[Home]SilverFist/Chat20040923

Robo Home | SilverFist | Changes | Preferences | AllPages

Interesting! :-) -- Pulsar

This is rather a rather funny picture:

-- PEZ

I got 1000+ momentum when I changed Shadow's package to abc.tron3. :) -- ABC

After 20 battles SF too has a momentum of 1000+. I wonder why the rating system refuses to give it a higher rating. The latest Shadow got 2050 after 3 battles. Doesn't matter much, but it adds to the voodoo feel of the ELO ratings. -- PEZ

Weird, very weird... But SF seems very strong, iīm very curious about how high it will go. -- Axe

What's weird about a bot with SS' movement and P's gun being very strong? :) -- ABC

2039 points! This could maybe be tweaked to go grab that crown. Hmmm... -- PEZ

Amazing, congrats both. I must say I wasn't expecting such a high rating. It must be that a GF gun is even better than I thought against good random movers. -- ABC

I was expecting it actually. The theory was that SilverSurfer might have movement that almost matches RaikoMX's and Shadow's. Now I think that maybe both SilverSurfer and Shadow might have movements that is better than RaikoMX's. The Raiko gun is probably still the best RoboRumble gun out there, even if I have been closing the gap considerably lately. Anyway, you can easily test what the Pugilist gun might do to Shadow's ranking. If you unpack SilverFist you'll see that it uses an almost pluggable version of the gun. It probably takes you only one minute to plug it into Shadow. Just as a test of course. I do understand you don't want to have your bot equipped with someone elses gun permanently. =) But it would be very interesting to see what rating we would get that way. Regardless of the outcome of such an experiment you would get very good input on what to focus on next with your bot. Like if you don't outscore your current version you should probably focus on the movement. And vice versa. -- PEZ

I have been tempted many times to test Shadow with another gun, I could even use Jamougha's gun since it's opensource, but that goes against my "do it your own way or lose trying" principle... ;) Btw, why do you say RaikoMX's gun is the best gun out there when you out-score him in both TC's? I'm sure that a current top10 TC would show very different results though, there are now 2 higher ranked bots that Shadow crushes to a pulp... -- ABC

TC's selection of reference bots is probably not enough to tell the real RoboRumble story. And as for crushing high ranked bots. That's not what RoboRumble rating is about. =) Though I am pretty sure if you keep insisting that your bot should be super strong against the high ranked then eventually you will out-score those against the low ranked too and the glory will be yours. That's my philosphy with Pugilist at least. I hate seeing green dots in it's mini LRP graph! Still think you could try one release with P's gun. Recall it immediately when it has ran its 500 battles and you will almost not have noticed it. =) -- PEZ

Ok, ok, I've just plugged your gun into Shadow, I'll think about testing it in the rumble after I finish some key tests. So far it seems to perform worse against RaikoMX (55%) and loses badly against Shadow 3.04 (under 40%). I'll test it against DT 2.41 next... - ABC

Most impressive, 59% against DT 2.41 and an amazing 53% against DT 3.02! -- ABC

Here it goes, I won't substitute Shadow, just test it in the rumble, it also loses against Tron 3.x. I like the fact that Shadow is currently the strongest candidate to the PL throne, only losing against DT in my tests. ;) -- ABC

Cool. I think that against DT there seems to be a divide between yours and my gun. Pugilist slices down DT's wall smoothing to small pieces and seems to find a pattern there. -- PEZ

Woke up this morning (not really morning :), and u have been quite busy! The Shadow P version is a good test indeed, that might reach the throne (I think that Shadowīs moving is the strongest than SS, but now weīll see how much). -- Axe

Well, after 150 battles it doesn't look very good, 2015 points and falling... -- ABC

I'm not so surprised if Axe has the better movement at the moment. He got over 2020 with SS, and I know his gun is not as good as Shadow's against most of the field.

PEZ, if you want to put together a PugilistMX? as a test then I'd be very interested in the results; I'd do it now myself, but I have my final exam in 2 days... :) -- Jamougha

Yeah, let's do that. Even if I'm not sure I want to see the results. =) I'll see if I can do it quickly. -- PEZ

One thing i know for sure: my Gun surelly needs to be improved a lot (but that would have to wait untill i'm done with WaveSurfing)... Allways that i try to raise my firePwr, the rating degrades. -- Axe

Well, that happens to me too. If I use power 3 instead of 2 as default I get 50 or so less rating points. -- PEZ

RaikoMXP? is out now, named jam.RaikoMX 0.29.p like abc. It's not supposed to hang around I reasoned. It was just a quick and dirty jamming in of the gun while I was waiting for my home made pasta to dry. So I might have broken something. -- PEZ

Ok, so it would seem to me, that out of Pug, SS and Shadow, SS is the best mover (as when all have Pug's gun, then the SS variant scores highest). However Shadow kicks the ass of the other 2 with its own movement and gun. Therefore the best combination would be silversurfer movement with shadows gun?? While everyone's in the spirit of multi-author bots, why not combine SilverSurfer and Shadow to make some sort of portuguese speaking bot eating monster. Just a thought --Brainfade

Lets see how things turn out when all the current combinations get enough battles, I like the idea of a pt.SilverShadow? being all-powerfull, though. :) -- ABC

There's still the possibility that RaikoMX movement and Shadow gun would be the most scary combo. The two main conclusions I think can be drawn right now is that Pugilist movement isn't all that great and that SilverSurfer's gun has some catching up to do too. Let's not mention CassiusClay's movement =).-- PEZ

Hmm... maybe 300-odd battles is too early to tell, but it doesn't look like my movement is too scary at all. :) Unless RMX.p is buggy, which it doesn't seem to be in my tests, that's a 40 point gap in gunnery. -- Jamougha

Same gap as Raiko has to Tityus and as RaikoMicro has to Aristocles. I think your gun is 40 points better than mine. -- PEZ

I'd wait for around 600 battles to reach any conclusions, Shadow.p gained a lot of points between 300 and 500 battles. -- ABC

Of course, if these scores hold, it also means that Shadow's current movement is around 20 points better than RMX', and SS' movement around 30 points. That would mean the best combination would be SS+RMX. -- ABC

@PEZ: Are u sure that RMXp isnīt buggy? Canīt believe that RMX gun is that far from yours... Maybe a Pugilist experience + RMX gun can clarify that... -- Axe

What about a table to make these xperiments results more clear (pls help to fill it up / correct)? -- Axe
gun/movementSS 2.21 P 1.9.9.4.1RMX 0.29 Shadow 3.04
SS 2.21 2020N/A N/A N/A
P 1.9.9.4.1 2039 2000 2004 2028
RMX 0.29 N/A N/A 2046 N/A
Shadow 3.04 N/A 2028 N/A 2036

Hmmm, quick question - is SilverFist saving gun data between rounds?? Because it sure doesn't look like it is... --Brainfade

Axe, I'm pretty sure there's no new bug in RMX.p. The gun is so easily pluggable that even I have a hard time introducing bugs plugging it. =). I did think beforehand that RMX's gun would prove better than mine. I would have guessed 20 points or so maybe (thought I had improved Aristocles gun) but I'm not too surprised by 40 points either. I'll try to disentangle RMX's own gun and wed it to a Pugilist movement. Tonight that'll be. -- PEZ

Brainfade, you're eyes are not lying to you. I've assimilated ABC's philosophy about data saving. And, I tell you, robocoding is much more fun without it. -- PEZ

I must say I'm very much surprised by RMX getting 40 less points with P's gun. That means we don't have good way of measuring gun performance in the TCs. It's probably the time to make a new TC, maybe with a mix of top random movers and top wavesurfers. I'm also tempted to test my recently improved gun with RMX/SS movement. I won't do it, though, I'll now go back to improving my own code, beeing stubborn made me reach the top before. ;) -- ABC

Paul stressed the fact that the TC needed a better mix of low/mid/high performance movers when we created it. Thoug I think the TC is about as good as it gets. The environment is not exactly the same as in real battle. Both RMX and Pugilist fire firepower 2 in real battle, but are forced to fire power 3 in the TC. And the mix of movers is very hard to get right if you have only 10 reference bots and the number of rumble bots is approaching 300. More reference bots than that would make it painful to run the test... At least the TC can guide you some in tuning your gun.

Here's another tought. I can make a pluggable movement of Aristocles just like I did with P's gun. Then people can equip their guns with that movement and guns can be compared. The reason I'd choose Aristocles instead of P is that the top-10 would be more often intact then I think. What about it?

-- PEZ

Nice idea - I think that's probably the only way to rate guns effectively, 'in the field'. I always had a feeling that it was the targeting that was more special about the Raikos, but I didn't think the gap would be that big. :) Nice to know I've got plenty of room to improve my movement, anyhow! -- Jamougha

Nice for you maybe. =) -- PEZ

The more I think about it, the more I'm amazed by such a big difference... That means SilverFist would be near 2080 points with Raiko's gun! There must be something about the combination movement+gun that Jim mentions, maybe it "solves" some of Raiko-movement biggest problem bots better than other types of gun? -- ABC

We can try a version of SilverFist with the Raiko gun. It wouldn't surprise me too much if it approaches 2080 points. First I have to disentangle it though. This part of RMX is a bit of a mess. Quite ironic. =) -- PEZ

I say this mostly because my tests indicate a very small difference between my gun's performance against SS and RMX movement. I know my gun is not "standard", but i believe in it ;), and it contradicts this results by saying SS movement is slightly more predictable than RMX. Then again, I still can't win by a decisive margin against my best random movement (Shadow 2.15), and with P's gun I can... -- ABC

P's gun is mainly developed to gun down Raiko and DT so its performance against random movement isn't a coincidence. The only thing I can see where RMX's and P's movements differs in a way that it could influence the gun is that they control their distance differently. P backs away all the time and RMX has a preferred distance of some 600 or something. I doubt it is important, but it could maybe be that the Raiko gun likes that fighting distance more than the Pugilist gun does. You should believe in your gun. But it can't alone give general answers. -- PEZ

Last time I looked, RMX.p had a specialisation index of around 58, vs 20-something with my own gun. That tends to indicate that the Pugilist gun works extremely well against some bots, but that RaikoMX's manages to catch everyone more-or-less evenly. Which makes sense given that P is so focussed on wall movement, while RMX just has oodles of segments for every occasion.

ABC, your gun is so far from the mean that it's difficult to rate a bot's movement by it. It definitely picks up patterns in wave surfers that virtually all other guns miss, excepting SS/Aleph? and perhaps Cigaret. That's why I never test against Shadow. :) -- Jamougha

I came to the same conclusions, Shadow & DT guns are probably the bests against WaveSurfers (but i don't think that SS or Aleph guns are that good). They both are my nemesis, Cigaret is also a big problem. ABC, that came naturally, or do you have a special treatment for WaveSurfers? -- Axe

I like the idea of a reference movement for gun comparison purposes. Actually I have used Raiko's movement to test my new gun in Locke. --Vic

@Axe: It came naturally for me, my gun is much older than anything related to WaveSurfing. I believe Paul made some specific changes to his gun to improve it's performance against wavesurfers, but I think his gun was allready harder to fool than the "typical" GF gun. I remember him talking about "timer based movements", that must have been when he dealt with it. -- ABC

Must have been post-1.91 then. Since Aleph could complete the CurveFlatteningChallenge. -- PEZ

Maybe it's my inability to target him that makes me overrate his gun. The version I first tried to dodge using wavesurfing was 2.41. -- ABC

Has anyone besides me come to the conclusion that the reason PM guns do so well vs WaveSurfers is that the surfers tend to only slow when the intend to reverse? This seems like a very easy thing to trigger on. If, as a mover, you are only making decisions at shot time, your escape envelope has been cut by 1/3 or more. Just thinking out loud. -- jim

Can you develop that? I think no wave surfer is making descisions at move time. Maybe I misunderstand you. -- PEZ

Don't think that jim's assumption is necessarilly truth (at least not for SS), it depends on how you surf. The biggest problem against PMs, i think, is the fact that it's intention is to fool GF guns, more specifically to avoid GFs areas, and by doing this you can be caught in some pattern. However, the how and when u move to these GFs can be the answer to avoid these patterns (is in that that i'm focusing now). -- Axe

SS wavesurfes in a very different way than Shadow, but I don't think the "problem" Jim mentions aplies to any kind of wavesurfing. You'll only stop/invert a lot when the GF's close to 0 are safer than the ones near GF -1/1. I have, nevertheless, already tried what you mentioned elsewhere, testing +8, -8, +4 and -5 movement, and other variants. I didn't notice any improvement against PM guns, though. The problem is probably that, like Axe said, certain movement profiles will repeat and your reaction will be repeating the same patterns. -- ABC

@PEZ I thought that direction changes were only considered at shot time, even if you considered all valid not reached me yet Waves. I clearly do not have a surfing movement so it appears I am showing my ignorance of the subject.

@ABC && Axe Maybe patterns do develop in the movement. I had not considered that. My thinking was that if you only ever decelerate because you intend to invert your direction, then that is a huge signal of what you are going to do over the next 5 - 13 ticks (5 from 8 to 0 and 8 from 0 to 8) that a pattern matcher could key on. In that 13 tick window a 3 power bullet is 143 pixels closer to you. At range 400 your escape envelope is approximately 23 bot widths wide. At range 257 (400 - 143) your escape envelope is 15 bot widths wide, a reduction of 8 bot widths. If you remove the ones you can not realistically get to you are really only left with 12 - 13 bot widths to fit into. Purely random shooting at that point should theoretically get an approximately 8% hit rate (better than I can get with BP or Jekyl). If I shoot a 2.0 power bullet the available escape area becomes even smaller.

Now one last question: if you think that a pattern emerges in the movement that a PM can pick up on, could a bot that shoots Waves every tick not look for a patteren in the GF's that were fired over the last N ticks to see if a pattern has developed and from that perhaps deduce which GF to fire at for the next shot? Maybe by iterating forward through all the waves until a projected bullet would hit, and pulling that GF out of the Waves buffer and shooting away. It's an idea that has been floating in my head for quite a while. Maybe someday I will get around to implementing it. Maybe it would be that long sought marriage between a stat gun and a pattern matcher. -- jim

jim, your concerns about escape envelope are more than justificated. I tried to maximize that escape envelope during my development of SSīs WaveSurfing (Iīm trying to find some time to finish my SilverSurfer/WaveSuffering page, there iīll explain the peculiarities of my movement, perhaps u will find it an interesting reading).
About that gun... Do you read minds, man :)? That idea is the basis of my future (not any more secret) GF-PM gun (recent talks here in wiki, makes me think that Iiley had already done something in that way - witch can explain a lot the performance of his bots against the WaveSurfers). I have a lot of experience in PM (my gun), and WaveSurfing gave me the GF & Waves knowledge, so...
The beauty of this kind of gun is that it can reduce the 2D problem into 1D, and also the wave-phasing can help a lot. I have a feeling that it would get together the best of two worlds, and bring some storm to WaveSurfers (here i am, again shooting my own feet, oops... :). -- Axe

Well, in Shadow I evaluate direction changes (and currently also stopping) every tick, it's very much like a minimum risk movement. That results in a movement where deceleration doesn't necessarily mean inversion. It is possible that it decides to go back to full speed after it stops or after it decelerates enough so that the GF zone it will end up in if it resumes full speed movement is safer than the inversion. Another factor that you are not considering is that the shots that are fired precisely when I'm deccelerating/stopping are a small fraction of the total shots, most of them will be fired while Shadow is traveling at full speed. Even pattern matchers will match the pattern from the current tick, not from the last inversion.

I believe Iiley's old PM gun is very similar to that GF-PM gun you guys are talking about. ;)

-- ABC

I believe iiley used waves as the basis for his PM gun. I am not so sure that he used the GF result from the Wave collision as the basis for matching on. As Axe alluded, it reduces a potentially multi-dimensional problem to 1D space. You would of course need to test this vs. something that moves in a consistent pattern. I would also suspect that it would be so heavily influenced by your own position that it might prove useless.

@Axe: It is only a matter of time until someone developes a FlashLight? gun. I call it that because a FlashLight? will remove the Shadows from the world. At the end of the day all WaveSurfing is deterministic movement. Someone will figure it out. They did with DT's stat guns already. -- jim

Ready for some PEZ style "hunch"-theories again? I've been thinking a lot about the different results when equipping some of those top bots with the same gun. I have started to think now that the answer isn't as simple as RMX's gun is 40 points better than Pugilist's. (I know I said I thought it was that simple before. But I think I have proven myself wrong there with my latest CassiusClay experiments with combining different versions of my movements and guns.) But I don't think it's as complicated as some elusive influence on the gun from the movement or vice versa. This is what I think. Imagine a table with three columns. One column of PBIs for a movement left to its own. Second column is the PBIs for a gun left to its own. Third column is the sum of the first two. Now for a gun to compensate the movement against the movement's problem bots it needs to be particularly strong against those. And the gun needs the same help from the movement against the gun's problem bots. Now what if RMX's movement and P's gun share more problem bots than the combinations RMX-move and RMX-gun? And what if many of SilverSurfer's movement problem bots are the same bots where P's gun is particularly strong? We would get the result matrix of these combinations as we did in the real test, wouldn't we?

Well, even if the theory holds, I doubt it will get us anywhere. What's needed then, but maybe hard or impossible to achieve is a way to create those two first columns of the table for a representative amount and selection of rumble bots...

Yeah, just a thought.

-- PEZ

Another point: You cant compare guns only by their points in the TargetingChallenge. Remember that there all guns shoot with bullet power 3.0 . But energy management of these guns is not the same !!! --deathcon

BTW, which parameters do use to calculate your firepower? --deathcon

I fire power 1.9 always. Except at really close ranges (sub 180 or something) where I fire power 3. -- PEZ

Doesnt seem to me very smart . --deathcon

But I thought a lot before arriving at that strategy. And after that I have tested it a lot and it seems to work very well. Though when I made the gun I was working with a mini. There is probably room for some more fancy strategies in a mega. Power 1.9 is not arbritrary. I posted a graph somewhere on the wiki where I plotted the relationship between score and max enemy escape angle. It indicated that a fire power of 1.8 something maximized score/escape_angle. Of course noone but me seemed to care about that theory. But I'm used to have my theories laughed at by others. =) -- PEZ

Poor PEZ ;) --deathcon

It was on the SelectingFirePower page by the way. -- PEZ

You made me laugh just now :-) Be assured though, not because of your firepower theory ;-) Poor PEZ indeed :-) --Vic


Robo Home | SilverFist | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited September 23, 2004 6:15 EST by PEZ (diff)
Search: