle that no one notices, until they incidentally found their bots not vulnerable to BSs. I know very few bots protecting themselves from BSs deliberately, most people get protected occasionally.
Are there bots like MoxieBot, which instead of intercepting HoT bullets, deliberately create a shadow to which they move on?
MoxieBot is quite cool, but unfortunately does not run reliably in the modern robocode engine. I think we still have a lot to learn from this bot.
I myself clearly have buggy bullet safety shadow code since I see enemy bullet in the calculated shadow zone quite often.
I think with a modern Passive Bullet Shadower, you can easily implement Active Bullet Shadower by brute-forcing firing angles — and actually with some math you should be able to calculate that directly.
I've been long thinking about some smooth transition between Bullet Shielder to Active Bullet Shadower and Passive Bullet Shadower. E.g., shield at first, then begin dodging. When there's a chance to shadow, do that deliberately if worth the miss.
What do you think about bringing the GigaRumble participants to the main roborumble page, and adding the gigarumble.txt and gigarumble.sh/bat files to the superpacks?
Totally cool with me, I do think it's a pretty interesting rumble. Worth noting that it requires a little maintenance to actually sort the main rumble by PL and update the participants list from time to time, but the top 30 doesn't change that often.
Also open to discussion about the number of bots/format/name and such, all of which I dictated when it was just me playing around.
Does this mean I should start keeping my own entry updated on the participants list? At the moment, XanderCat 11.4 is still in the GigaRumble, which is quite a bit out of date.
Once you are on this page, you have your ticket in. Feel free to edit.
Voidious, I'm sure you could automate the page generation?
Er, yeah, I guess I could. Sort main rankings by PWIN, take top 30, update GigaRumble participants list - right?
I really should get all this stuff running from the RoboWiki server. It's funny sometimes when I turn off my laptop for travel, then come back online and see a few @roborumble tweets come in. ;)
As a note, I just updated robocode-archive.strangeautomata.com to ensure that participants-latest.zip always includes versions in the GigaRumble participants list as well as the other versions. This is to cover the case of the main participants list getting a new version before GigaRumble.
(Also tomorrow makes for the 1-year anniversary of the cron job for robocode-archive.strangeautomata.com being started, and hasn't failed since :))
I nominate the following robots for admission into the GigaRumble: Nene, Neuromancer, CunobelinDC, RaikoMX, and SandboxDT.
If anyone else would like to nominate a robot for the GigaRumble, feel free to use this thread to do so.
Well, Nene has more than earned its place at #11 in PL. As for the others, my vote is definitely to stick to "top x bots" instead of hand picking bots we like.
Top 50 would include all those bots... I still like 30 and personally feel like even that's erring on the high side. (A bot at #30 has almost no chance vs DrussGT.) What do other people think?
For:
- Nene, top 10 bot. I have no problems with someone entering it, but its turned for simple bots.
Against:
- SandboxDT, impressive for its age. But not impressive enough.
- RaikoMX, see SandboxDT.
- Neuromancer, great at Melee, only decent in One on One.
- CunobelinDC, entering it in the gigarumble just isn't fair.
Nominating:
- Seraphim, decent post-debug, holds #10 Vote and #22 PL (PWIN). Tuned against surfers originally.
If you want Sandbox DT you should do a separate "Classics" league. ;)
That crossed my mind too. :-) Then I thought it's kind of silly to have an active rumble for bots that never get updated. And also, how to treat a bot like Shadow? Totally classic, but updated a lot more recently than others like DT.
It's not a sign of disrespect to leave those bots out of a "strongest bots rumble". Pretty sure most of those authors would want to earn their spot anyway, if they were active.
I should know better than to even look at the results after so few battles, but this already seems pretty interesting: [1].
First thing I notice is Diamond/DrussGT about even in APS, but DrussGT almost 5% ahead in survival. Of course, I start thinking, "maybe some better bullet power management would rocket me ahead of DrussGT!" But my experience is generally that more conservative bullet power selection will increase survival, but tend to lose the same in bullet damage, for the same overall score in the end. Perhaps DrussGT could even gain some score by sacrificing some survival...
Shadow is lower than I expected.
Pris does awesome, as I expected.
This LiteRumble thing is pretty cool! =)
Like I expected, GresSuffurd is dangling at last place. Btw, is there a way to check if a robot extends Robot, without looking in the source?
Given how strong GresSuffurd already is, I bet you could move up to the middle of the pack in a night's work, adding rolling average to surf stats and/or an Anti-Surfer gun.
About Robot, do you mean for accounting for wall collisions in energy tracking? I was just thinking about that the other day (very old "TODO" in Diamond's movement code)...
I'm pretty sure Robocode doesn't tell you, but you could detect it pretty quickly by noticing if a wall collision affects their energy. With gun heat tracking, since most collisions won't happen at gun heat = 0, I bet it would usually take only one wall collision to tell if they're Robot or AdvancedRobot. (A bullet being fired is the only other reason the AdvancedRobot might have lost energy, and you could rule that out.)
First page |
Previous page |
Next page |
Last page |