Difference between revisions of "Talk:Ocnirp"

From Robowiki
Jump to navigation Jump to search
(lot of text)
 
(16 intermediate revisions by 4 users not shown)
Line 6: Line 6:
 
Other topic: You'd laugh but I don't know about jikes. I'm not very into the low levels of programming, I'm rather interested in game programming and robocode. I just start NetBeans and have fun. But I'd be glad if you tell me more about jikes.. <br />
 
Other topic: You'd laugh but I don't know about jikes. I'm not very into the low levels of programming, I'm rather interested in game programming and robocode. I just start NetBeans and have fun. But I'd be glad if you tell me more about jikes.. <br />
 
Another topic: Yes, I know about GF nanos, simonton and kawigi has one. GFnano D and Thnikkabot I reckon so. Your movement is strongly vulnerable to them, because it's very flat. --[[User:Robar|HUNRobar]] 15:09, 15 February 2009 (UTC)
 
Another topic: Yes, I know about GF nanos, simonton and kawigi has one. GFnano D and Thnikkabot I reckon so. Your movement is strongly vulnerable to them, because it's very flat. --[[User:Robar|HUNRobar]] 15:09, 15 February 2009 (UTC)
 +
 +
I've not look at [[FunkyChicken]] or [[Kakuru]] code yet. But I know that NanoLauLectrik and WeekendObsession has '''difference''' type of gun. I know that it's symbolic pattern matching, but WeekendObsession do SingleTick matching while NanoLauLectrik does ordinary pattern matcher. For jikes, I not sure for nano bot, but for minibot, it is. On [[Komarious]] page on the old wiki, he said that jikes would save him 18 bytes. Also, when I try to compile new Conriantumr om my machine using javac, it's codesize is around 1503 which should not be occur for mini bots. To use jikes, go to robocode editor! Are you using Netbeans? Well, I've both Netbeans and Eclipse on my computer and I saw that Eclipse is better for robocode than Netbeans!
 +
 +
I think you should misunderstand something. My movement is '''not''' flat so the GF will hit me well! But it is flat enough to dodge simple GF. Note that flat movement is effective against GF, but truely flat movement will vulnerable to PM. I saw that current version of Ocnirp can dodge StatistBot GF better than my flat movement with [[Waylander]] flattener!
 +
 +
One more thing, I should deny that Ocnirp is mine. My PM gun is lift directly (the word 'modified' on its page mean only increase match length) from [[WeekendObsession]]. And I took only 10 minutes create random orbiting movement! Although I take 2 more hours to find the factor suitable for every bot.
 +
 +
Last thing: you can split your comment into many paragraphs as you like. Not just mix everything into one paragraph!
 +
 +
&raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 17:37, 15 February 2009 (UTC)
 +
 +
----
 +
 +
Depending on your distance, BFT can be anywhere from +1 to +3.  Give some different numbers a shot when you get a chance.  I find it very funny that all my tweaks thus far add up to a +2 to a timer :) --[[User:Miked0801|Miked0801]] 08:51, 24 May 2009 (UTC)
 +
: also, I found that having setAdjustGunForRobotTurn(true) disabled cost me 2-4% pts overall.  --[[User:Miked0801|Miked0801]] 08:55, 24 May 2009 (UTC)
 +
 +
My distance is 160 (!!) so almost every robot will move aways from me (except the one without distance controller). This is to trash YOU, Miked0801 ;) About setAdjustGunForRobotTurn(true), I figure this out when I'm creating NanoKitty with HOT, which can beat some stop-and-go robot for its gun slippage. But it was Robar who conclude that having setAdjustGunForRobotTurn(true) for PM gun make you lose points. You will not believe me that Ocnirp 1.0 has setAdjustRadarForGunTurn(true) too! &raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 09:46, 24 May 2009 (UTC)
 +
 +
== Distance controller ==
 +
 +
Just wondering. Actually my one now is more Pugio's than Neophyte's/Moebuis' but am I think correct? Because I hard-coded BFT, I need to control get to the controlled distance quickly. My version how have huge offset, but that will make me do a zig-zag at the desired distance. But there may have performance enhancing bug again like setAdjustGunForRobotTurn(true). &raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 10:26, 24 May 2009 (UTC)
 +
 +
:As I read the source, it's the same as my version in Pugio. :) The only difference from my other bots could be the calculation of offset angle: somewhere I divided Math.signum(xyz) in spite of multiplying by the radian value. There's no reason, maybe it was easier to type. :D --[[User:Robar|HUNRobar]] 15:55, 25 May 2009 (UTC)
 +
 +
== Worse, worse and worse! ==
 +
 +
Start from 0.2 to 1.0, a big JUMP! Now from 1.0 to 1.4, all the worse and worse =( If 1.5 don't get better, I start to have no ideas what's going on as I can't figure out what distance is best for 0.92 yet. &raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 13:51, 24 May 2009 (UTC)
 +
 +
:Hi, I can't test your robot now, but I took a look at the source.
 +
:Try to make your movement react to enemy fire. I have done exactly the same with Pugio and it gained 3% APS and 5 places. :) After every tweak I test my bots with NanoAndrew, NanoLauLectrik and Acero for 2-3 battles. For example, if a tweaked version of Pugio can't reach 89-90 % against NanoAndrew, 59-60% against Acero and 40-45% against NanoLauLectrik, than I trash or improve the tweaks.
 +
:Considering Ocnirp 1.4 at 90-95th place, there is some kind of bug or big problem with it. With a PM gun it shouldn't score lower than 50th place. --[[User:Robar|HUNRobar]] 06:57, 25 May 2009 (UTC)
 +
 +
: Thanks for the advices. 1.4 is at #59 actually, but 1.5 is at #91 =( I now think the close problem is not good for Ocnirp, I'll try the longer distance (250 or 300 again). Only bot I tested with is [[Pugio]] 1.4 actually... (first version is tested with WeekendObsession and all Neophyte) And PM gun now are hard-coded BFT so that may be a problem (46% against [[N]] is quite unacceptable for once top-20 nanobots)
 +
 +
: I may work around with this robot a bit, if it doesn't work and it even reach #150, expected 2.0 to be 1.0 again. &raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 12:01, 25 May 2009
 +
 +
:: You really need to test against [[Acero]] and one other simpler, HOT bot.  Acero's a great bell weather bot.  He's anti-pattern matching on movement and uses head on targeting.  If you are getting destroye by him, look at your movement to make sure it is protecting against HOT. --[[User:Miked0801|Miked0801]] 13:23, 25 May 2009 (UTC)
 +
(UTC)
 +
 +
::: Really? Surely don't! I have a lot of problem with this Acero, even the OcnirpSNG get 37% it! (and this 1.5 get 27% by the way) I actually didn't think there will be any HOT bot out there so I didn't think about it. (I always think my movement is hard to hit with PM/LT/CT, which use more common in NanoBot, but the 1.1-1.5 have prove me wrong) I've may fully retire this and make something better! (I'm creating statistical bullet dodging, but it didn't fit with PM so I need to squeeze it.)
 +
 +
:: You're wrong! There are many HOT-bots out there. Just write some by heart: Acero, Bluebot, Bichephal, Hawking, Hotspur. And also watch out for random-targeting bots, such as DuelistNano, Smog (maybe, I'm not sure), Squirrel (actually LT/Random), Splinter (same as Squirrel), or my first bot, Breeze (LT/Random again) and many others. I always test my bots carefully before releasing them. It's an old mistake to submit bots to rumble untested. So I recommend you:
 +
:::*[[Acero]] as very strong HOT,
 +
:::*[[NanoAndrew]] as LT, because it shoots very small bullets, and [[Avesnar]] a very good anti pm bot with LT
 +
:::*[[NanoLauLektric]] plus your WhiteWhales as top-bots
 +
:::*[[Splinter]] as LT/Random Targeting bot
 +
:::*[[NanoDeath]], [[SledgeHammer]] and [[Tide]] are pure rambots, but also don't forget [[Bichephal]], which is a StopNGo/Rambot with HOT!
 +
::Anyway, I've made a nano Anti Pattern Matching movement, which pattern matches itself and tries to move differently than the estimated movement. It's only in an experimental state but if you want I can submit it to the repository. ;)
 +
--[[User:Robar|HUNRobar]] 15:55, 25 May 2009 (UTC)
 +
 +
Hi. I have to ask what happened to Ocnirp, as 1.5 is just the 93th in the nanorumble? It's still a very strong enemy against PM, but it's almost 80 places lower in rank than the 1.0. --[[User:Robar|HUNRobar]] 16:44, 19 June 2009 (UTC)
 +
 +
Yes, it is really 77 places lower than 1.0. Few I ideas I have is too close distance with too random gun. The 1.0 is much depend on the starting environment since it never control its distance and use the same 0.92 constant at every distance. I  need to run a ton of battles to figure out which value is the best. The 'too random gun' mean that I tuned my gun to be accurate but with no setAdjust...(true) =) &raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 17:09, 19 June 2009 (UTC)
 +
 +
== The Difference of Pattern Depth ==
 +
 +
Hmm... Does anyone know that how much the pattern depth can effect on the pattern matching gun?
 +
 +
I must say that [[Ocnirp]]'s movement is ''very'' weak against simple targeting , so the main point the give Ocnirp current position in the rumble is gun, which is based on Simonton's pattern matching gun.
 +
 +
Now, 1.61 use 30 as pattern length while 1.61b use 45 as pattern length. There is a quite noticeable difference between them. (see [http://darkcanuck.net/rumble/RatingsCompare?game=nanorumble&name=nat.nano.Ocnirp%201.61&vs=nat.nano.Ocnirp%201.61b]) Fine-tuning this constant should probably give me points, especially when Miked08011 isn't here to update LittleBlackBook =) &raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 13:12, 13 September 2009 (UTC)
 +
 +
In [[Toorkild]] I have a pattern depth of 66, but I multiply my depth by 2/3 every time I look for a new pattern, not just subtract 1 like you guys in nano do =) Basically, the longer the pattern the better, until you get to the point where it starts skipping turns. --[[User:Skilgannon|Skilgannon]] 13:30, 13 September 2009 (UTC)
 +
 +
Nah, it depends on enemy. See [http://darkcanuck.net/rumble/RatingsCompare?game=nanorumble&name=nat.nano.Ocnirp%201.61&vs=nat.nano.Ocnirp%201.61b this] (the same link as above), 1.61b use longer pattern but it is worse again some enemies (in my test: Weekendobsession, Avesna etc.). In my test both robot's skipped turns rates are about the same. I wonder what pattern depth like Waylander will do (512 /= 2) (in 1.72, tomorrow release) &raquo; <span style="font-size:0.9em;color:darkgreen;">[[User:Nat|Nat]] | [[User_talk:Nat|Talk]]</span> &raquo; 13:57, 13 September 2009 (UTC)

Latest revision as of 14:57, 13 September 2009

Congratulations for your nano! It's very good for a first pattern matcher, but you must prepare, because I'm developing a super-duper nano which is not a Pugio-like beat-the-weak-hard-and-forget-about-other-pm-bots robot but a real pattern-matcher fighter! :) --HUNRobar 21:31, 14 February 2009 (UTC)

Ha ha... No, it isn't my pattern matcher, it's Simonton's! I just try my movement. I didn't expect any more ranking on this bot. My interest is in mega bot now! I've some problem on codesize-limited bot because I like javac than jikes! I think you already knew that jikes generated less codesize than javac! One more thing you need to know, there are some GF around nanoboy, too. My Ocnirp movement can beat top 4 bot, but not these GF bot! » Nat | Talk » 01:44, 15 February 2009 (UTC)

Ok, I thought that the pm code is not yours, but the robot is. My robots have Funkychicken's code, I think it's easier to understand and tweak. Anyway, there are 3 types of nano pm codes. One is the NanoLauLectrik, WeekendObsession etc. type of gun, second is Funkychicken's approachment and Kakuru has an own gun which works with doubles instead of a string.
Other topic: You'd laugh but I don't know about jikes. I'm not very into the low levels of programming, I'm rather interested in game programming and robocode. I just start NetBeans and have fun. But I'd be glad if you tell me more about jikes..
Another topic: Yes, I know about GF nanos, simonton and kawigi has one. GFnano D and Thnikkabot I reckon so. Your movement is strongly vulnerable to them, because it's very flat. --HUNRobar 15:09, 15 February 2009 (UTC)

I've not look at FunkyChicken or Kakuru code yet. But I know that NanoLauLectrik and WeekendObsession has difference type of gun. I know that it's symbolic pattern matching, but WeekendObsession do SingleTick matching while NanoLauLectrik does ordinary pattern matcher. For jikes, I not sure for nano bot, but for minibot, it is. On Komarious page on the old wiki, he said that jikes would save him 18 bytes. Also, when I try to compile new Conriantumr om my machine using javac, it's codesize is around 1503 which should not be occur for mini bots. To use jikes, go to robocode editor! Are you using Netbeans? Well, I've both Netbeans and Eclipse on my computer and I saw that Eclipse is better for robocode than Netbeans!

I think you should misunderstand something. My movement is not flat so the GF will hit me well! But it is flat enough to dodge simple GF. Note that flat movement is effective against GF, but truely flat movement will vulnerable to PM. I saw that current version of Ocnirp can dodge StatistBot GF better than my flat movement with Waylander flattener!

One more thing, I should deny that Ocnirp is mine. My PM gun is lift directly (the word 'modified' on its page mean only increase match length) from WeekendObsession. And I took only 10 minutes create random orbiting movement! Although I take 2 more hours to find the factor suitable for every bot.

Last thing: you can split your comment into many paragraphs as you like. Not just mix everything into one paragraph!

» Nat | Talk » 17:37, 15 February 2009 (UTC)


Depending on your distance, BFT can be anywhere from +1 to +3. Give some different numbers a shot when you get a chance. I find it very funny that all my tweaks thus far add up to a +2 to a timer :) --Miked0801 08:51, 24 May 2009 (UTC)

also, I found that having setAdjustGunForRobotTurn(true) disabled cost me 2-4% pts overall. --Miked0801 08:55, 24 May 2009 (UTC)

My distance is 160 (!!) so almost every robot will move aways from me (except the one without distance controller). This is to trash YOU, Miked0801 ;) About setAdjustGunForRobotTurn(true), I figure this out when I'm creating NanoKitty with HOT, which can beat some stop-and-go robot for its gun slippage. But it was Robar who conclude that having setAdjustGunForRobotTurn(true) for PM gun make you lose points. You will not believe me that Ocnirp 1.0 has setAdjustRadarForGunTurn(true) too! » Nat | Talk » 09:46, 24 May 2009 (UTC)

Distance controller

Just wondering. Actually my one now is more Pugio's than Neophyte's/Moebuis' but am I think correct? Because I hard-coded BFT, I need to control get to the controlled distance quickly. My version how have huge offset, but that will make me do a zig-zag at the desired distance. But there may have performance enhancing bug again like setAdjustGunForRobotTurn(true). » Nat | Talk » 10:26, 24 May 2009 (UTC)

As I read the source, it's the same as my version in Pugio. :) The only difference from my other bots could be the calculation of offset angle: somewhere I divided Math.signum(xyz) in spite of multiplying by the radian value. There's no reason, maybe it was easier to type. :D --HUNRobar 15:55, 25 May 2009 (UTC)

Worse, worse and worse!

Start from 0.2 to 1.0, a big JUMP! Now from 1.0 to 1.4, all the worse and worse =( If 1.5 don't get better, I start to have no ideas what's going on as I can't figure out what distance is best for 0.92 yet. » Nat | Talk » 13:51, 24 May 2009 (UTC)

Hi, I can't test your robot now, but I took a look at the source.
Try to make your movement react to enemy fire. I have done exactly the same with Pugio and it gained 3% APS and 5 places. :) After every tweak I test my bots with NanoAndrew, NanoLauLectrik and Acero for 2-3 battles. For example, if a tweaked version of Pugio can't reach 89-90 % against NanoAndrew, 59-60% against Acero and 40-45% against NanoLauLectrik, than I trash or improve the tweaks.
Considering Ocnirp 1.4 at 90-95th place, there is some kind of bug or big problem with it. With a PM gun it shouldn't score lower than 50th place. --HUNRobar 06:57, 25 May 2009 (UTC)
Thanks for the advices. 1.4 is at #59 actually, but 1.5 is at #91 =( I now think the close problem is not good for Ocnirp, I'll try the longer distance (250 or 300 again). Only bot I tested with is Pugio 1.4 actually... (first version is tested with WeekendObsession and all Neophyte) And PM gun now are hard-coded BFT so that may be a problem (46% against N is quite unacceptable for once top-20 nanobots)
I may work around with this robot a bit, if it doesn't work and it even reach #150, expected 2.0 to be 1.0 again. » Nat | Talk » 12:01, 25 May 2009
You really need to test against Acero and one other simpler, HOT bot. Acero's a great bell weather bot. He's anti-pattern matching on movement and uses head on targeting. If you are getting destroye by him, look at your movement to make sure it is protecting against HOT. --Miked0801 13:23, 25 May 2009 (UTC)

(UTC)

Really? Surely don't! I have a lot of problem with this Acero, even the OcnirpSNG get 37% it! (and this 1.5 get 27% by the way) I actually didn't think there will be any HOT bot out there so I didn't think about it. (I always think my movement is hard to hit with PM/LT/CT, which use more common in NanoBot, but the 1.1-1.5 have prove me wrong) I've may fully retire this and make something better! (I'm creating statistical bullet dodging, but it didn't fit with PM so I need to squeeze it.)
You're wrong! There are many HOT-bots out there. Just write some by heart: Acero, Bluebot, Bichephal, Hawking, Hotspur. And also watch out for random-targeting bots, such as DuelistNano, Smog (maybe, I'm not sure), Squirrel (actually LT/Random), Splinter (same as Squirrel), or my first bot, Breeze (LT/Random again) and many others. I always test my bots carefully before releasing them. It's an old mistake to submit bots to rumble untested. So I recommend you:
Anyway, I've made a nano Anti Pattern Matching movement, which pattern matches itself and tries to move differently than the estimated movement. It's only in an experimental state but if you want I can submit it to the repository. ;)

--HUNRobar 15:55, 25 May 2009 (UTC)

Hi. I have to ask what happened to Ocnirp, as 1.5 is just the 93th in the nanorumble? It's still a very strong enemy against PM, but it's almost 80 places lower in rank than the 1.0. --HUNRobar 16:44, 19 June 2009 (UTC)

Yes, it is really 77 places lower than 1.0. Few I ideas I have is too close distance with too random gun. The 1.0 is much depend on the starting environment since it never control its distance and use the same 0.92 constant at every distance. I need to run a ton of battles to figure out which value is the best. The 'too random gun' mean that I tuned my gun to be accurate but with no setAdjust...(true) =) » Nat | Talk » 17:09, 19 June 2009 (UTC)

The Difference of Pattern Depth

Hmm... Does anyone know that how much the pattern depth can effect on the pattern matching gun?

I must say that Ocnirp's movement is very weak against simple targeting , so the main point the give Ocnirp current position in the rumble is gun, which is based on Simonton's pattern matching gun.

Now, 1.61 use 30 as pattern length while 1.61b use 45 as pattern length. There is a quite noticeable difference between them. (see [1]) Fine-tuning this constant should probably give me points, especially when Miked08011 isn't here to update LittleBlackBook =) » Nat | Talk » 13:12, 13 September 2009 (UTC)

In Toorkild I have a pattern depth of 66, but I multiply my depth by 2/3 every time I look for a new pattern, not just subtract 1 like you guys in nano do =) Basically, the longer the pattern the better, until you get to the point where it starts skipping turns. --Skilgannon 13:30, 13 September 2009 (UTC)

Nah, it depends on enemy. See this (the same link as above), 1.61b use longer pattern but it is worse again some enemies (in my test: Weekendobsession, Avesna etc.). In my test both robot's skipped turns rates are about the same. I wonder what pattern depth like Waylander will do (512 /= 2) (in 1.72, tomorrow release) » Nat | Talk » 13:57, 13 September 2009 (UTC)