[Home]Tityus/DiscussionsAsOf20040104

Robo Home | Tityus | Changes | Preferences | AllPages

No diff available--this is the first major revision. (no other diffs)
In the scanned robot event you have :
        setTurnGunRightRadians(Utils.normalRelativeAngle(
            enemyAbsoluteBearing + maxEscapeAngle(bulletVelocity(enemyFirePower)) *
		sign(deltaBearing) * mostVisitedFactor() - getGunHeadingRadians()));
should that be bulletPower instead of enemyFirePower?? Was wondering why it wouldn't beat DuelistMicro... :)

- Jamougha

Phew! Thanks for telling me. I often have problems against the small Duelists, but this bug might be giving it an unfair advantage. =) -- PEZ

Now a new Tityus is out there. I added it to the RoboRumble@Home again now that it might have a chance to enter the top-30. I'm curious Jamougha, how come you tested Tityus? It has been dormant for some while now.

Kawigi, if you read this, please update Tityus before the next season of RobocodeLittleLeague. It might be version 0.5.1 or 0.5.2 then, if I find something to tweak. =) -- PEZ

I've been coding away from home, and rather than re-write my old gun I decided to borrow one for the duration; Tityus' is almost identical to mine, though much more elegantly written. I was quite shocked when it did worse than head-on targeting :) but I fixed and changed a few things. Congrats on the #8 re-entry, btw! - Jamougha

You are entirely welcome to borrow Tityus gun. Just remember it's RWPCL so you must show what you changed. =) Again thanks for helping it recover from that ugly bug. I was very, very frustrated when T started to fall down the rankings and I couldn't fix it. Seeing the company in the top-10 in mini #8 isn't too bad! -- PEZ

Thanks, I may just keep using it and release the bot as open-source in a few days. Right now it barely beats 0.5.1 over a thousand rounds, and my movement is definitely not as good, so I figure you can get Tityus up a few more places yet. :-D - Jamougha

I'm sure Tityus gun can be improved. It's not exactly the best gun around. =) But I'm not sure Raiko beating T 0.5.1 over 1000 rounds really shows this. It depends on what you aim at. Tityus gun is opted for RoboRumble@Home short battles. Then your high segmentation might slow learning a bit. But it pays in long battles for sure. And even if T 0.5.1 movement is better than Raiko's (which I am not sure it is) it might not be better against T's gun. None the less I will look at Raiko's gun and see what I could consider to tweak in my gun. This is the beauty of open sourcing a bot. I would never have found that bug without your help and the gun will surely evolve faster with two or more people using it than with just me. I hereby promise never to close the source of any of my bots again. -- PEZ

Yeah, open source is great. Saved me a good few days. ;-) I've had problems with over-segmentation in shorter battles too. For over a week my best gun vs most bots was completely unsegmented! Now that you're firing a wave every round you have a lot more leeway, though. Segmenting on whether or not a bot was approaching the walls was a big breakthrough for Raiko, do try it in Tityus. The other segment is a load of junk. :-) Hmmm, the nicest thing would be to have oodles of segments and remove most of them depending on how much data is in the bucket. Unfortunately I don't think I can fit that into a mini, but maybe for Raiko's eventual big brother. - Jamougha

FloodMini does that segmentation, but it's the first one I take out when I do guess-factor guns in Micros, not because it's not good (it helps against bots like Cigaret if I remember right, it was the segmentation that got me beating Cigaret, which was a couple versions ago of both bots), but because bots don't spend as much time in different segments there. -- Kawigi

That's an interesting observation. Maybe it's that different segmentation is needed? Like wall proximity and such. I'll try such a segmentation in Tityus once I have gotten the movement to work. Once you have a basic gun, movement is more important for your performance than targeting. -- PEZ

A real basic gun may also leave you with glaring weaknesses, but a reasonably segmented guess-factor gun without any coding bugs usually doesn't generally have that problem. And adding a segment for whether or not the opponent was approaching the wall jumped FloodMini from 30th to 12th in the ER (in retrospect, I'm surprised it made such a distance - more than half-way to the top). -- Kawigi

For the record. With "basic gun" I meant a "reasonably segmented guess factor gun". I'm not one of those counting GF targeting as "advanced". It's simpler than most other techniques. -- PEZ

Jim, if you test BlackPearl .69 against T 0.6.2, remove T's data file on BP first. I accidently packaged it with a saved file from a 1000 round battle against that BP. It was my last test before releasing... Oh, yes, BP won that fight pretty comfortably. -- PEZ


Robo Home | Tityus | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited January 4, 2004 14:15 EST by PEZ (diff)
Search: