1 2
19 20 21
Active player (422)
Joined: 9/7/2007
Posts: 329
Lag after the boss seems to based on what the boss is doing (whether it adds extra sprites is likely), but the variance of lag of a few frames is not enough to bother worrying about too much. I suppose you could try delaying the last hit by a frame or 2 to see what happens. It looks like screen transition lag is always the same, unless there are extra sprites on the screen which would cause lag (and before). But even then I doubt trying to reduce that lag would be worth it since it would also only add a few frames of lag. There is a lot of random lag in the stages. I doubt that can be reduced without losing time, unlike the simulations. :) So I will focus on finding better boss patterns. Is the Bloodpool Act 2 boss the first one with randomness? Also, should I bother trying to find a better pattern for it? The fight looks really optimal.
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
The Bloodpool Act 2 boss (apparently called Zeppelin Wolf) is indeed the first boss where randomness is important. The thing about that boss is I would be able to get a better time if I could get it to show up on a lower platform the second time it materializes. In order to be able to take a hit against the boss I had to kill a goblin way back in the room where you dodge spikes instead of just getting hit by the goblin. That lost me 16 frames if I recall correctly. If Zeppelin Wolf showed up on a lower platform getting hit wouldn't be necessary for getting a reasonable time. Also, I would be able to start hitting Zeppelin Wolf quicker. Therefore if you could find a way to get Zeppelin Wolf to show up on a lower platform the second time around without a significant loss in time (greater than about 60 frames?) I would try to hex it in. I suppose it would be OK if it showed up on the highest platform the first time. What's important is the second time.
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
Here's Aitos Act 1: http://www.mediafire.com/?nlo1l6w10u0d58z An improvement of 43 frames. EDIT: And here's Aitos Act 2: http://www.mediafire.com/?3ta2c2dpwah3x8z An improvement of 118 frames.
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
Marana Acts 1 and 2 completed: http://www.mediafire.com/?dm71xscdv1bvqbo 121 frames saved in Act 1 and 126 frames saved in Act 2. Unfortunately, I got the same pattern for Kalia as I did last time. I tried varying the wait time before entering the boss chamber but the other pattern I got was no better. Dunnius, if you're investigating boss patterns, focus on trying to manipulate Kalia's pattern. Getting a better pattern for Kalia would save more time than getting a better pattern for Zeppelin Wolf. Hotarubi, you had a good attack pattern for Kalia (the boss in Marana Act 2) in your TAS. Was that good pattern luck or did you do something to make it that way?
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
Not sure what to do to improve luck against Kalia, so I'm moving on for now. Here's Northwall Act 1, 195 frames saved: http://www.mediafire.com/?ayy79p7s0lcflq7
Sir_VG
He/Him
Player (39)
Joined: 10/9/2004
Posts: 1911
Location: Floating Tower
Damn, you're going to town! Good job. :)
Taking over the world, one game at a time. Currently TASing: Nothing
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
OK, here's a candidate for submission: http://www.mediafire.com/?qm893ua05b66c0y 233 frames saved in Northwall Act 2, 86 frames saved in Death Heim. Total number of frames saved is 1279 frames, or almost 21 and 1/3 seconds. For those of you who want some idea of the difference in lag between my run and Hotarubi's run, the difference in lag by the end of our runs according to the lag counter is 386 frames, or about 6 and 5/12 seconds, most of that derived from increased transition times due to the change in emulator version.
Active player (422)
Joined: 9/7/2007
Posts: 329
I researched Zeppelin Wolf and I think I discovered how his pattern works; and it could be the same for all bosses. There is a 2 frame rule on the door, so everything is in twos. He alternates between 2 patterns, switching off every 2 frames. After 128 times (every 256 frames) he alternates between 2 other patterns for the next 256 frames. This explains why it is difficult to get bosses to change patterns; there is a 256 frame "boss pattern frame rule". Delaying the entry to the boss door by 20 frames (first fadeout at frame 17432 instead of 17412) causes Zeppelin Wolf to start on the top platform, then go to the bottom platform. Hopefully that pattern will work better. If it is possible to use some of those 20 frames to avoid an additional damage, it would save some time after the boss battle. (or during) I noticed that in the room with the disappearing platforms, you don't jump next to the spikes due to bad timing luck. Do you get better luck by taking damage from the goblin in the room that goes dark? Even if you can't save time with the trick, you can still jump next to the spikes and wait there because it would look awesome.
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
Thanks for your findings on Zeppelin Wolf. Unfortunately, that doesn't really help in this case. What would really save time is a pattern where both of the phases are on either the first platform or the second. If one of the necessary phases is on the top platform, as in this case, then a hit will still be required regardless of whether it's the first phase or the second. Since waiting 20 frames gives me the option of having the last phase be one where the enemy is on the first platform instead of the top platform, I tried it but it actually gave me a worse time. I'll see if your findings are any help against Kalia or Arctic Wyvern. EDIT: unfortunately, Kalia doesn't seem to alternate patterns like Zeppelin Wolf does. I don't have to wait too long to get another pattern but that one isn't good either. I have to wait a long time before getting a third pattern, too long for it to be helpful.
Post subject: actraiser: youtube + video file
Joined: 3/18/2006
Posts: 971
Location: Great Britain
youtube http://www.youtube.com/watch?v=G1bu43f8GNw video file http://www.wupload.com/file/128212966/actraiser_fin_k.avi http://www.fileserve.com/file/sFrB8UA/actraiser_fin_k.avi
zidanax wrote:
OK, here's a candidate for submission: http://www.mediafire.com/?qm893ua05b66c0y 233 frames saved in Northwall Act 2, 86 frames saved in Death Heim. Total number of frames saved is 1279 frames, or almost 21 and 1/3 seconds. For those of you who want some idea of the difference in lag between my run and Hotarubi's run, the difference in lag by the end of our runs according to the lag counter is 386 frames, or about 6 and 5/12 seconds, most of that derived from increased transition times due to the change in emulator version.
Active player (422)
Joined: 9/7/2007
Posts: 329
Kalia uses the same pattern for 256 frames (128 times), then switches patterns. The next pattern occurs by waiting 16 frames, but it is unlikely to save frames. Arctic Wyvern also uses the same pattern for 256 frames (128 times), then switches patterns. The next pattern occurs by waiting 216 frames. Of course, by being 40 frames faster, you would also encounter a different pattern. Death Heim's Zeppelin Wolf has the same patterns if you enter his room sooner. You should defeat Minotaurus as quickly as possible because there is much better luck this time. You have at least 100 frames before the previous set of patterns. Hopefully this won't mess up the luck on later bosses. I also noticed that there is a 2 frame rule on Minotaurus dying. Weird.
Active player (499)
Joined: 6/9/2010
Posts: 12
Location: Japan
I can't come to see easily because I was busy so I was surprised because your work was very advanced. I was making some improvements. snes9x1.53 filmoa 1 http://www.mediafire.com/?xc4civ8tqiekirx AIOS 1 boss http://www.mediafire.com/?1q7k30wzqsdyekb
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
http://www.mediafire.com/?claawry3j3wapw1 Hotarubi, I took your new version of Filmoa and looked to see if I could find any improvements. I managed to save some time, although it's difficult to compare to your version due to the slightly different amounts of lag in Snes9x 1.52 and 1.53. The biggest timesaver is that when I was experimenting with the boss I unintentionally found a good pattern that is 10 frames faster but takes the same amount of damage. This new version is 49 frames faster than my previous version.
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
Here's another version through to the end: http://www.mediafire.com/?519bg1dbqh4udwq This version incorporates Hotarubi's better Aitos Act 1 boss fight along with the revised Filmoa act 1 posted above. I also found a better way in Bloodpool Act 2 to bypass the goblin in the dark room. The lag in Casandra Act 1 is much less this time. The rest except for bosses is pretty much the same because I hex edited in a lot of what was in the previous version, redoing a few parts to account for randomness. This time I had very good luck in boss fights so I saved a ton of time as a result. Overall, I saved 509 frames, or almost 8.5 seconds, over my previous submission candidate.
Sir_VG
He/Him
Player (39)
Joined: 10/9/2004
Posts: 1911
Location: Floating Tower
Man, you're awesome man. I was looking over v3 and SNESMaps and thought of a couple of crazy idea. I'm sure they won't work or will negate something later, but I thought I'd throw it out there. Bloodpool Act I - Use the Y quick fall accel from the bridge just before the partial apple. Marana Act I - I assume the minor pause is to allow the 1st snake to fall enough that you miss it? Also, would a Y quick fall accel allow you to save a few frames from the ledge where the 1-UP is? Or would you lose it from the extra damage you'd take? Northwall Act II - Could you get the Partial Apple during the first wall climb to try to save some time? Or does it not save anything? BTW, nice of Kalia to FINALLY cooperate. :)
Taking over the world, one game at a time. Currently TASing: Nothing
Active player (499)
Joined: 6/9/2010
Posts: 12
Location: Japan
The following points worried me. This one is useful at many places however, it is difficult to consider the pattern of step. Snes9x1.52 movies http://www.mediafire.com/?q7ax0y15tj20b8g MARANA ACT 1 http://www.mediafire.com/?20m0aiim7rq275r FINAL BOSS http://www.mediafire.com/?fitf6k09nn9f1io
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
In response to Sir VG: Bloodpool Act 1: Quick fall on the bridge actually loses time (you press against the bridge for several frames before moving forward again). Besides, I want that partial apple because it saves 16 frames at the end of the stage. Marana Act 1: Yes, the wait is to let the first snake fall far enough to bypass it. A rather clever idea on Hotarubi's part (positioning yourself so that only the first snake falls). I had tried quick fall before but as you suspected I lose too much time due to taking two extra hits. Northwall Act 1: This idea, on the other hand, worked, thus saving me about 20 frames. I've put up another SMV to just before Tanzra: http://www.mediafire.com/?aorrsyyx4r9o01x The reason I haven't bothered to do Tanzra again yet is because the Tanzra fight is rather hex unfriendly due to randomness and lots of lag, and I want to look at Hotarubi's stuff first. Hotarubi: thank you for the clever ideas. I will try to integrate them, and hopefully they will not affect my luck on bosses. I knew about the trick where you jump boost before getting hit because I saw it in your run, but I wasn't able to do it consistently. As you said, it is difficult to execute, but perhaps the example you just provided will help me figure out how to execute it consistently.
Sir_VG
He/Him
Player (39)
Joined: 10/9/2004
Posts: 1911
Location: Floating Tower
I'm glad one of my ideas actually worked out successfully. Thanks for the explanations. :) This run is gonna be freaking awesome when it's finally done.
Taking over the world, one game at a time. Currently TASing: Nothing
Active player (422)
Joined: 9/7/2007
Posts: 329
Bloodpool Act 2 - I'm surprised that you got lucky. There is obviously a way to manipulate the patterns further, but I don't know what it is yet. I am researching it. Marahna Act 2 - Very lucky indeed. Northwall Act 2 - Obviously with better luck you can save time. The next pattern is 202 frames later (not better), and the previous pattern is 54 frames earlier. But if I can figure out how the luck changed in Bloodpool Act 2, time could be saved here. Bloodpool Act 2 Refight - Good luck, and without having to delay on Minotaurus. Marahna Act 2 Refight - Good luck. It seems like the bosses are a bit faster in the refights because you weren't able to get as many hits in when Kalia comes down. Northwall Act 2 Refight - Again luck could be better. The next pattern is not better. Again, if I can figure out the Bloodpool Act 2 luck change, it may be possible to get a pattern where the refights patterns are good.
Sir_VG
He/Him
Player (39)
Joined: 10/9/2004
Posts: 1911
Location: Floating Tower
Marahna Act 2 Refight - Good luck. It seems like the bosses are a bit faster in the refights because you weren't able to get as many hits in when Kalia comes down.
Yes, some of the bosses move about twice as fast. This makes Kalia a bitch to fight.
Taking over the world, one game at a time. Currently TASing: Nothing
Player (215)
Joined: 2/12/2006
Posts: 373
Location: Oregon
I was afraid that incorporating Hotarubi's ideas might mess up my good luck. Unfortunately, that is indeed the case. I got a worse pattern for Kalia that nullified my savings(I've kept a SMV of that). I tried pulling out improvements from before Marana Act 1 (i.e., hexing together the old version's pre-Marana stuff and the improved Marana). That gave me good luck on Marana Act 2's Kalia, but my luck ran out in Death Heim where I got a worse pattern for that Kalia (I've also kept an SMV of that). In order to get good luck in Death Heim too I cut out Marana Act 1's improvements too, which is really too bad because that's where the most substantial improvements are. I got worse luck with the eyes in Northwall Act 1, so my improvement at the moment is a grand total of 3 frames over the version that has Sir VG's apple trick. Luck could probably be maintained with a more extensive set of improvements if someone was willing to find out which set to apply, but I don't have that much time on my hands given all the hexing and redos that would entail, and it probably wouldn't amount to more than about 2 seconds improvement unless they actually managed to make luck on the bosses even better. I think I will go ahead and redo Tanzra and leave it at that, unless someone manages to find some other trick. EDIT: Here's the next submission candidate: http://www.mediafire.com/?t2c4hqn9j6mjk63
Active player (422)
Joined: 9/7/2007
Posts: 329
How many frames saved does each group of improvements have? It should be possible to get good patterns based on what I found in my previous post. I suppose I can take a look at it to see if I will be able to figure it out.
Joined: 11/27/2018
Posts: 5
So I started up a new TAS after a dilemma in the speedrunning route between going for a 38 or 39 cycle strategy. They both have upsides and downsides. I'm only a spectator of sYn_stream on Twitch so I don't actually speedrun the game, and my understanding of the sim sections was lousy so I wanted to TAS it to understand it. I ended up spending 40-70 hours on the first action segment though. Much of it was spent simply learning the ropes and seeing if I can further understand the AI by inspecting the RAM. I kinda wanted to get it over with, but there were many obvious time saves to me despite the 400iq needed just to fucking jump around thanks to the fact that every jumpboost you do has an effect on your jumpboost windows for every single jumpboost thereafter, causing immense amounts of permutations and making certain maneuvers possible or impossible. I'm doing this on bizhawk/bsnes. It has much longer loading times than the Snes9x 1.51 TAS, probably due to emulation accuracy. I lose 41 frames before level 1 even begins, however the action segments suffer no differences so far. Here are my current time saves on Fillmore Act 1. video: https://youtu.be/MVlSyWPyC2g -2 frames On the very 2nd jump, I fit in a full 4pixel jumpboost -2 frames On the enemy immediately following the above jump, I damage boost him as opposed to jumping over, because there isn't enough space to do a 4pixel jumpboost. An ordinary unboosted jump will cost you 4 frames of crouching idle upon landing. A 4pixel jumpboost will push you forward the equivalent of 2 frames of running, thus halving the cost of jumping, down to 2 frames. A damage boost will cause 2 frames of idling during the animation, thus it is equal to a 4pixel jumpboost. -9 frames I precisely damage boost the gorilla to be launched onto the branch. There's only a two-frame window to execute this, but it's a major frame save, as opposed to the slow jump over his head from the 2011 TAS -x frames Due to oddities in hit registration, I take damage from the hornet on the hopping tree a frame earlier than I would if I was going a frame slower. That is to say, despite the fact that the hornet has been spawned for the same amount of time, and despite the fact that all of his RAM addresses are 100% identical, and the fact that we meet at the same exact coordinates, we register impact 2 pixels sooner than we would if I came into the area on a different interval. This fact doesn't slow me down, but it does shuffle my jumpboost opportunity around and makes me have to jump a frame earlier to get a 4 pixel jumpboost, meaning I land 2 pixels less far than the other TAS. This was a problem because I couldn't 4pixel jumpboost onto the following stump in a position that let me complete the following section as fast as the old TAS, no matter how many permutations I tried, and I always lost a single pixel because I couldn't get the 1 frame of falling accel boost onto the 2nd hopping trees hand. However, I was able to jump off of the tree while it was a few pixels higher in the air, letting me fall 1 pixel further, thus making up for the lost pixel described above. -2 frames I land next to and damage boost the brambles -12 frames While falling into the bosses room, I shorten my distance by 2 pixels to offset myself by 1 frame due to a hit registration issue in the middle of the fight. If I don't offset everything by 1 frame, I'll get hit on the frame one of my attacks is supposed to come out. It was very difficult to find new ways to get hits in. At one point, I turn around and do a back slash to evade taking damage from a mysterious hitbox that immediately disappears, which is also why I have to turn back around to keep attacking. This would be a 13 frame save, but for reasons unknown, I get an extra lag frame loading the boss, as opposed to my bizhawk reconstruction of zidanax's TAS. all in all this is the first serious TAS I've attempted. I quite like it. edit: I've overhauled the 2nd paragraph, the explanation of the first -2 frames, and the 1st paragraph of the -x frames explanation
Active player (422)
Joined: 9/7/2007
Posts: 329
One thing to keep in mind when comparing runs is the difference in the amount of HP after beating the boss. Because the health bar refills, the difference of HP makes a very small difference in the number of frames. Of course damage boosting has priority, but damage on a boss should be avoided if it doesn't save time over the refill time. I looked through the TAS resource to refresh my memory on things, and I remember why I never got around to this. The movement on the Action is a bit complicated. I made a lua script for the jump and crest boosts to know when it is optimal to jump or use the height boost, which I can find if you need it, unless you made one. But yes, the planning for chain of optimal boosts kept me away from TASing immediately, so I am glad you are taking this up. As I had mentioned, the plans I have for the simulations involve reducing the amount of time for the Town Under Construction cycle. I noticed while working on Kasandora that the pointer movement affects where things are built, and thusly the time it takes. I didn't have time to mess around too much with that back then, but now that I am familiar with bot testing it is a big focus. I need to watch the TAS again to remember what other ideas for improvements I had. My bot testing will take a while, so I think you should work on the JP Special Mode first. You won't have to worry about Stardust in that mode! Plus, it will make it easy to redo parts if you discover something later on. Having to redo the simulation parts is a bit annoying.
Joined: 11/27/2018
Posts: 5
dunnius wrote:
One thing to keep in mind when comparing runs is the difference in the amount of HP after beating the boss. Because the health bar refills, the difference of HP makes a very small difference in the number of frames. Of course damage boosting has priority, but damage on a boss should be avoided if it doesn't save time over the refill time.
I did test this, and the health refill had technically cost 2-4 frames in the ending sequence, however, it still took the same amount of frames for me to load into the sim as the version with 1 more health. Basically it delayed the sequence, but the delay took up part of the loading screen and evened out. I want to test again.
dunnius wrote:
I looked through the TAS resource to refresh my memory on things, and I remember why I never got around to this. The movement on the Action is a bit complicated. I made a lua script for the jump and crest boosts to know when it is optimal to jump or use the height boost, which I can find if you need it, unless you made one. But yes, the planning for chain of optimal boosts kept me away from TASing immediately, so I am glad you are taking this up.
Currently I can use RAM Watch to read the memory addresses of those two values, unless you mean something more than that.
dunnius wrote:
I noticed while working on Kasandora that the pointer movement affects where things are built, and thusly the time it takes.
Could you elaborate on this? I can't tell if you're talking about a level of detail beyond what it sounds like.
dunnius wrote:
My bot testing will take a while, so I think you should work on the JP Special Mode first. You won't have to worry about Stardust in that mode! Plus, it will make it easy to redo parts if you discover something later on. Having to redo the simulation parts is a bit annoying.
I must admit I have slightly more interest in the sim than the action. That's what brought me here. My next step is to work on the Fillmore sim if only to familiarize myself with how sims truly work, and then decide from there what is next. I would not be able to tolerate only doing the 12 action stages. It sounds miserable, especially since I use the time consuming strategy of not even referencing the old TAS until after I finish the level. I feel that strategy is responsible for most of my ideas, improvements, and fun.
1 2
19 20 21