Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
zeromus wrote:
As if things arent confusing enough, he tells us all about his insane workarounds and then tells us that things seem fixed anyway.
Sorry for not clarifying.
The first translation about the workaround was related to this post where he used the EmuCR build.
The second translation about BizHawk 1.9.3 was from his last post. It indicates that there was no problem with disk change in BizHawk 1.9.3 (thus, any problem with disk change lies with the EmuCR build).
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
ALAKTORN wrote:
ディスク交換はいったんメモリカードセーブしepsxでロードして行い
そのデータをセーブしたものを
Bizhawkで読み込みnowで作成しました
No idea what he’s saying here.
I'll try translating:
For disk change, I temporarily save to memory card, and then load in epsx, and then I load
the data that I saved into Bizhawk and record a movie starting from "now".
AmaizumiUni wrote:
Translation:
Yes, for BizHawk 1.9.3 I was able to have no problem.
There was no problem for Valkyrie Profile, Gundam 0079 (Mobile Suit Gundam).
I cannot try other games that use disk change because they are too long.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
The mob has finally beaten the game for the first time (about a day ago). Pokemon caught is now at 86 out of 151.
Other stuff:
- Safari Zone is 3000 steps instead of 500, making it way too easy to get Gold Teeth and HM Surf.
- There was a glitch that prevented Kadabra from evolving from a trade. I guess it is fixed by now.
- On Day 16, the mob started playing with the PC, and killed off 13 Pokemon. Of course, not as much drama as on Day 4, but whatever. (They have only killed off 1 Pokemon since.)
- The Mob manages to catch Zapdos with Master Ball.
- The Battle Tent glitches out again (twice more), forcing resets. Then the next time, the Battle Tent doesn't break their game, and the Mob goes on to beat the 10 trainers there. They are awarded with 999999 Pokedollars, apparently another glitch. The intended reward was apparently 30000. (Later the streamer cut their money down to about 30000.)
- The mob was trapped on a 1x1 square in Seafoam Island (reachable by Surfing) for a long time.
- The Route 22 ledge (toward Victory Road) was beaten in Anarchy at least once. Of course, they would just black out in Victory Road and have to do it again. Speaking of which, Victory Road was simply too hard to do in Anarchy (although they managed to do a couple rock puzzles successfully) so Democracy was activated.
- It took 17 attempts to beat E4 and Champion Blue, at Day 28.
- Checking under the bed triggered the Dream Red event and battle.
- Using Strength on the SS Anne Truck triggered a Mew event, leading to a mysterious cave.
At this point, reporting on TPP is much like reporting on a space probe on its way to Mars, so don't expect too many updates. The last few Pokemon will probably take them an insanely long time to catch, I think.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
For more information on ending movie, see http://tasvideos.org/MovieRules.html#MovieMustBeComplete
There are occasionally different schools of thought about when exactly to end a movie, but the common theme is that you shouldn't end the movie with blank inputs because all it does is make the final time appear to be greater than it actually is.
If you decide that there is a good reason to go through the ending and credits (say, to clear a dialog box at the end or enter a name using inputs), then by all means do it.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Spikestuff wrote:
Anyways if it was up to me, I'd have this rejected for being suboptimal.
I tried all other options just now.
As White, AI's first move is always c4 or e4.
As Black, AI's first response to e3 or e4 is always .. e5.
In other words, I see no reason to believe that the two other scenarios you listed are possible.
Also, AI's response to 1. e3 e5 2. Qh5 is always d5.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
ALAKTORN wrote:
“ Uses speed/entertainment tradeoffs ”
Does that still apply? Where?
The blackout glitch is not used. This glitch is where, after beating a boss, if you press start on the frame where X starts walking toward center, the screen blacks out. It saves about 20 frames (by not having fadeout) but adversely affects the music by having it play at soft volume (like when in the submenu) until you open the submenu again.
For this reason, Hetfield90 does not use the glitch in the TAS.
“There are about 1000 values out of the 43534”
I’m actually confused by this: why are there 43534 values total? If the RNG is 2 bytes, shouldn’t there be 65536?
Better RNGs (such as linear congruential generator and linear feedback shift register) would have periods which are powers of two, or very nearby. Any so-called RNG that was put together in an hour has no such harmony (i.e. it may miss out on whatever number of values).
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
By the way, after testing more door phantom grabs, I now understand by what parrot14green means by the distance between phantom grab door and normal door.
If the door is first produced in the boomerang (not when X touches it) too far to the left, the boss won't spawn.
So, this rules out any bosses where the door can only be approached by horizontal scrolling, which are Octopus and Mammoth.
For Kuwanger's case, if you phantom grab the door so that it scrolls very little and you can see the ladder, then Kuwanger won't spawn. I'm pretty sure now Hetfield90 has the best strategy.
For Penguin's case, you'd have to try phantom grab from below. But this is the farthest right I've been able to get the door. Even if you were somehow able to escape from the wall, Penguin still won't spawn. I tested by teleporting X into the room.
Example: http://tasvideos.org/userfiles/info/21001241970948079
For Chameleon's case, you'd be very far away once you phantom grab from below.
Example: http://tasvideos.org/userfiles/info/21001253808618399
I don't think it is possible to ice sled zip in Mandrill and then do boomerang phantom grab. You'd have to be in the wall to load the gate and you can't stop while in the wall.
I've stopped testing, so if parrot14green doesn't come up with anything, I don't think any skip glitch will appear any time soon.
ALAKTORN wrote:
Huh, the optimal pattern is that rare is it. How do you keep track of what value in the sequence you are in the RNG? Is that a Lua script?
I have a supplementary table (which is too large to put into Lua script or post in a message) which has all the RNG values, their position in the sequence, the wire crossing pattern generated when it appears as "Current value" in the Lua script, and the number of frames from Bospider descent to when he opens his eye. (Technically, the value at "Current value" doesn't generate the pattern; it is the last RNG value used after the pattern is generated. A value 10 or so positions in advance starts the generation of the pattern.)
This table can be generated in any programming language.
There are about 1000 values out of the 43534 which give perfect Bospider (no crossing). This actually differs from estimation by true randomness (3/512 probability, or about 255 values); the reason for this is because the RNG is flawed. If a value is the same parity (even or odd) as the previous value, then the next value will very likely be the same parity. Likewise, if a value is of different parity as the previous, then the next will very likely be different. This affects the wire pattern greatly. Ideal patterns rely on long sequences of values that switch between even/odd.
But for the same reason, many of the ideal values all tend to clump together. So there are a lot of regions where ideal values don't appear for a long time.
As for the pattern Hetfield90 manipulated, the RNG value he used is actually about 20 positions prior to a perfect value. Actually, there are a whole bunch of perfect values clumped together in that region; this is because when the RNG value is hex FFFF, the following values will be FFFE, FFFD, FFFC, FFFB, FFFA, FFF9, FFF8, ... , FFAA (yes, this is another flaw in the RNG) and every four values will give a perfect pattern.
Now that we know that we are 20 positions short, we can of course go back to the last time we can advance the RNG 20 (or more) positions without losing time and redo everything from there. The problem is, I don't think Hetfield90 wants to do that, considering "last time" is back at Armadillo boss fight. Hex-editing input to change RNG is not an option.
Edit: Oh, I misunderstood Hetfield90's last post. Apparently he did get to the next perfect pattern, but at the cost of a lot of frames in the previous room.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Wow, TPP is ruined.
The streamer implemented a thing called "Battle Tent" (like Battle Tower/Frontier in later games) in this Red Anniversary hack, which I suppose was the thing where the mob could grind for money.
Apparently, no quality testing was ever done, because the game glitched and caused the entire Pokemon party to be lost. Red is now blacking out every 4 steps. Also, the game was saved.
Before that, there was another incident (about an hour ago) where the game glitched and caused Missingno. to appear in the party. Then the game crashed. Fortunately they could go back to their last save.
On another note, the mob actually managed to get through the Rocket Hideout maze completely on anarchy about three hours ago. Then this happened. (Watch the chat closely for their reactions.)
I've never seen so much failure in the span of four hours.
Edit: Looks like the streamer went back to an earlier savestate to fix it. But as long as the Battle Tent exists, who knows what will happen.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Time for an update.
The mob couldn't even wait three days and three hours before they started a PC killing spree that claimed 75% of their Pokemon, including their starter (this was some time after beating Misty and they were in Vermillion City). As a result, they were cursed to wander the world until they could beat Blue on S.S. Anne well into the sixth day.
The stream has mostly lacked drama since then. Route 9 was done pretty fast. Rock Tunnel took a while but they managed to box their Dig Pokemon (probably remembering what Digrat did last year). They reached Celadon and managed to beat the gym there (while cutting all the trees) and cut their way out; I'm impressed with how fast they did that.
Looks like they are going to do Rocket Hideout. I'm interested in seeing how long it takes them to do that. Don't think Democracy has been activated yet.
(Or can they use Pokedoll skip on ghost Marowak in the Pokemon Tower?)
ALAKTORN wrote:
Probably check the map ID for no changes within an allotted time? Or some version of that.
I heard that it is only implemented for problem areas, where if they spend too long in an area (where leaving the area only pauses the timer), then Democracy will be activated. That's just what I heard though.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Here is my attempt to improve bkskip1 method:
http://tasvideos.org/userfiles/info/20924232140952311
Even at fastest, I think it is still slower than both no skip (hadoken) and parrot14green's bkskip6_earlyenemy example.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
I don't have all the details, but from what it looks like, this is a hard mode version of Pokemon Red. All wild and trainer Pokemon have been changed and are at higher levels. For example, I just saw the player get knocked out by a hiker using level 20-something Electric-type Pokemon.
So for obvious reasons, the mob is going through a bit slower than last year. That, and they can't decide whether to catch Pokemon or progress the plot.
I found a site that updates everything about TPP, including the previous games (this is basically an advanced version of the Google Doc info linked in the past). For this installment, it keeps a list of known gym leader Pokemon, and a list of known wild Pokemon by route. The site is below:
http://twitchplayspokemon.org/
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
TPP Pokemon Red Anniversary is underway (yes, it's been a year since the whole TPP thing started):
http://www.twitch.tv/twitchplayspokemon
It is Pokemon Red under Anarchy, switching to Democracy only if too much time is spent in one area. Not exactly sure how it works. Also, there is a new objective which involves catching all Pokemon. I don't know how it works either. According to tvtropes:
According to the creators of the special TPP Version being played, there is a brand-new "True Ending" that has been added into the game. To unlock it, 3 conditions must be met, and then the Mob must challenge the Elite Four again. The conditions are as follows:
1.) The game must be cleared once (Player must be registered into the Hall of Fame).
2.) The Diploma must be obtained (Catch all Pokemon).
3.) The bed in the player character's house must be checked after meeting the above two conditions.
Before you ask, nobody knows why we have to check the bed. We just do.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Hetfield90 wrote:
Fractal, have you ever messed with these dash speed gap jumps in the climb before Vile in Sigma 1?
https://www.mediafire.com/?l0g04e45e1b82dr
They're pretty easy to set up in Sigma 1, but I am wondering if it is possible to do on the first ladder in Kuwanger's stage. It seems like the ladder gaps in both stages have the same width and distance from the wall, but I can't seem to get it.
I think it depends on the player X and Y position. I don't know too much other than that.
By the way, the file you uploaded has no extension, but I eventually figured out that it was a SMV file. Now I know why the SMV format requires the file to begin with "SMV".
ALAKTORN wrote:
I wish I knew how to disassemble games. Do you know any guides I could learn from?
I used Geiger's Snes9x Debugger, which I think is the best tool so far, even though it is a little buggy (aren't all emulators like that anyway? :) ). It supports breakpoints, which is by far the most useful thing to have when disassembling.
Be warned that TASing, even now, is at the low end of reasons for disassembly. For the majority of stuff, you will never need to use disassembly. The main reason I use disassembly is to find out how RNGs work, and how the game uses the RNG to make decisions (basically, stuff that is almost impossible to figure out without disassembling and is relevant to the TAS).
Before you even start disassembling, you will need to know how to find addresses using RAM Search. Finding the RNG isn't always easy either (but one thing that helps is that it is more chaotic than almost all other addresses). Once you know the RNG address, you then use the debugger to set read/write breakpoint on the RNG, and then whenever the RNG is read from or written to, the debugger will breakpoint and you can then follow the code. (There is a bug where if you load a savestate, you have to open the breakpoint window and click OK for it to work again.) From there, there really isn't any single process that must be followed.
See this page for more information.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
I found that BizHawk 1.9.1 will crash when the following steps are done:
1. Load a ROM.
2. Load Lua console and load a Lua script.
3. While the emulator is paused, close the Lua console, and click No when asked to save session.
4. Load Lua console again and load a Lua script.
BizHawk 1.9.1 will then crash with the error dialog as follows:
I also reported it to the bug tracker.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
I am aware that Hetfield90 has had trouble with Vile, so I'll post what I figured out by disassembly, if it helps.
There are four courses of action that Vile may take, possibly based on distance between Vile and X, or based on RNG, or both:
- If the distance between Vile and X is 160 pixels or greater, then Vile may either begin by dashing/walking, or jumping back. If the RNG value at that time is even, Vile will begin by dash/walk. Otherwise, Vile will jump back.
-- If Vile begins by dash/walk, then the next RNG value will determine whether Vile dashes or walks. If it is 2 or 7 mod 8, then Vile will dash. Otherwise Vile will walk.
-- This also occurs at the beginning of the fight.
- If the distance between Vile and X is between 32 and 159 pixels, then Vile will choose between dash/walk. If the RNG value at that time is 2 or 7 mod 8, then Vile will walk. Otherwise Vile will dash. (Note that this is backwards from the above.)
-- This also occurs when Vile begins walking from 160 pixels or more away and first closes the distance to less than 160 pixels.
- If the distance between Vile and X is less than 32, then Vile will choose between jumping backward or forward. If the RNG value at that time is 0, 3, or 4 mod 8, then Vile will jump forward. Otherwise, Vile will jump backward. (This hasn't been tested well, since it doesn't factor into the optimal strategy.)
-- This does not apply for the situation below.
- If Vile has chosen to walk or dash up to X, then when he is close enough, he will either jump back or jump straight up. If the RNG value at that time is 0 mod 8, then Vile will jump straight up. Otherwise, Vile will jump backward.
Applying to TAS, the optimal manipulation is as follows:
- First, manipulate Vile to start dashing right away (RNG value is even, and the next value is 2 or 7 mod 8). The alternative, walk first and then dash 5 frames later when within 160 pixels of X, is a bit slower.
- X should be walking towards him at the same time. When Vile is close, X should shoot him; this freezes him in midair about 8 pixels off the ground when he attempts to jump. Whether Vile attempts to jump straight up or jump back before he is frozen doesn't seem to matter much (jumping back just moves him back one pixel).
- At this time, X should jump up, remaining close to Vile but at least 32 pixels away, and get hit by the ball that Vile shoots.
- At about this time, Vile will (since X is at least 32 pixels away) choose to dash or walk in midair. He should be manipulated to walk, so RNG should be 2 or 7 mod 8. Vile will then attempt to walk, and simply fall down to the ground. (If he attempts to dash, he will still fall to the ground, but towards X, which doesn't allow the following to work.)
- About 20 frames later, Vile will be on the ground and choose to dash/walk towards X. He should be manipulated to walk, so RNG should be 2 or 7 mod 8.
- As a result of this, Vile will arrive just as X does his "helpless" pose; this immediately triggers the next scene and Vile doesn't have to walk any distance.
This is very difficult to manipulate; the flexibility of this strat is almost nothing. In particular, the manipulation for RNG to be 2 or 7 mod 8 after Vile is frozen requires two RNG values which are 2 or 7 mod 8, with one of them 24 values in advance of the other. (I did get 23 in advance a few times, but that was when I shot Vile with a green charged shot, which doesn't seem to be possible in this strat.)
By the way, some other MMX disassembly I did a long time ago: http://tasvideos.org/forum/viewtopic.php?p=284826#284826
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
I'm currently modifying Pasky13's Lua script (don't know where it originated, but Hetfield90 gave it to me).
I'm also in the process of disassembling intro stage Vile's behavior, which is a major roadblock. I now have a good idea of what is required to get optimal behavior from Vile.
Edit: Oh, the script is from the BizHawk 1.9.1 Lua folder.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
dwangoAC wrote:
Also, in case anyone is curious, the balloon takes you up 10 levels but I out-climb it, or to rephrase, the only way I'd be able to reach the balloon would be to slow down which would obviously not help.
Oh? And I thought that the balloon was just one more thing trying to kill you.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
mklip2001 wrote:
However, I think the game still gives you the ability to progress if you beat up the old man instead of letting him show you the map.
I think that the game here intends that, after you sing the karaoke, you beat up some NPCs and then the old man will show up and trigger the map menu, then you choose either the 5-minute option or the 1-hour option, and then after that you beat him up.
In this TAS, you beat him up and leave before the map menu is triggered. Since this run glitches to the end, the only thing that matters is beating the old man up, and not the map.
Edit:
To whoever is encoding this run: Please include the secret message at the end. It occurs about 5 minutes after the apparent end screen.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Hi agwawaf.
The glitch works as follows:
Addresses 7E:1628, 7E:1658, 7E:1688, 7E:16B8, ...
are memory for objects like items, moving platforms, and doors. Each slot is 48 bytes. When an object is produced, the game places it in the lowest empty spot in memory. When an object is removed, the game removes it from memory without changing any other slots.
The boomerang glitch happens when an object is produced at the exact same slot and at the exact time that the game removes an object held by the boomerang cutter.
The method in your movie (which I tried earlier) doesn't work because the life recovery item was produced while the platforms were already in memory. As far as I know, the only way to start the glitch in this area is with the heart tank (life up).
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
The Brookman wrote:
I'm on a Dell with intel core 2 duo cpu p8400 2.26 ghz with 4.00GB ram.
Bizhawk is running my SNES game at ~25 fps. and i can't increase or reduce the frame rate significantly even at 6400% increase it runs at ~25 fps.
Would disabling "Use Ring Buffer IO" improve the fps for you? (When SNES ROM is loaded, SNES -> Options -> Uncheck "Use Ring Buffer IO".)
I also reported it to the bug tracker.