User talk:Skotty

Bitten by poor web application design again. After having to pause one of my edits to take care of another matter, I returned a bit later, finished my edit, and submitted it. However, my session had timed out. Much to my surprise, MediaWiki did not preserve my edit. I thought MediaWiki was classier than that.

It is a newbie web developer mistake to have a web application lose someones edit due to a session timeout. The proper action is to save the edited data, give the user a chance to log back in, and then apply the edit or return them to the edit page without losing the edit. I'm often surprised at how many web applications exhibit this glaring flaw...so many, that I almost consider it my own mistake for not copying the text to notepad before submitting. But it is not my own mistake. It's the web application's mistake. A mistake that I am really growing tired of seeing again and again and again...

I can't believe how long it has been since I was last active.

Life keeps me quite busy, but I still think about RoboCode from time to time. I'm tempted to come back for awhile and update RoboJogger (finally get it to a stable 1.0 version), and maybe even work on a robot (maybe take a new look at the missed turns issue XanderCat had in the first couple of rounds).

I hope everyone is doing well!

I'm honestly not too excited about trying to improve on XanderCat in 1:1 combat. Deminishing returns and all. I'm actually more interested in updating my robot framework to fully support melee combat, and then building my first real melee robot. Not sure if I will just expand XanderCat for this, or create a separate robot (maybe XanderClowder).

I did a quick review of XanderCat vs Spitfire. Spitfire is my bot that only does bullet shielding. If you do a compare of these two robots in the rumble, you can see that the bullet shielding implementation in XanderCat is still failing frequently. Spitfire outperforms XanderCat by wide margins on many robots vulnerable to bullet shielding. So while I was thinking of just doing melee work, I may also go back and see if I can fix this deficiency in XanderCat again, as doing so could make a significant impact on it's overall performance. There are two things I can do for this.

First, I may need to do another round of trying to cut some fat out of XanderCat to address the missed turns issue that was previously (and I assume still is) a bit issue related to this. First on the chopping block would probably be the scenario for ramming other opponents, which was just there for fun. I'm sure there are other places that can be improved as well.

Second, I may need to tweak XanderCat into sticking with the bullet shielding for longer before jumping to other modes. Possibly making it continue trying to bullet shield if the number of missed turns is enough to be interfering. There is a balance to be achieved there between improving against bullet shielding vulnerable opponents and degrading performance against opponents who are not vulnerable to bullet shielding. But based on comparisons, I think trying harder to bullet shield is probably the way to go, as XanderCat is some 20 to 30 APS points lower than Spitfire against quite a few opponents. That's pretty significant.

I find that it is convenient to treat my bot the same as enemies bot inside the code. I.e. each bot has full information on past and present battlefield situation. My bot would fire real bullets, enemies would make "virtual" decision on firing, which mine is dodging. That assumes that enemies using similar guns and strategies. So I might even have a rough guess at which opponent an enemy fires in melee (if they use same decision tactic) and dodge or not dodge such bullet.

One can even attemt to extend it, and predict which way the enemy would dodge in melee taking in account not only my bullets but other enemies too. But this is CPU expensive and leads to missed turns, so I do not use it. Though, I feel it can be done relatively fast.

In version 12.8 of XanderCat, I partially fixed my bullet shielding components that were being seriously hampered by skipped turns due to what we believe to be a garbage collection issue that occurs on some platforms for bots that produce a lot of garbage (yeah, XanderCat is a trashy bot). This partial fix now has me neck and neck with Gilgalad (up from 7th to roughly tied for 4th).

The skipped turns remains a problem, but the bullet shielding is working much better now regardless. Originally I thought the problem was due to the time loss, but the real culprit was with how the opponent waves were being generated. When XanderCat sees an energy drop from scan to the next in an opponent that cannot be explained by anything else, it logs it as the opponent firing a bullet and creates an opponent wave for the assumed bullet. This is fairly standard. However, what happens if your robot has missed scans in a time period when the opponent has fired a shot (in my case, due to skipped turns)?

XanderCat would just assume that the opponent bullet was fired on the previous tick. For example, assume you have a scan from time 40, then have skipped turns, then the next scan you get is from time 50. XanderCat would assume the bullet was fired at time 49, when it could have been fired at any time between ticks 41 and 49. This resulted in incorrect opponent waves often being created. This error doesn't affect other strategies all that much, but it does break bullet shielding quite badly.

My initial fix for XanderCat 12.8 is fairly simple. If the scans are not back to back, the wave is just flagged as "estimated". Any components that then rely on processing opponent waves can take that flag into account if needed. I just have my bullet shielding controller ignore "estimated" waves so that I do not attempt to shield those waves and don't count misses against those waves in whether or not to continue shielding.

A better solution, which I may consider, is to try to actually figure out what missed tick was the tick when the shot was actually fired, and then use that tick in the opponent wave creation. This could be done by analyzing opponent gun heat. Most bots will fire as soon as their gun heat reaches 0, so if you know when the previous shot was fired, you can figure out on which tick the next fired shot should have happened on. How much additional help might this be?

Now, most 1v1 bots don't have problems with missed scans or skipped turns, so concerning oneself with this issue may not be worth the trouble. But it's something interesting to think about.

Nice work and congrats!

It's hard to guess how much the gun heat tracking would help, but to me it seems to have potential for a decent gain and thus seems worth trying. Even if it just lets you shield one more wave once in a while, that can be a big APS difference when you're holding opponents to so few points (as I'm sure you've noticed =) ).

Nice improvement indeed. In my opinion, gun heat tracking and assuming they shot as soon as possible will gain you approx the same amount of APS you got with this update, say 0.25. Not enough to threaten #3 yet, but surely a noticable jump. I think that the "improved estimated wave" can safely be regarded as "true wave", but that is something for you to test and decide. Note that such an improved estimated wave not only helps your shielding, but also your regular surfing.

Good job, you finally pushed past my bot (and 2 others for that matter). I was wondering when that would happen. I might have to get back to work on that one robot I never finished and just stole the gun from instead.

Everyone -- I've been pretty busy with work and home stuff lately, and have had little time for Robocode. I left RoboJogger in a reasonably stable state, though not entirely finished. I will get back to it in time. If anyone wants to get a hold of me for any reason, since I'm not checking this site very often right now, make use of the "email this user" function (assuming that works?). I checked to make sure my email for the site is one I actually check regularly.

I look forward to getting back to Robocode and the Robocode community in time, but feel free to poke me from time to time if you haven't heard from me in awhile.

I think everyone that's been into Robocode long enough has had long hiatuses at times, so I didn't really think much of it. But definitely good to hear you're not permanently checked out. As far as I know the email feature works - I certainly get email notifications for page updates and stuff, and I have to think it all uses the same mechanism.

Hey everyone. I've been away, busy with other things in life. Just thought I'd ping the site and let everyone know I still think about and really appreciate the Robowiki/Robocode community. I might have a chance to do a little robot work again early next year. I was somewhat stuck with improving XanderCat any further, but there are plenty of other things I never got into that could be fun (like melee or teams). Best of luck, and see you out there...

Hey, good to hear. :-) I've also been working on other stuff. Things seem to go in waves around here. Will look forward to you triumphant return. ;)

Anyone played around with trying to use an AffineTransform when painting on the battle field? If I so much as make an exact clone of the existing AffineTransform associated with the Graphics2D object and set the clone as the transform, nothing gets painted to the battlefield. Not sure what is going on there. Maybe something weird Robocode is doing. Example:

public void onPaint(Graphics2D g) {
	AffineTransform clone = new AffineTransform(g.getTransform());
	g.setTransform(clone);
	// paint something here
}

This will result in nothing being painted. Remove setting the transform, and it paints fine.

I'm trying to figure out how to shift the paint position for a paint framework I am trying to write. The general idea in play here is developing a means for moving informational displays around on the screen. I can't get the AffineTransform to work. I also tried to cheat and use graphics from a BufferedImage, then draw the image onto the battle field, but Robocode doesn't allow the use of the drawImage methods.

I would appreciate any information anyone has on using AffineTransform in Robocode, or any other ideas on how I might accomplish what I am trying to do.

FYI: I worked around this problem by creating a proxy class for the Robocode Graphics2D object called WindowGraphics2D. The proxy class uses a subset of delegate methods to support many of the Graphics2D paint methods; in those methods, it manually shifts the text or shapes to the appropriate screen positions. To use a window, you extend the abstract WindowPainter class, which in turn allows painting through an abstract method with signature paintWindowComponent(WindowGraphics2D graphics, T paintable).

My alternate robots, currently named Fiona, Mikey, and SamAxe, are meant to demonstrate different gun ideas (they replace my obsolete robots Nissa, Simon, and Rowdy). However, on my first release, they were more competitive than I was expecting. In the latest versions I will be rolling out over a period of hours today (versions 1.2, 1.1, and 1.1 respectively), I have swapped out the drives in each for something simpler. The old versions had poorly segmented wave surfing drives, while the new versions have simple anti-linear drives. This should tone them down enough to not cause ripples in the ranks and hog top 100 spots, while still being good enough to show off their different gun skills. The only robot of mine I want making ripples in the ranks is XanderCat.