Archived talk:Dookious 20061129
Well, it's nice when the bug-hunt for what turns out to be a moronic typo leads you to finding other (possibly) significant bugs in your tank... :) -- Voidious
That's more like it! I was wondering what happened to Dookious in the last release. That's always the worst when you know you were thinking the right thing when you wrote a line of code but typed the wrong thing anyway. --wcsv
Hey, i just noticed that the current release of Dookious (version 0.972) is on top of the rumble. Congratulations to that. It could even be the new king. Im reinserting the top version of Ascendant now to check that. But either way, a very impressive release! --mue
Thanks, mue! It was more than I expected for this release, for sure. It's not the king, though - it's below both Ascendant 1.2.6 and 1.2.10 (which are the same code), in both rating and per-opponent comparisons. I'd have a hard time claiming the throne with less than a one or two point lead, anyway, since I think there's at least a 1 point chance factor. =) -- Voidious
You did it, a 5 point lead. You are the new King of the Rumble. Let me be the first to congratulate you, good work! -- ABC
Thanks!! =) I knew it was possible by the time I went to bed last night, but I wouldn't have dare hoped for a 2081 rating. Wow... I think mue deserves some kind of grace period to reclaim it, though, as he has been the King for well over a year now. But maybe I'm just afraid of the responsibilities of Kingship. :P -- Voidious
After a reign of 17 months, the dictator Ascendant has finally found its master. Congratulations with this excellent result. (Although you are not the King of PL ;-) ) -- GrubbmGait
Congratulations,now you'll be able to release Dookious 1.0 ;-) -- Florent
Bloody excellent, v! You really deserve the crown and I hope you'll be allowed to keep it a while. -- PEZ
Congratulations from me too. Really well done. And i dont think anyone needs grace period as you put it (wouldnt be of much use anyway ;-). Time to put Dookious on the eternal list of roborumble kings. There is a page for them on this wiki, i just dont remember the name... --mue
Thanks, guys. Florent, yes, v1.0 is on the way ;) ... Mue, I of course know right where that page is, as I've been looking at it longingly for a while now, hehe: History/KingsOfRR. I just don't quite feel right adding myself to it... -- Voidious
Gratz, Voidious! Long live the king. -- Greywhind
For data saving you might want to look at how CassiusClay chooses to do it to save space. Yes, I only save VG stats, but still, storing all opponents in the same file saves lots of space. -- PEZ
I've seen how you do it in CassiusClay, I just had no idea it would save much space. I actually thought it would take up more space, since the enemy's name is stored in the filename instead of the file contents when you use separate files. But I'll give it a try. -- Voidious
That's an interesting thought. I have actually played with the thought to try encode data to save in the file name. But that was when I was into data saving more than I am today. Now I am more of a convict to ABC's school of learning_speed > data_saving. =) -- PEZ
Well, I certainly agree with that, too. I'm not sure how much the data saving is helping, but I think it's worth something, and some of my comparisons between versions with >= 2 battles are showing that. I've waited until now to implement data saving, because I always knew I'd end up spending more time on it than it's worth once I started tinkering with it =) The 0.98 gun got my best TC2K6 FastLearning scores without the data saving, too. Sometime soon I'll try that challenge with data saving turned on. -- Voidious
I should've listened to the LittleMan inside me that said not to post a new version so quickly =) I think I'll re-post 0.98 once 1.0 hits 1,000 matches. I think I'll try the same code and no data saving for the next version... -- Voidious
Well, 1.02 has lost some points in rating (so far), but it seems to be particularly crushing some of the top bots in the rumble. It could just be because the Main Gun is worse, so the AntiSurfer gun gets chosen more easily. I'll definitely have to investigate... (I really must learn to make less changes with a version update, unless I'm very confident in all of them.) -- Voidious
Man, I really should just sit back and enjoy that top spot for a moment, but I can't help myself from trying to find some points in my gun! =) -- Voidious
Since Ascendant's gun's true power doesn't show itself in the TargetingChallenge's, I have been using the RRGC to benchmark my gun progress. I gained 4-5 points in my gun there, and with data saving turned off in 1.045 compared to having it on in 1.0. But, so far, Dookious 1.05 (gun from RRGC 1.045) is quite a ways behind Dookious 0.98 (same gun as 1.0) and Ascendant 1.2.20. Maybe it will turn around, but if it doesn't, I'm going to be so confused! -- Voidious
Sometimes things can be so simpel . . . and sometimes not. I think that the problem is your movement! Before you are getting more confused, let me explain my thoughts. Dookious uses WaveSurfing, meaning that no HOT-, LT- or CT-gun can touch it. It does not matter if you hit your opponent 10 times or just twice, you win with 98-2. Raiko's movement is relatively vulnerable against simple targeters, especially HOT-bots. So with DookiousRRGC it does matter if you hit your opponent 10 times or just twice. My conclusion (and only mine) is that gun v1.045 does slightly better against the lesser bots and slightly worse against the better ones (maybe only a few promille on average). In Dookious you do not have the positive influence because of its surfing, but encounter only the negative influence. So the best gun and the best movement put together does not always result in the best bot. But that is one of the charmes of Robocode, even if you are the bot to beat! -- GrubbmGait
Thanks for your insight, Grubb, I'm sure you must be right about this. I knew how infrequently simpler targeters hit Dookious, but I was under the (false, apparently) impression that it would still pay to improve my gun against them, since they still hit him sometimes. I guess, at this point, the only way to benchmark Dooki's gun is attached to Dookious! -- Voidious
And of course we should wait to see the results for Dookious with this gun before we can form any sort of theory. =) -- PEZ
Hehe - yeah, I get a little anxious at 1% per-opponent after 100 pairings =) -- Voidious
The latest version of Dookious is a testament to the power of taking a break from a tank to clear one's head... Hours and hours of work since 0.98 have resulted in barely any rating boost, but recently I've been working on Lukious or nothing at all. Yesterday, I spent about an hour experimenting with some changes to bullet power stuff in Dookious, and after almost 300 pairings, it's halfway between the best Dookious and Tyranius. Yay! -- Voidious
Hope it works for me too... I haven't worked on Strength for a month or so - I was getting annoyed at the lack of improvement past 97% against WSCBotA - maybe when I start trying again I'll find a fix. -- Greywhind
I definitely expect 1.070 to lose some rating points, but I would really like to see where these changes put me right at this moment. I've basically tweaked it to the point where it does only slightly worse than 1.06 against my primary test bed, so it will also tell me how good a TestBed it is. -- Voidious
Congratulations, it looks like 1.09 is the PL king, with 56% against Shadow after 3 battles. Did you run some 500/1000 battle tests? -- ABC
Thanks! No, but I ran 10 35-round battles with gun data saving turned off in Dookious - Shadow won 9/10, with almost 58% total. I'm a bit shocked at the victories in the rumble, but I'll take 'em =) How goes the rewrite / refactoring of Shadow? I never mentioned how flattered I was by your comment about "dedication and attention to detail". Best of luck. -- Voidious
That is good to know :). Shadow's rewriting is currently stopped, my initial attempts resulted in decreased performance, and I don't currently have the time/energy to do it from scratch. I had some fun with ShadowTeam recently, I still have some ideas to try with it. I know I'll be back to full Robocode addiction one day, but right now I'm enjoying the summer and my new DS Lite... ;) -- ABC
Man, I updated my test bed info above with some new research I just did... scary!
| Challenger | Dookious 1.09TC | Shadow 3.66d | Score | Comment | Dookious 1.10 | 60.40 | 59.80 | 60.10 | 2 seasons | Shadow 3.66MC | 70.62 | 67.65 | 69.14 | 2 seasons
Note for those who can't read my thoughts: That's 500-round TargetingChallenge style scores, Shadow and Dookious guns on the left vs Shadow and Dookious movements on the top, over 2 seasons. -- Voidious
This is exciting!! Will you be the first to officially break the 2100 barrier ?????? -- GrubbmGait
Not if I can help it! -is testing out some mods to DookiSaber for use in DoctorWho- :-p Plus he's down to 2095 now. :-/ -- Krillr
Yes I will be, Grubbm, but not with this version =) I'm very excited that such a simple CurveFlattening tweak yielded such an increase in points, though. And btw, the 2100 barrier will not be broken with improvements to DookiSaber, I'd put money on that! I'm not sure even Mue could squeze many more points out of the current top Rumble guns. -- Voidious
Ya, you're probably right. But, if it does happen, I always include my source with my bots so you can see what mods I've made to get there (which I probably won't :-p). -- Krillr
Wow, not such an increase in points after all... but I'll take what I can get! -- Voidious
It's insane how touchy movement changes can be ... but I'm not going to give up on a change that brought a victory against Shadow. =) -- Voidious
"Yes I will be" That remains to be seen. ;-) --David Alves
You know, I regretted that arrogant comment later in the day, but it slipped my mind to comment on it again =) But I am certainly going to try... and I think this new TestBed really helps. -- Voidious
2106 points after 480 battles . . . will this be The One ?? -- GrubbmGait
Nah, but hopefully it'll get to 2096 or 2097. It's pretty close to 1.125 in the comparison... Maybe the next one, knock on wood. =) -- Voidious
You sure had me scared for a while... very nice work. Your next one will make it for sure. --David Alves
- That makes one of us that is so sure about that... -- Voidious
- And maybe that "non-data-saving" TestBed is exactly the worst place to test CurveFlattening changes. :P But maybe I can still learn something. -- Voidious
|http://www.dijitari.com/void/robocode/Dookious_saved_data_comic.png
..wow..awesome -- Alcatraz
Congrats in advance on 2100. (I'm headed to bed soon) Are you going to enter yourself in The2100Club or wait to do it without preloading data? Either way you've definitely earned your spot in it, very well done! --David Alves
WOW! 2107 nice work! I wonder if 2200 is possible, what do you think? :D Or is your next project a 2k team? --Krabb
- From the long evolution that bots have taken so far, it's tough to say what's "possible" =) But it seems to me that 2200 is maybe maybe possible with pre-saved data, and 2130 or 2150 maybe without it. So hard to say, though! -- Voidious
Yuk, pre-saved data. Will you keep it that way? -- PEZ
I have very mixed feelings on the subject. When I started implementing it, I thought I was just playing catchup to David's "wicked" version 0.666 of Phoenix, which got off to a super strong start in the rumble. On principle, I've never had a problem with pre-saved data: the Robocode packager packages it and extracts it to the cache, it's been agreed upon to allow it in the rumble, there are other rumble bots that do it, and even one RoboRumble King did it. But I've also never been the one doing it, and it sure feels dirty!
On the one hand, it feels like setting yourself "outside" of the RoboRumble. What I mean is... Something like 80% of the bots in the rumble (or more) will probably never be updated again. To fight them with pre-saved data is really straight up unfair, and it's no longer fighting on equal ground; at that point, it's less like I'm competing with them, and more like I'm competing with other active bots to see how badly I can beat them. Then again, the whole top 20 is kind of like that, anyway. Ultimately, I think the most logical and fair way to view it is this: it's allowed or it isn't, and it has been the accepted practice that it's allowed.
As someone who has competed seriously in sports and games quite a bit, I think it sucks in a way when something is "allowed but looked down upon". What's "cheap" one minute is "stupid not to do" once everyone learns the wisdom of it. So for The2100Club and the KingsOfRR listings, I think pre-saved data (as long as it's under 200 kb) isn't really an issue. It reminds me... When I was playing Quake semi-seriously, and mainly 1v1, I initially thought camping for weapons and armor was cheap and detracted from the fun; once I started playing other serious gamers, though, I soon considered it simply unwise to go into battle unprepared!
So that's how I feel about it at a high level. Will I keep this pre-saved data in Dookious, though? The cut throat competitor in me says it's silly to pull it out. But it really just doesn't sit well with me. I think I will return him to his normal data saving ways, and: a) keep this trick up my sleeve if I need it to defend the throne =), and b) maybe explore the depths of the "dark side" later and see how far I can push the pre-saved data advantage. Whew, I'm outta breath!
-- Voidious
To me it's just inellegant. It doesn't make a bot stronger in any real sense. And from a competetive side it risks hiding other potential room for (real) improvement. I would certainly have tested it like you have now, it's interesting to know what effects it can have. There's the reason why CassiusClayWT never entered the RoboRumble, since SilverFistWT already had given us a pretty good idea on what preloaded WikiTargeting would do to CC's scores. For the record, say a non-preloaded Dookious was dethroned by a preloaded bot. In my book it would still be Dookious that is the King, regardless of what the official story is. As I see it the RoboRumble@Home is there mainly for us to explore how to make stronger bots. Exploiting the flaws in the RR@H environment itself seems not to be in that spirit. But that's just me. All is fare in love and war as they say. =) -- PEZ
I don't think data saving is "the dark side". It sure is an itresting part of a bot and it makes him stronger! In my eys just an other part, like targeting and movement. You can compare it with real world copetitions (like Volleyball, what i play :/). You first train general strategies (could be compared with the programming of a bot) when you have learned the basics you optimate the strategies (could be compared with parameter tweaking). In a match against a new Team you try to realise the learnd strategies, especially the ones wich work well in the current situation (for example GF-Targeting in robocode). But after that match you analyse your faults to win the second leg. If you can adapt your strategie successful you sure have an advantage. Same goes for robocode. But why should data saving be bad? I agree with PEZ, that new strategies could be harder to test with data saving, and bots can become really complex. But for me this only shows, that it is hard to do data saving. You have not endless disc space, you must choose between quantity and quality, and you must be aware of adaptive movements :) --Krabb
Well, I don't know if you realize this or not, but we are also talking about the idea of "pre-loading" data about RoboRumble opponents. I see pre-loading as a very different thing than just data saving. Lots of older bots save data on Dookious between matches, and Dookious saves data on them, too. They are on equal footing, learning from the same battles. But if I make a bot version that comes pre-loaded with data learned from hundreds of rounds against each of those bots, that is really just kinda unfair (in my opinion), and definitely a different thing than just saving data between matches. -- Voidious
So, are the terms for being TheKing going to specify that you can't steal the crown from someone else by using preloaded data? Personally, if Phoenix 0.666 had been a success, I was planning to enter myself in The2100Club but not crown myself as TheKing. --David Alves
Yeah, that sounds good to me. -- Voidious
Yes, data-saving is a totally different story.
I don't think we can alter the "rules" for RR@H and say that pre-loading is now forbidden. We can make that a rule for RR@H 2.0 maybe. But there's nothing wrong with a truce like that one you guys now entered. Way cool. Let's hope we can convince the next throne challenger too. =) -- PEZ
Ok, i missunderstood the "pre-loading" thing :) And there isn't a real example in my real world comparison, because you cant try your new strategiel vs. a virtual team... And i must affirm this kind of datasaving is inellegant.
EDIT: But may be "pre-loading" is a renewal of the 1on1 robocode competition. Nothing really changed in the last few years, the only improvements new top bots have are optimated strategies and technologies which where developed before i started robocoding. "pre-loading" could be a new field, with new learning algorithms and other undiscovered possibilities. "pre-loading" might be the only pssibility to get a much higher ranking, like Voidious already said. --Krabb
It isn't new in any way. -- Voidious 22:03, 10 November 2007 (UTC)
The general idear is not new, but it is not that squeezed out. I thought of new technics wich might develop from data pre-loading. --krabb
As far as targeting is concerned I think it is pretty squeezed out. A slower or easier saturated gun will gain more than Dookious did, especially since D collects most of its points through movement. But even at the very top of gunning and dodging there are still some points to gain by carefully selecting what bots to pre-load data on. But maybe, for a learning movement there are even more points to gain with preloaded movement data. -- PEZ
Wow ... Dookious 1.144 lost 7 points from 1.14, and all I adjusted was weights in the surfing. I halved the RollingAverage depth when the CurveFlattener was enabled, and slightly lowered it against high-ish hit-% guns when the flattener was disabled. On the one hand, losing points is no fun - but now I have much hope that I can find some points in tweaking these weights / rolling depths. Fingers crossed... -- Voidious
The-glass-is-half-empty-comment: My experience with those finding is most often that I magically got it exactly right the first time and no amount of tweaking can improve it. Hopefully not the case this time. =) -- PEZ
Yeah, weird but true =) Makes sense that you might tweak (other things) around stuff like that. But I'm still hopeful, darn it! -- Voidious
I was getting some somewhat godly TC2K6 fast learning scores with 1.152, but as soon as I released it, I realized I had a bug in that new segment I added. :-\ I couldn't help but fix it... I guess it's time to run more tests without that segment, hopefully I can keep those scores. (And hopefully they mean RoboRumble points.) -- Voidious
- Hmm, not to worry, the scores are looking even more godly with that buggy segment removed =) Still, we'll see if it actually gets me any rating points. -- Voidious
A 49.6% loss against Aleph, unfortunately, keeps Dookious 1.18 from being undefeated... But I think Dookious may finally average over 50% against Shadow! I'll run a couple hundred matches overnight to get some accurate results (1.18 w/o data saving vs Shadow 3.66d), but it's very very close to even over the first 10 matches of my tests: 5 of 10 wins, 49.43% for Dookious 1.18. -- Voidious
- Killing them softly huh, a 0.1% at a time. -- Chase-san
You got me worried for a moment there... :) -- ABC
No need to worry yet... but check back in a couple days ;) -- Voidious
Didn't run enough tests with the final version of 1.181 to be completely sure, but I've got Shadow to basically a draw at this point. Got my AntiSurfer gun (alone, no VirtualGuns) from about 64 to about 69 in a 500-round TC, and tried to make my VirtualGuns less biased against the AntiSurfer gun. Hopefully I won't lose too many rating points with all of this. -- Voidious
Ugh... (No pun intended.) No offense to Martin, but that 49.9 against Ugluk bothers me. At first, I thought maybe it was a rammer loss, which I could maybe understand, but Ugluk 15.7 doesn't appear to ram Dookious. I ran 10 matches against Ugluk, the closest was about 2:1, average 70.5%. Also, it came from my client! Couldn't duplicate less than 75% against Neutrino, either, which 1.181 got 62% against. Oh well, perhaps my VG changes really are that dangerously unstable... (And if it was indeed error free, kudos to Martin... dammit. :-P) -- Voidious
- In general I would guess any drastic performance hit against Ugluk is more a problem with your movement than your targeting. My movement isn't really all that great, but my targeting is fairly competent. Then again my Grid movement is as close to random as I have conceived so far, while still incorporating dead-on targeting avoidance and to some extent guess-factor targeting, which is dependent on how similarly I segment compared to my opponent. -- Martin
- Not that I doubt what you're saying about Ugluk's targeting and movement... but in this case, my movement is very stable and I haven't changed it at all, while I did make significant gun changes, so I'm really hesitant to think it was a matter of my movement. -- Voidious
Well, it's quite clear that this is bombing, almost 1% per opponent after 160 battles... Now to make sure that reverting the VirtualGuns changes solves the problem. If it does, I can try to find a middle ground. -- Voidious
62% against Neutrino? I dunno... I really doubt that's a VirtualGuns issue. You should get a lot higher than that regardless of whether your VirtualGuns system chooses your main or your AntiSurferGun. Maybe you have (had?) an intermittent bug that caused Dookious to crash? Did you check to see if that battle came from your client? --David Alves
1.183 not looking good for PL... maybe something went wrong in those AntiSurferGun improvements? =( --David Alves
I did remove the onBulletHit thing for the AntiSurfer gun, maybe that is causing the VirtualGuns to choose the AntiSurfer gun incorrectly a lot more often. Other than that, it really is just changes to the AntiSurfer segmentation. Weird! -- Voidious
- Speaking more in terms of rating here. Lost 11 points in rating, mostly not against surfers. -- Voidious
- Wow, went up 4 points after passing 1200 battles? Weirder and weirder... =) -- Voidious
Hmm... these 1.18* versions have confused me. I expected 1.18 to maybe take a rating hit, as messing with the AntiSurfer gun could greatly affect the VirtualGuns, but it rated almost exactly the same as 1.161. So I made some further, probably equally significant, changes to the AntiSurfer gun in 1.183 and 1.184, both of which really lost some points. I guess it could be as simple as 1.18's AS gun being more consistent and not being chosen accidentally, maybe even being straight up better against surfers, too.
Now I'm starting to get some fresh ideas about AntiSurferTargeting and identifying surfers (to augment straight VirtualGuns), and I'm starting to really feel constrained by Dookious' fairly bloated code structure. I developed it while in the process of learning about all aspects of targeting and I adapted the code as I went; that same system has housed non-wave VirtualGuns with dozens of instances of simple targeters, PatternMatching, horribly miscoded GuessFactorTargeting, true GuessFactorTargeting, and wave-based VirtualGuns. Considering all that, I'm fairly pleased with how well the structure has held up, but I think its time is about up... I'm gonna start writing a new gun system from scratch to replace the current DookiSaber. Hopefully I won't have too many headaches trying to get the nitty gritty details to be functionally identical. *Knock on wood.* Then I can, hopefully, rewrite the movement (that code makes me ill), integrating with some of the general gun classes like it does now.
(Not that many people particularly care how Dooki's gun is coded, but maybe writing it out will serve to remind and motivate me to get it done...)
-- Voidious
Hmmmmm. I think 1.20 fell victim to refactoring. Lot of wins but some strange losses... --Corbos
Finally *blush* switching to radians in the gun rewrite, my trig was still a little sloppy when I started Dookious, so it's all in degrees. The culprit was: if (((Math.abs(lastBotScan.getBearingRadians()) < 25 in my rammer detection; 1.20 thinks everyone's a rammer, eventually, so starts firing 3.0. Version 1.21 coming soon... -- Voidious
Alright! I think I can safely chalk up 5-6 of the 7-8 point rating difference to VirtualGuns subtleties and lack of data saving. Maybe 1 more to not doing the windowed smoothing stuff in the AntiSurfer gun. The 1.22 gun is 30% faster than the 1.161 gun; making it only aim when gunHeat() < .4 makes it more like 40% (for the next version). Finally not a SlowGun. I'm happy =) -- Voidious
Ah, 2107 with the new gun makes me so happy =) And saying "40% faster" isn't quite accurate, it takes 40% less time to execute, so it's close to twice as fast. -- Voidious
No matter how you say it, your score and speed are impressive. Nice work. Refactoring is the most difficult part of enterprise software and you pulled it off with grace. --Corbos
Whoa, 2113?! Holy crap. Nice work. --David Alves
Thanks. (2112 now - go Rush!) I have a very rigorous and scientific development process to thank for these points:
- Tweak AntiSurfer gun by itself in TC500 vs Shadow.
- Release.
=)
-- Voidious
Hah, nice process Voidious, You are a scientific genious. =) --Chase-san
Man, 1.26 beat Shadow 12 of 15 matches, total about 53.26% in favor of Dookious. Really hope I can get that one back =) Also, there's still time for it to change, but I will be somewhat shocked if I lose 8-10 points over a fast learning change in my AntiSurfer gun... the only real explanation, I think, would be that it is now interfering more in the VG than it was before. Then again, the 1.25 rating might have been a bit of a fluke, as it seemed that way to me already. -- Voidious
The 1.271 AntiSurfer gun gets around 71 in TC500 against Shadow, by far my best score, though comparable to what Phoenix's AntiSurfer gun gets. Another benefit of this AntiSurfer trick I'm trying (see /VersionHistory) is that it may make the AntiSurfer gun significantly worse against non-surfers, helping my VirtualGuns. -- Voidious
This movement rewrite may be a much longer road than I anticipated... =( -- Voidious
Ok, 2089 is something... There was a behavior in the old Dookious that seemed so ugly and unnecessary to me when rewriting it that I decieded to try to drop it. In 1.32, it's still not implement correctly, so I'm not 100% sure it's the big problem with the rewrite, but it might be. When Dooki gets within a certain distance of an enemy (about 75 usually), he starts WallSmoothing away from the enemy instead of towards him, and keeps smoothing that way until he's not smoothing anymore. It's a real pain to get it predicting all of that correctly. But I guess I will muddy up my rewrite with all this weird smoothing logic after all... -- Voidious
Finally found a bug that could be the main issue with the new movement: my "find closest wave" method is searching through all the waves, not just firing waves, so it could be matching bullet hits to nonfiring waves. Interestingly, this would also be affecting the VirtualGuns, AntiSurfer gun, and that AntiSurfer trick I've tried; I'm going to leave the bug in place in the gun for now for sake of consistency. The MC2K6 for 1.352 is running now and looking quite good, but I'd trade MC2K6 scores for rating points any day of the week =) -- Voidious
One question pops up: Why should you keep track of the enemy's non-bullet waves. As far as I see it, you can't do anything with them. Good luck with your pointhunt. -- GrubbmGait
I use the non-firing waves as part of my curve flattener (when it gets turned on), and I also surf a non-firing wave if there are no firing waves in the air. I got one or both of those ideas from PEZ originally. -- Voidious
Well, I have been looking through the old (and incredibly messy) Dookious movement code for discrepancies that still remain between it and the rewrite. I laughed out loud when I realized that I broke that "smooth away" behavior by accident at some point. So while the complex code is still in place, the most recent old Dookious didn't do that! I can safely remove that stupid behavior from the Dookious rewrite...
Now to comb through my past versions and find out when it got removed. I bet it was when I implemented the precise escape angle in the gun; it might have even been part of that 8 point rating boost. (The TC scores still went up a whole point with that, though, so the precise escape angle is still worth doing IMO.)
-- Voidious
- What a guess, it was with 1.16 that I broke that smooth away behavior. I bet it contributed to my breaking 2100! Talk about a PerformanceEnhancingBug... -- Voidious
Oy... Yes, I did break the "smooth away" functionality in 1.16, but not completely. Oddly, this same "broken" way of doing it is how I first re-implemented it in 1.32. At least it's cleaner to code it this way... Hope it yields some points. -- Voidious
Revision 2197 . . November 25, 2006 15:46 EST by Voidious [Dookious 1.365] Revision 2196 . . (edit) November 25, 2006 15:41 EST by Voidious [revert to Dookious 1.362 (again!)] Revision 2195 . . November 25, 2006 15:32 EST by Voidious [Dookious 1.364 (bug fix)] Revision 2194 . . (edit) November 25, 2006 15:27 EST by Voidious [revert to Dooki 1.362 (bug...)] Revision 2193 . . November 25, 2006 14:56 EST by Voidious [Dookious 1.363] Revision 2192 . . November 25, 2006 10:49 EST by Voidious [Dookious 1.362] Revision 2191 . . November 25, 2006 8:49 EST by 83.234.48.236 Revision 2190 . . November 25, 2006 2:32 EST by David Alves [Phoenix 0.859] Revision 2189 . . November 24, 2006 23:06 EST by Voidious [Dookious 1.361 ...] Revision 2188 . . November 24, 2006 21:36 EST by Voidious [Dookious 1.36] Revision 2187 . . (edit) November 24, 2006 20:34 EST by Voidious [reverting CassiusClay to 2pi.08 (at PEZ's request)] Revision 2186 . . November 24, 2006 17:21 EST by Voidious [Dookious 1.352] Revision 2185 . . November 24, 2006 13:41 EST by Voidious [Dooki 1.351] Revision 2184 . . (edit) November 24, 2006 13:38 EST by Voidious [doh, bug fix coming...] Revision 2183 . . November 24, 2006 13:12 EST by Voidious [Dookious 1.35] Revision 2182 . . November 23, 2006 22:42 EST by Voidious [Dookious 1.34] Revision 2181 . . (edit) November 23, 2006 21:21 EST by Voidious [revert to 1.32] Revision 2180 . . (edit) November 23, 2006 20:12 EST by Bayen Revision 2179 . . November 23, 2006 19:59 EST by Voidious [Dookious 1.33]
12 edits in less than two days? Don't you ever sleep?! =) --David Alves
I won't sleep well until Dookious is back to 2114. :-P -- Voidious
If it turns out the lost points from the gun in the refactoring, and it's looking that way, I'm going to be very happy! I have been going insane looking over the new movement code, which seems quite correct and bug free to me. If so, I'm going to have to figure out what about the precise escape angle stuff in the gun rewrite gained me about 10 points... -- Voidious
Huh? I don't understand what you just said. --David Alves
That'll teach me to post after a few shots of rum. =) I don't understand it either. But here's what I meant:
Since I rewrote the movement, I have still been 8-10 points below Dookious 1.25, my best version. I have been assuming that my gun after the movement rewrite was equivalent to the gun before the rewrite (in 1.25), but I think some changes to code common to the gun and movement may have messed up the gun. Dookious 1.367 is the 1.3666 gun and the pre-rewrite movement; if it does no better than 1.3666, then the movement rewrite is OK and I messed up the gun in the process. -- Voidious
