[Home]LittleEvilBrother

Robo Home | Changes | Preferences | AllPages

Difference (from prior major revision) (no other diffs)

Added: 95a96,97

Nah, WhatTheHeck may beat some really good bots, but it does crappy against bots who don't wait for their gun to turn before firing. So it would just have a really high specialization index. -- Kawigi

Little Evil Brother (LEB for short)- trying to be annoying

A NanoBot by PEZ

This bot is just in the early stages of an idea in my mind. I have had nano ideas before and they have all turned out to be crap and/or way too large. Which means you should maybe not be too impressed before LEB actually is released.

The goal is to at least for a day or so rule NanoBot land, and break the 1800 barrier in general RoboRumble@Home rating.

Here's the features I want it to host:

Feature Bytes when implementedEstimated rating points worthComments
Movement and Targeting92 1200 This is about what you get for free if you just move and fire (I think)
GuessFactorTargeting 200
Distance DataSegmentation 100
Velocity DataSegmentation 50
Acceleration DataSegmentation 50
Flattening RandomMovement 150
WallAvoidance 25
HeadOnTargetingAvoidance 0 40
Total realized 92 1240

Bytes left: 157


Questions / comments

1800? in general rumble? good luck, you'll definitely need it =) -- rozu

Thanks! It's all actually Kawigi's fault. He raised the question the other day on how far a nano can reach. And he stated the importance of thinking outside the box. And that's what I will try to do now. -- PEZ

Hmmmmm, yeah, I think I might have set my goals a bit too high... The bytes are flying away under my hands! But if my estimates are anywhere close to reality I now (at 92 bytes) have should have some 1240 rating points. Of course, something like that could be said about Walls too I guess. =)

1st: pez.nano.LittleEvilBrother 0.1       8201	3650	730	3236	568	16	0	78	27
2nd: davidalves.net.DuelistMicroMkII 1.0  2163	1350	270	505	37	1	0	27	73
-- PEZ

i must be doing somthing terribly wrong...gun not working probably. i have every feature there and my ranking is awful.--andrew

1760 isn't awful. "We can't all be heroes, someone has to sit on the sidewalk and clap as they go by" - Will Rogers :-)

Marshmallow had even more then that and still lost against DuelistNano :). robocode isn't a science it's a religion. -- rozu

...or an art, maybe. -- Kawigi

All hail the robocode gods! How does LEB fare against the regular DuelistMicro? DuelistMicroMkII was a miserable failure. :-( --David Alves

Well, I don't know yet. But tomorrow I can tell you. The test against MkII? was just to show that the HeadOnAvoidance? was in place. Though it's removed now, because now way in hell that I can fit it! =) LEB is now 239 bytes. With all but the last two features there. (And the flattener isn't really flattening.) -- PEZ

MkII? doesn't use head on targeting. :-p --David Alves

Hmmm after watching it fight Eve... maybe it does use head on targeting... It wasn't supposed to. Must be a bug in the gun. :-p --David Alves

I can't imagine that you got distance, velocity, and acceleration segmentation in a guess-factor gun in a nano. Is this a typo? I sure hope so, or I've been outdone! -- nano

I think it's still a speculation. -- Kawigi

No it's not speculation. The gun is very small. What I can't seem to fit is wall avoidance. Even if I think head-on avoidance would pay better, it feels silly to see the bot bang itself to death against the walls. -- PEZ

David. So it's a bug! Think of for how long that bug has been there. We are quite a few people who has thought for quite a while that it should be using head-on. It's one of the bots giving us head-on avoiders rating points. -- PEZ

I recalled David saying somewhere that it was meant to work as a pattern matcher, but it's pretty apparent from the LRP graphs what it really does. :-) *Checks code* interesting, I tried to develop a gun like that several times. They were pretty lousy even without bugs, though. :- Oh, and if you can fit all of those features into a nanobot then I'm truly shocked! That's an incredible work. -- Jamougha

Well, it's out now. Check it out. It might lessen your shock. =) -- PEZ

I did. :-) I kind of got the idea when I saw the file size. Interesting, I'm still surprised it fits. -- Jamougha

It's an old technique used by SandboxMini, some versions of Tityus and surely a host of other bots. RaikoMX is going good! I'll try find some time to check it out later today. -- PEZ

Wow. It's going to be a hard struggle to push this bot above 1800 in general... -- PEZ

Clever, but cheese. SandboxMini never did that. SandboxMiniMelee has the ability to load data without the corresponding save ability, but it only uses it to load an array storing the points to drive to in it's corner movement pattern. It doesn't use another bot to do its learning for it, so I wouldn't call SandboxMiniMelee cheese. --David Alves

BTW thanks for pointing out that DuelistMicroMkII was using HeadOnTargeting. I looked through the code for 1.0 and the gun was nowhere near complete... I don't know what I was thinking when I released it. I went back and fixed the gun. Of course, knowing my luck its rating will now plummet even lower. :-p --David Alves

I like cheesy! And Little Evil Brother likes to be annoying. =) And using another bot or hardcoding the values or commenting out the data gathering / saving is about the same in cheesy I think. -- PEZ

I wrote a bot like this once. It was in the minibot challenge for one week and took first in nano 1-v-1, but I withdrew it because I felt it was a cheap way to win. Daniel Pereira did the same the following week (his was better) but also withdrew it after 1 week. Of course back then prelearned data was a brutal advantage, but the general quality of bots now is so much higher that it's not as effective. --David Alves

Indeed. I have released tons of bots preloaded with data. That has been my cheesy strategy since I implemented data restore in Marshmallow eons ago. -- PEZ

I don't think loading a bot up with data is cheesy if that bot contains the code to learn and save data. I just think it's cheesy when you drop the code to learn and save so that your bot can make it into a lower weight class. --David Alves

Well, I have done that before too. I don't mind cheesy at all. In love and war... -- PEZ

Ouch. Very ouch. :P -- nano

And yet you want everyone to show their source? In love and war... And if there really were no rules, then we already know what bot is the best bot. WhatTheHeck wins by a large margin. -- nano

Ah, but we have worked with the rules quite a lot. It has been quite clearly stated that preloaded data is cheesy, but allowed. Remember that I just want everyone to release their bots as open source. I do not in anyway demand it.

Anyway, I wanted LEB to be annoying, which seemed to work with David. But I don't like it to be offending, which it seems to be for you. So I'll withdraw it. Nano land has done well without a bot of mine for very long. It can be without it a while longer. I don't think you would have liked it at all when I released the bug-fixed version with tweaked movement. =) -- PEZ

Go ahead and release it...I'm interested in where it would stand. -- nano

Nah, WhatTheHeck may beat some really good bots, but it does crappy against bots who don't wait for their gun to turn before firing. So it would just have a really high specialization index. -- Kawigi


Robo Home | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited February 28, 2004 17:48 EST by Kawigi (diff)
Search: