DoDonPachi Dai-Ou-Jou Black Label "VERY HARD" TAS

xy2_ · Post by **xy2_** » Sat Jan 30, 2016 5:38 pm

Hello! Apologies if this is the wrong section to put this in. I'm currently making a DoDonPachi Dai-Ou-Jou Black Label TAS on the VERY HARD difficulty, and would like some feedback. This took me 3 weeks to TAS the very start of 1-1, and I'm looking for some feedback atm.

https://www.youtube.com/watch?v=JakvcJdzI0I

Movie file (play back with FBA-RR v.0.0.7)

Strikers1945guy · Post by **Strikers1945guy** » Sat Jan 30, 2016 5:43 pm

Very hard as in going into the menu and changing the difficulty? I thought that didn't do too much in DOJ since it barely even has rank unless you're playing at a god like level? Does adjusting the difficulty in DOJ make the patterns harder or just make the bullets move slightly faster?

CStarFlare · Post by **CStarFlare** » Sat Jan 30, 2016 5:52 pm

What is the purpose of this TAS? Score? Weave through as many bullets as possible?

If it's score I think you need to spend some time with HFD's videos and do things like collect bees as late as possible. If it's for show then I think it's going to be hard to pull that off without just making a video that looks like it's shot with your hitbox turned off.

SMC · Post by **SMC** » Sat Jan 30, 2016 6:12 pm

Look up the base values of enemies and kill the highest value ones as early in the chain as possible.
Collect bees on the very last possible frame before they leave the screen. There shouldn't be more than the minimum number of enemies required to continue the chain left on screen when you collect a bee.
Get as much mileage as possible out of your hypers. Stall a lot so that your hyper period covers the largest number of enemies it possibly can.
Point blank enemies during their initial invulnerability period as much as inhumanly possible.
Point blank hyper bosses constantly unless it's literally impossible.

xy2_ · Post by **xy2_** » Sat Jan 30, 2016 7:20 pm

Strikers1945guy wrote:Very hard as in going into the menu and changing the difficulty?

Yeah. VERY HARD changes a little bit in regard to difficulty, with normal enemies shooting much more than in NORMAL and shot speed being faster (?). Don't know if boss patterns change currently.

CStarFlare wrote:What is the purpose of this TAS? Score? Weave through as many bullets as possible?

If it's score I think you need to spend some time with HFD's videos and do things like collect bees as late as possible. If it's for show then I think it's going to be hard to pull that off without just making a video that looks like it's shot with your hitbox turned off.

As a TAS, the primary goal I aim for is score and the secondary one is entertainement; in short, score while trying to be entertaining/impressive. I've looked in-depth into HFD's superplay, and use it as a base to beat.

SMC wrote:Look up the base values of enemies and kill the highest value ones as early in the chain as possible.
Collect bees on the very last possible frame before they leave the screen. There shouldn't be more than the minimum number of enemies required to continue the chain left on screen when you collect a bee.
Get as much mileage as possible out of your hypers. Stall a lot so that your hyper period covers the largest number of enemies it possibly can.
Point blank enemies during their initial invulnerability period as much as inhumanly possible.
Point blank hyper bosses constantly unless it's literally impossible.

The first one I didn't know about, thought it was the other way around: preserve the higher point enemies for as late in the chain, which I do as much as possible in the current run. I'll be redoing the run, then.

For bees, the bee collecting pattern I use is fully optimised, and I've looked in advance into making myself TAS clearing patterns such as to optimise bee collection, collecting them as late as possible while currently not in an hyper. This is very tricky to route, and while I've tried to separate from HFD's route as much as possible (getting a few more hits than his route along the way) his route is really good.

Getting as much out of your hypers I know about.

The point blanking for enemies and bosses I also know about: the behavior regarding invicibility and hyper shots is very wierd; you are able to hit them with one shot before switching to hyper laser and grinding more hits out of them while in invicibility, for usually a one hit gain. Point blanking large enemies with hyper is also very wierd in the way that they can give you hits around 25 frames after completely dying if you position yourself properly between the two hitboxes of the large enemy you're trying to grind hits off; I have no idea why this happens.

Shepardus · Post by **Shepardus** » Sat Jan 30, 2016 7:58 pm

The way the chain works is every enemy you kill adds the values of all the previous enemies in the chain to your score. Kill more valuable enemies first so their value gets added more times.

I think which weapon you use to kill an enemy also affects how much hyper you get out of it.

Look into the "deadzoning" technique where players sometimes let an enemy almost drift off the bottom of the screen, then hyper and sit on top of it while it's invulnerable (in the dead zone) to get some more hits. HFD does this on stage 2 after the midboss, and at one of the stage 5 midbosses (the one based off the DDP stage 3 boss), but there may be more opportunities for this in a TAS that would be impractical for a human player. SPS also uses this technique to chain to the 1-3 midboss in White Label, and I've also seen this done on one of the large tanks near the end of stage 5 that HFD doesn't do, if I remember correctly.

rjosal · Post by **rjosal** » Sat Jan 30, 2016 9:05 pm

About the deadzone/aura technique, it's the same as grinding the stage 4 boss's gears. The more enemies that are blue under your aura, the more crazy hits and hyper value you earn. I expect this can be abused in many places.

I bet there are some more optimal hypering strategies. It could be that picking up a bee the instant it's on screen when you're off hyper turns out better than hitting it at the bottom when you are in hyper.

I'm stoked to see what this ends up looking like.

Erppo · Post by **Erppo** » Sat Jan 30, 2016 9:13 pm

rjosal wrote:It could be that picking up a bee the instant it's on screen when you're off hyper turns out better than hitting it at the bottom when you are in hyper.

It most definitely is. Charge is always worth more than couple extra bee points.

But yeah, the biggest improvements you can do in a TAS probably come from being able to aura-abuse way more enemies. Especially enemies invulnerable when entering the screen, HFD doesn't really pay much attention to that after 1-1. Kill order optimizations too but I doubt they will ever change the score much.

Also I fail to see the point of using very hard. That way you're not directly comparable to the human records and the difference is really minor anyway.

xy2_ · Post by **xy2_** » Sat Jan 30, 2016 9:30 pm

rjosal wrote:It could be that picking up a bee the instant it's on screen when you're off hyper turns out better than hitting it at the bottom when you are in hyper.

Picking up bees without an hyper raises your hyper gauge significantly (based on your number of hits); it's worth it in all cases, as a huge boost to charge towards another hyper which can lend a very significant amount of points (especially so considering the hyper abuse I'm gonna be doing in the run) is huge compared to 10 or even 100 extra hits on a bee pickup.

Erppo wrote:Also I fail to see the point of using very hard. That way you're not directly comparable to the human records and the difference is really minor anyway.

I chose to use VERY HARD to make for more entertainement, as the situations you're in can be tighter. Other than that, there are absolutely no differences scoring wise, and the extra difficulty VERY HARD provides doens't really matter in a TAS, due to the fact that it's a TAS; technically, it would still be comparable to an human run.

rjosal · Post by **rjosal** » Sat Jan 30, 2016 10:39 pm

@Errpo, @xy2_, every case for sure? Because it doesn't have to be that way either. It will affect your overall hyper flow which may cause you to be in hyper while you pick up 2 bees later. I'm mostly a WL player so maybe this isn't much of an issue. I just think about how if I drop a chain in a spot where I can reach 200 hits again before the next bee, it doesn't affect my hyper flow, but if I can't, it will and it might cause me future chain breaks and really mess up my run.

Another thing to make sure to do: get all 3 ships at the beginning of stage 3. I think there might be multiple ships the same way before the stage 3 mid boss, but I've never tried cuz that's crazy.

xy2_ · Post by **xy2_** » Tue Feb 09, 2016 9:51 pm

https://www.youtube.com/watch?v=IvNieBHu4eo

Movie file. 15497 rerecords. For reference, a rerecord is made each time you load a savestate.

I've made some progress and redid the segment to a level I'm somewhat satisfied with. In the first part of this section, nothing much has changed; some score optimisations that are not terribly interesting. Since there is a lot of downtime TAS-wise in these sections, I tried to cram in some entertainement.

The second part is where real stuff goes down. As soon as I unlock the top-left bee, the game becomes a desesperate race of getting the hyper to grind as many hits off the tower as possible. From there, the route becomes complex; you effectively have to juggle between scoring on the popcorn enemies right before the hyper, without actually destroying the flying popcorn you grind hits off on the turret, all while trying to kill all enemies.

The solution I came up with looks a little sloppy, but it's the best by 1 hit and a bit of score: grind on the tower as hard as possible. This makes it so three tanks are left behind. Once I destroy the 8 rotating.. discs? around the tower, I start making my way back to the tanks I left behind earlier, destroying one and making two out of bounds tricks of varying lengths on the two left. While this makes me lose some hits due to not destroying the thanks and having a slightly lower score, the extra hit make up for it, and I end up with 451 hits instead of 450 at the end of this section.

A little trick I use at the end of the miniboss; by keeping it at 1 hitpoint for a few extra frames, I can delay its time of death without losing any score nor chain, and this way wait for a pattern to be fired to gain some extra points.

youtube · Post by **Softdrink 117** » Wed Feb 10, 2016 2:15 am

This is an interesting project. In my mind, the resulting TAS is less important than the actual research that will come from the process of doing it (though I'm sure the TAS video will be pretty interesting too!) -- the routing and optimizations in a game that is already very well optimized should be pretty interesting to see. I hope you continue to document this thoroughly throughout the process; this is pretty cool stuff so far!

rjosal · Post by **rjosal** » Wed Feb 10, 2016 5:09 pm

A general comment on the entertainment aspect - With a TAS of course you can swim through bullets the whole run long, and so to me I lose interest in that aspect. Part of practicing a shmup is optimizing risk. It'd be cool to see a TAS which gets the optimal maximum high score at all costs, but also minimizes risk as a secondary goal. I bet going through the game with such a fine tooth comb as you are, you may be able to find some reasonable risk score optimizations that normal players should be able to do, but that would be hidden if you're just swimming through bullets on purpose. Maybe make bullet proximity a reverse goal to reduce risk. And safe spots. Tricking enemies and bosses to fire away from you.

But I get it if it's a goal of the TAS to swim through bullets like that. If it is, you may want to make some obvious abuse of the every-other-frame collision detection. I noticed that on that stage 1 tower it fires some fast downward bullets and the center one is aimed directly at your ship. You can fly directly up straight through it every time if you want.

xy2_ · Post by **xy2_** » Wed Feb 10, 2016 7:46 pm

rjosal wrote:A general comment on the entertainment aspect - With a TAS of course you can swim through bullets the whole run long, and so to me I lose interest in that aspect. Part of practicing a shmup is optimizing risk. It'd be cool to see a TAS which gets the optimal maximum high score at all costs, but also minimizes risk as a secondary goal. I bet going through the game with such a fine tooth comb as you are, you may be able to find some reasonable risk score optimizations that normal players should be able to do, but that would be hidden if you're just swimming through bullets on purpose. Maybe make bullet proximity a reverse goal to reduce risk. And safe spots. Tricking enemies and bosses to fire away from you.

But I get it if it's a goal of the TAS to swim through bullets like that. If it is, you may want to make some obvious abuse of the every-other-frame collision detection. I noticed that on that stage 1 tower it fires some fast downward bullets and the center one is aimed directly at your ship. You can fly directly up straight through it every time if you want.

TASes are meant to showcase what a theoritical superhuman could do - and that would include "dodging" bullets at incredible reflexes. But your argument makes a good point: while this is the start and the game bullet management is something that I've overlooked because it can waste precious frames that could be used doing score management. And playing around with bullets while waiting can be not terribly interesting at times. This is something I'll try to look into much more than what I've been currently, because it is possible to alter situations without losing score.

xy2_ · Post by **xy2_** » Mon Mar 14, 2016 6:47 pm

https://www.youtube.com/watch?v=g3xpUAQ ... e=youtu.be

Movie file.
22005 rerecords. For reference, a rerecord is made each time you load a state.

Small update on this..

HFD | 8 478 920 score, 587 hits
XY | 8 524 990 score, 599 hits
Score at powerup ship.

This TAS is put indefinitely on hold until an emulator with rerecording features exists (BizHawk is on that path with importing a MAME core.) I've also made an embarrassing scoring mistake at the start, so that's why I'm planning to restart this anyway. This will probably delay it by a few months at best, unfortunately.

1-1 is much more irritating to TAS than I initially thought, for a few reasons that I'll outline, but the main one is definitely the hypers. In this first half of the stage, most of the scoring opportunities are present while hypering. This already leads up to some roadblocks, that I'll call 'limits'; these block completely any further scoring until you pass that limits. Hard limits are impossible to get through, and are represented in the form of bees and hyper timeouts; since they control hypers, it means that scoring is forcibly limited to these hard limits. The other type of limits I will talk about are 'soft' limits: these are limits that can be extended with skillful play, and are mostly chaining related (how much enemies do I need to leave to continue my chain?).

And this is where TASing comes in! While normal scoring segments can't really be improved upon (although they can with a particular glitch I'll talk about later), these soft limits can be extended almost infinitely, both through the use of precise play and glitches. In fact, the only soft limit I meet in this segment is fully extended to its corresponding hard limit (hyper timeout). I'll mostly talk about how I managed to break these soft limits here, as they are the main limiter of scoring gameplay.

The other huge hassle was the tower miniboss. The problem with the tower miniboss is that it structures all the future hard limits we come across, because we're dependant on it to grind a bunch of hits on it thanks to our first hyper, and so the chain of hard limits gets structured as this:

First hyper (tower) > Timeout > Second hyper > Timeout.

We have very limited control over when hypers start, and because of their fixed duration there's not a whole lot of breaking new strategies you can do (especially in 1-1), so my route for this segment is akin to HFD's route and the hard limits are the same. Given that, there's is only one soft limit in this segment: at the moment of the first hyper's timeout, just before the second powerup ship. In scoring, you would need to leave some enemies unscratched by the hyper, even though it would be possible to hit them and get more hits + score. This is the soft limit I was referring to earlier. So how do I get past it?

Before I continue, I first need to explain how big ships work. Take for example the medium-sized turret ship, after the turret miniboss. One particularity of these ships, found in all big ships is that they have two hitboxes: one at the front, smaller and in the middle, and one in the back and wider. This interesting quirck makes for some interesting results that allow for more scoring, thanks to the hyper's behavior. First, we know that the laser is actually two weapons: laser + aura. This would not make a single difference if we weren't under hyper. Furthermore, there is a failsafe present to avoid hitting the two hitboxes with the same weapon; it will simply not register the hit. But what happens if we hit the two hitboxes with two separate weapons? There isn't a failsafe for that, and so we get more hits. However, this is extremely precise to do, as if any of the weapons touch the hitbox that doens't belong to them, the hit is cancelled. I refer to this as 'diagonal grinding', and is also the source of the wierd movement you see when I laser enemies.

Going back to our soft limit, the way I handle it is give zero fucks in regards to that limit and extend all the way to the hard limit. This results in a big problem; we only have a single ship, the powerup ship, to somehow keep the hit counter up for a ridiculous amount of time. We know that hitting a big ship with laser will set our hit gauge to a fixed height. What I do is what I call hooking. Because of the gauge set mechanic with hitting a big ship with the laser, if we hit the powerup ship for one frame, then hook off him, our gauge still get sets, then lowers slowly. At the last possible frame, simply hit the powerup ship again to get the gauge set, and repeat! Because the gauge reset with lasering lasts exactly 9 frames, we are able to hit the powerup ship once every 8 frames instead of once every other frame, essentially making the powerup ship last four times as long and keeping our chain easily.

All of this had to be done by hand, since FBA doens't have ram search, any precise values and doens't work at all in general. I believe the switch is the best decision I can take, so you'ill have to wait a few months on the run.

rjosal · Post by **rjosal** » Tue Mar 15, 2016 4:46 am

That is totally awesome, and what a great write up of your discoveries! Really good read. What you did with the second powerup ship is probably something that will come in handy many times. Human players have to do it while hypering after the stage 4 1-up on a railgun to keep the chain alive. The part about getting extra hits off a big ship with 2 hit boxes seems like a pretty huge discovery. I have no doubt that when you finish the stage you will have the absolute maximum possible score. Too bad you have to take a break.

Patashu · Post by **Patashu** » Wed Mar 16, 2016 7:52 am

The laser grinding trick is really cool. I'm glad to hear Bizhawk is finally integrating MAME (I guess it was waiting for it to become FOSS?) and I hope that the TAS in Bizhawk with rerecording will be as amazing as you can dream!

xy2_ · Post by **xy2_** » Mon Jun 20, 2016 5:15 pm

Posting here to let you know this TAS will rise from its grave soon.. The main reason for this is that a non-garbage rerecording emulator won't come for probably several years, so as it is currently I might as well do whatever I can.

Since this WIP, there have been a good deal of interesting strategies that our own moozooh suggested to me:

- Use the laser grinding trick on large enemies. The laser grinding (in case it wasn't obvious already) breaks the game pretty much everywhere: BL's chaining is normally very tight, but the ability to make large enemies last for four times as long will allow for a lot of cool alternate TAS-only strategies.
- Do not pick up any powerups. This might come as a surprise, and one which would be out of the question normally, but if you look at DOJ's mechanics it makes a lot of sense. First, hyper has a fixed damage output regardless of shot power or ship type. This makes being on low power a lot less risky, as you still keep the same damage. However, the real advantage comes because of dealing less damage: laser grinding on large enemies increments at a fixed rate (1 hit every 8 frames!) This means having less damage is actually more optimal in this case, because you will gain more points from the fixed tick rate, gain more hits and overall more points. One might argue that the low power might backstab at more chaining heavy parts, however I don't think the damage output difference will be critical enough to deter scoring negatively. Remember that bullets don't matter in a TAS, so we're only looking at how long enemies stay on screen, rather than 'being overwhelmed' by enemies. This also comes in handy for abusing aura: the tick rate is the same, so exploits are still possible and more advantageous (including double hits on enemy with laser+aura, getting hits after enemy death, abusing out of bounds enemies)

Van_Artic is working on a testrun with the no powerup principle, and I will make a quick TAS testrun in the following days to test my theory. If you have any other suggestions, be sure to list them here, too.

rjosal · Post by **rjosal** » Tue Jun 21, 2016 4:55 am

Makes sense with the no powerup. You can get an even weaker laser if you use shotia. It might affect your hyper placement like in the thick areas of stage 5, but that's probably where you want to be hypering anyway. With your discoveries so far it seems like the game is ripe for a TAS. Looking forward to it.

ilitirit · Post by **ilitirit** » Wed Jun 22, 2016 12:14 pm

Doesn't mame/mess-rr support DOJ?

http://tasvideos.org/EmulatorResources/Mamerr.html

Or FBA-RR?

http://tasvideos.org/EmulatorResources/Fbarr.html

xy2_ · Post by **xy2_** » Wed Jun 22, 2016 8:54 pm

I'm using the latter, fba-rr to TAS. The former is broken garbage and the latter is functional garbage (albeit for different reasons), so I chose the lesser of two evils. All my movie files in fbm movies are provided along with encodes, so even in case of HDD failure (like recently) I don't lose the TAS.

xy2_ · Post by **xy2_** » Thu Jun 23, 2016 4:12 pm

rjosal wrote:You can get an even weaker laser if you use shotia.

<Van_Artic> shitta
<Van_Artic> well that pretty much sums her up
<Van_Artic> actually

I've looked into it and there's a few complications:

- Laser tick rate is lower.
- Aura range is lower - it has trouble hitting offscreen enemies.
- You move slower when lasering.
- The most detrimental - laser has no pierce, for example when killing popcorn enemies.

xy2_ · Post by **xy2_** » Sat Jun 25, 2016 6:54 pm

One of my most remarkable failures: https://www.youtube.com/watch?v=E5AoxeCvE7k

Movie file

Unfortunately, not picking up powerups seems to have some complications.. apparently, the hyper counter increases slower for the same amount of damage. Even with the 12 extra hits from the bee compared to my older WIP, I still can't break through.

xy2_ · Post by **xy2_** » Fri Jul 22, 2016 7:33 pm

Posting to remind I'm not dead yet. Apparently my wish has been heard and a non-shitty rerecording emulator is coming out soon (in the form of FBA itself implementing rerecording) which means another redo from scratch.

I want to specify one thing: powerup strategy isn't dead, it was my mistake (I fired too many shots on the powerup ship and didn't get enough hyper, hence not being able to get the hyper at the right time.) From my observation powerup level DOES NOT affect hyper gain from laser.

moozooh · Post by **moozooh** » Thu Jul 28, 2016 9:59 pm

Instead of HFD, I suggest using this guy first loop routes as a reference. He has found improvements over HFD's chains in every stage, around 50 million in total. His stage 3 route in particular is amazing and worth 22 million more than HFD's.

xy2_ · Post by **xy2_** » Thu Nov 03, 2016 6:47 pm

This TAS is put indefinitely on hold until an emulator with rerecording features exists

After a long wait, for the first time in years, we finally have a competent arcade TASing emulator: FBA! No more dealing with the broken and terribly outdated FBA-rr anymore, and most importantly, correct sound emulation in DOJ!!!!

Added re-recording functionality [blip, dink]

trap15 · Post by **trap15** » Thu Nov 03, 2016 7:12 pm

xy2_ wrote:After a long wait, for the first time in years, we finally have a competent arcade TASing emulator: FBA! No more dealing with the broken and terribly outdated FBA-rr anymore

Heeeeeellllllllllllllllllllll yeah

xy2_ · Post by **xy2_** » Sun Feb 19, 2017 6:57 pm

This post is more of a dumb bump than content: feel free to ignore.

So why hasn't been there any activity from me in several months with no apparent reason? The reason is pretty simple: I go to college. Unfortunately this gives me close to no time to do anything else. I expect to get the time to work on TASes when school ends (around June), which will give me two months to hopefully restart the Zing Zing Zip TAS (I had two hard drives failures..) and test out some ideas and some things I've discovered with DOJ.

In terms of emulators, feos has also worked on getting MAME-rr, a fork of MAME for rerecording, first usable, then up to date. Check it out at https://github.com/TASVideos/mame-rr.

shmups.system11.org