A crazy idea I had to add extra data to the regular surfing stats, by marking places on waves where you know the enemy didn't fire on previous waves (because you were there and didn't get hit) as safe. It runs like a flattener, but in reverse, marking places you've been as safe instead of dangerous. I've got no idea if it will actually work, but I see no reason why it shouldn't. --Skilgannon 11:53, 13 April 2009 (UTC)
Take a look at (very buggy) BlackHole 0.1.11, it does implements this in order to dodge AntiSurfer targeting, although weight fairy low. Not sure is this actually work since I change both buffer and movement algorithm in this version so I don't know what make it perform really bad. If you implements this alone, you will end in sitting duck until hit, and sitting at new location again.
- Of course, by itself it is fairly useless because we don't know where they are shooting. But when used along with Hit surfing I think it could make my surfing against simple, non-adaptive guns more accurate. --Skilgannon 13:04, 13 April 2009 (UTC)
Actually, my ideas is to implements this along with normal stats with crowd movement. But no time for now since I need to create a really work wave surfing robot first.
- Yes, I've seen that, but here I'm doing it for a different reason: instead of only updating my dangers based on what is dangerous, I am also taking into account where is safe. On the AntiAntiSurferSurfing we are doing that because that is what the enemy is (maybe) doing. --Skilgannon 13:04, 13 April 2009 (UTC)
I've considered this type of thing before really. I came to the conclusion that it wouldn't be worth it, because while it could help against non-adaptive targeting and maybe some anti-surfer guns, it would make you far more vunrable against plain GF targeting by showing up in the same place more often. --Rednaxela 12:32, 13 April 2009 (UTC)
- Fast rolling Crowd Movement will work I think. But it will only work for top pair of GF/AS gun in rumble. » Nat | Talk » 12:38, 13 April 2009 (UTC)
- My experiments in CrowdTargeting show that there is too litte data to roll crowd weights fast. With movement, there is even LESS data to work with (only BulletHit and BulletHitBullet, instead of all firing waves), thus I'm very certain that such a thing as "Fast rolling Crowd Movement" is quite impossible to be implemented in a useful way, there is just too little data. --Rednaxela 13:43, 13 April 2009 (UTC)
Yes, this is something that I would only want to enable against non-adaptive targeting, so only if the enemy hitrate is < 4% or so. --Skilgannon 12:59, 13 April 2009 (UTC)
I definitely tried this in my earliest attempts at surfing, and it didn't work out. But I was quite the newbie at that point, and not the RoboRumble champ, so I'm curious to hear if you find some benefit from it. :-) It does seem like you should clearly be able to pull some value from this data. --Voidious 17:21, 13 April 2009 (UTC)
I tried this and as all mentioned, it works against simple targeting but is very bad against anything that learns. What I ended up doing in YersiniaPestis is try to figure the probability of the enemy firing HOT/Linear/Circular and use those as factors to dodge weighted accordingly, it worked much better than the safe spots. Against GF it quickly stops using this factors, and against the simple ones it was better than the safe spots because it doesn't need to learn across all segments. --zyx 17:39, 13 April 2009 (UTC)