Archived talk:RoboRumble Gun Challenge 20090505
- RoboRumble Gun Challenge Sub-pages:
- RoboRumble Gun Challenge - Results - RaikoNMT - Archived Talk 20090505
|
I made the /RaikoNMT movement pluggable (PEZ style). Download BeeRRGC for the source and usage guide. -- PEZ
- thanks, Pez, i used your version to quickly plugin mini.Wodan's gun. Question: i noticed the results of RaikoRRGC in the ratings, but could not find its entry in the participants list. Can you use RoboRumble@home to focus on one bot that resides on the local system where RoboRumble@home is running? --Loki
- RaikoRRGC is in the wiki.mini package, not the wiki.rrgc package. --Vic
- thanks, found the package! But is it possible to focus RoboRumble@home on a single bot? I believe you had different versions of Locke in the rumble results, but one entry in the participants list. --Loki
- PEZ and Albert are working (any progress?) on a new RR client that can give priority to a single package. The effect you saw was probably a known peculiarity in the RR system that causes old versions to reappear in the rankings. Yesterday I witnessed three CassiusClays, two PulsarMaxs and two Lockes in the list at the same time :-/ --Vic
- No progress yet. I haven't put any time on it actually. It shouldn't be very hard to do it the way Albert described. But to comment on Loki's question. We really don't want several versions of a bot in the participants list. As Vic says it's just a glitch in the matrix that several versions of a bot sometimes appear. It fixes itself once all clients "forget" about all old versions. -- PEZ
Vic can you re-release RaikoRRGC and somehow make it a MegaBot (mayve just using the pluggable version of the movement might push it beyond the CodeSize limit.) It's ranking might get a bit contaminated by the fact that it has an unproportional amount of pairings against minis. I know this shouldn't happen, but I have seen Pugilist general rating fluctuate often enough while its mini battles is filling up. I think we should add it to the rules of the competition that the entrant must be a MegaBot. -- PEZ
No problem. --Vic
This (the MegaBot rule) then applies to WodanRRGC too Loki. -- PEZ
- should be no problem: i will add some junk-code (same as all my other code...) --Loki
Well PEZ, only two points behind RaikoRRGC! What do you have to say for yourself? Congrats with improving your gun more than you expected yourself. Now that I think about it, Jamougha may have improved his gun since Raiko as well.... It is strange that Locke lost a lot more rating points than Raiko did after removing the MusashiTrick. But now I know where I stand compared to you and Jamougha (I would be very interested in how other people's guns perform, we need more RRGC contestants :-). A lot of work still to do.... --Vic
- What do I have to say... Well, I'm still not sure I have improved the gun. As I thought out loud on the SilverFist page it might be a matter of your gun compensating your movement's ProblemBots. Maybe Bee does that for Raiko's movement as good as Raiko's own gun does. No, I don't think RMX contains any changes to the guns behaviour. At least I would hesitate greatly before changing anything in a gun performing like that. =) I don't think it's too strange Locke lost more by removing the MusashiTrick than Raiko did. Given that Locke's gun still has some catching up to do on Raiko's the MusashiTrick hid this since it is all about feeding on a flaw in the rating system with your movement. -- PEZ
Loki, it seems you have more Rammer problem bots than you should. I had that too some version ago. Could it be that before firing you wait for the gun to stop turning even when it is cool? Removing that 'accuracy check' gained me almost 50 points. But I must say my gun's waiting back then was extreme. Sometimes it would not fire for a long time when Locke was taking a long sweeping turn. --Vic
- no, i even don't wait for the gun to cool: i use the simplest type of gun-control and 'fire' every scan, which will fire a real bullet when the gun has cooled down. Maybe the problem is that the gun has not yet been alligned with the rammer when i fire. Although, when i think of it, rammers must have a large GF-spike around zero GF, so the gun will always more or less stay pointed directly at the other bot. So i just don't know. --Loki
- a problem i still have is to handle stop&go bots. Pez pointed me at a solution, but i have still have not implemented this. --Loki
Great work, deathcon. It seems like you have a pretty mean gun as well. Not that I'm really surprised, since Shiva does really well... --Vic
Yeah, thanks. And it shows that there are still 10 point to gain. --deathcon
Well it is official: my gun sucks. I have known for a while that it was a limiting factor but it appears to be a real limiting factor. This is not good. Well now that the gun is jammed into it's own class I can do the same with movement and then I will be able to start plugging in different guns to make the next leap. I am not looking forward to this part. I have spent way too much time in the past working on gunnery and still have apparently not got it right. Ah well, a robocoder's work is never done. -- jim
No 1950+ bot sucks in any way in my opinion. I can still only dream of such scores. But I do know how it feels when you find out you have to catch up on 60 points with only gun tweaks. But PEZ assured me that this can be done, and in just a month or so I already gained half of those points. So go for it jim! You know now that gun tweaks alone can get you into the 2k club. --Vic
- Thanks Vic. I am refactoring everything into their own classes (Encapsulation anyone?) so I can be sure that any gun tweaks do not break anything else first. I should be done with this within an hour or so. I will then release this bot as an interim to make sure that everything is still working and then the testing starts in earnest. One thing that I know for sure is that if I can shoot better I will vault quite a bit. -- jim
- About earning 60 points with the gun alone; 1878 - 1819 equals ... ? =) -- PEZ
- Well I just finished refactoring DarkHallow and I am reluctant to replace the gun. I tested 500 vs. RaikoMX and beat him 55% to 45%. I tested 500 vs. CassiusClay and beat him 53% to 47% (I only ran these tests once so take the results with a grain of salt). The question is: should I keep a gun that is clearly not the best against bottom rung bots but is against top shelf bots? I know I should as there are more points to be made more easily vs. the bottom of the ladder than the top. I am going to release a refatctored DH and see where it is. It should prove to be interesting. -- jim
- I can pretty easily make my gun effective against all surfers. Jam knows how to do it too. But it makes the gun work less good against the majority of bots. Frankly I think surfers assume a learning gun is a good gun. And when it isn't, well then all bets are off for the surfer. -- PEZ
Way to go with the Shadow gun! Didn't expect it to be quite that good actually. -- PEZ
Thanks, neither did I... -- ABC
Great result, ABC! Just wondering, did you change much to the gun since Tron 2? --Vic
Not much, I changed some weighting parameters recently, but it's still the same algorithm/code. -- ABC
I've been playing around with weighting also, and found it can make a real difference. One recent tweak gained me 6 points, and another 7 points. Locke is trailing ShadowRRGC by just 12 points now. I think I will tweak some more and go for that crown ;-) --Vic
YAY! eeehmmm....CC WT = 2055 + 27 = 2082 rating points? --Vic
=) I'm hoping for 8 points or so. I think CC's surfing takes care of many of the bots in a way that it doesn't matter what gun it uses. -- PEZ
Added GrubbmOneRRGC to the participants list. Lets see if it is my movement or my gun that is crap.
Probably both . . . --GrubbmGait
Well, it's rather clear. GrubbmOneRRGC gets a rating of 1695 (not that bad for a CT-gun) against 1525 for my original bot.
This means my movement should improve at least 150 points.
Noting the difference between the 'big' guns and my own, my gun should improve 150 points also.
First goal: rating 1600 with better guns.
GrubbmOneRRGC will now be removed from the participants-list. --GrubbmGait
In fact your movement can improve 180 points I think, compared to Raiko. 30 points are denied RaikoNMT since it doesn't have the MusashiTrick left. -- PEZ
Ultimately I have to improve approx 300 points in movement and approx 200 in my gun, otherwise I never join the 2K-club. For now (this year) I will settle for 150 points each =) -- GrubbmGait
Yeah, but the 30 points from the MusashiTrick are easily collected. Might I suggest you make a version with that trick implemented and see where it takes you? Then you can remove the trick and have it in store for when you are starting to see improvements in your general performance. (The MusashiTrick can hide some aspects of your bots performance if left in.) -- PEZ
Does this meaning Pear has a not good Gun, I am wondering .... Pear 0.58 can beat many top bots. It's strange. -- iiley
Pear doesn't have a bad gun. But if you think about what kind of bots are "top bots" you can probably see how a bad gun would be quite effective against them. I know one or two bugs I can introduce in Bee to make it eat surfers for lunch. Anyway, you have 20+ points to collect by raising your gun's standard to the level of the top guns. -- PEZ
A VirtualGuns array of only simple targetters is just 120 points behind the top-GF guns. -- GrubbmGait
Well it seems I still have some catching up to do with my gun. -- Florent
Note that almost all results date from a year ago, I think you can subtract approx. 5 points due to drift and 'aging'. --GrubbmGait
Yes indeed maybe even more. Somone should run some of the other robots in that list again. -- Pulsar
Ok, that's a bit conforting. But it is still not as good as Bee, Acsendant or PulsarMax's gun. -- Florent
Yeah, I wouldn't mind seeing updated results for some of those other bots... Dookious RRGC isn't stable yet, but it's looking about the same as Toad. While I'd love to find that there are still so many points in my gun, I am a bit skeptical... (Though I'd also guess that the point differences are bigger in the RRGC than they are with an expert WaveSurfing movement.) -- Voidious
With BeeRRGC dropping almost 20 points, I'd say all of those ancient results are pretty far from accurate =) I guess our guns aren't as bad as we might've thought at first, Florent. -- Voidious
As far as I know Locke is a RRGC-bot and it has dropped 26 points, although I don't know it is the exact same version. If you extrapolate this, Ascendant and Bee have the only guns that scores better than Dookious and Toad. -- GrubbmGait
I figured there should be some note about the old scores, but I wasn't completely sure how many were out of date. Maybe we should just make a new table, and pull an old table from an older Revision and just list them separately? ...or is it fine like this? -- Voidious
This challenge is surely the most accurate way of judging overall gun performance. However, I've had this idea in my head for a while... What would the rest of you think of something like a "WaveSurfer's Gun Challenge", using the movement from, say, Komarious? It's small, fast, and is a fairly decent WaveSurfing movement. The idea came to me after I used the RRGC exclusively to develop Dooki's gun for a little while, only to find out that gaining 5-6 points in RRGC lost points in the actual tank. We've updated the TargetingChallenge and the MovementChallenge, so why not this, too? Thoughts? -- Voidious
In theory, and probably practise also, this would indeed be a more realistic measurement for the top-guns, as they are mostly coupled with a top-movement (read WaveSurfing). Would the BasicSurfer provide a good enough movement, or is it just to basic to be representative as a surfer. -- GrubbmGait
I think it's a little too basic - no segmentation and no attempt at distance control. Komarious uses a lot of similar code - I referenced it while writing that tutorial - but also features built-in HOT avoidance and a lot of other nice tweaks. If I don't hear any good reasons to not do so, or a suggestion of a better movement to use, I will make the movement pluggable and post the page in the next couple of days. -- Voidious
I made the pluggable version of the movement: KomariMover.java (package wiki.kmove). It's pluggable just like the RaikoNMT, except that it also needs the onHitByBullet method. Here's my first implementation of it: DookiSaberWSGC.java. If nobody objects, I'll make proper pages tomorrow and post DookiSaberWSGC 1.10 tomorrow night. -- Voidious
This seems like a cool idea, It'll be interesting to see how the order of the rankings compares to the RRGC. --wcsv
Yeah, I think there will definitely be differences. There's still something to be said for how well a gun does without relying on WaveSurfing movement, but that's not really how everyone needs to design a gun these days. I noticed that 1.10 dropped 3-4 points in the RRGC compared to 1.06, while it helped me squeeze another 3-4 points out of Dookious and brought him to the top of the PL. -- Voidious
I'm very interested in this, since I'm pretty good at hitting surfers... =) --David Alves
