Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
I've seen that "end input early and let the game beat itself" thing too often in the past to be excited about it nowadays. Many times it comes across as a cheap gag that will never be funny more than once. At least this TAS is uninspiring enough that it justifies using this trick to raise its entertainment.
The other thing which no one else (other than Spikestuff) commented on was removing the music for quicker loads. I mean, sure, a TASer has the right to do everything to post the smallest possible time on the submission thread's title, but cutting out the music for that? Really?
So, of the ~5.5 seconds that this TAS appears to improve over the previous, only a tiny fraction of it is actual gameplay improvement.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Hm, so it is actually a finished TAS.
GlitchMan, are you going to submit this TAS sometime soon? Perhaps you are redoing the TAS because of the left+right walljump glitch found in http://tasvideos.org/forum/viewtopic.php?t=18038 ?
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Amaraticando wrote:
I was thinking of flat earth bullshit and this problem came to my mind. The answer that I got was pretty big and complicated for a seemingly simple problem, let's see if some of you finds out.
Click to expand.
Let h1=h+a and h2=h-b. Then
cos-1 (r/(r+h1)) + cos-1 (r/(r+h2)) = 2 cos-1 (r/(r+h))
based on the angle subtended by the line segment between the tower and the horizon from the center of the earth, for each of the towers h (twice), h1, and h2. Solving for h2:
h2 = (r/D)-r
where D=cos( 2 cos-1 (r/(r+h)) - cos-1 (r/(r+h1)) )
Then take b=h-h2.
If you want D in terms of algebraic elements, then using that cos(cos-1(n))=n when -1≤n≤1 and sin(cos-1(p/q))=sqrt(q2-p2)/q when 0≤p≤q gives:
D= {2(r/(r+h))2-1}{r/(r+h1)} + 2{(sqrt(h2+2hr)/(r+h))(r/(r+h))}{sqrt(h12+2h1r)/(r+h1)}
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
All this problem is, is what Bobo the King said above with the ratio tending to 3:2:1, and the negative binomial distribution applied to the expected number of total drops required to get one of each gem.
If probability of success is p and probability of failure is q=1-p, then the expected number of trials required to obtain success is 1(p)+2(qp)+3(q2p)+... = p(1+2q+3q2+...) . Using the identity 1/(1-x)2 = 1+2x+3x2+... , the expected number is p(1/(1-q)2)=p/p2=1/p. Ex. The expected number of rolls of a 6-sided die required to roll a 1 is 1/(1/6)=6.
To calculate the expected number of total drops required to get one of each gem, condition on the order in which the types of gems are first encountered. That is, we calculate:
1 + P(Common)*{E(# of drops to get Uncommon or Rare) + P(Uncommon|Uncommon or Rare)*E(# of drops to get Rare) + P(Rare|Uncommon or Rare)*E(# of drops to get Uncommon)}
+ P(Uncommon)*{E(# of drops to get Common or Rare) + P(Common|Common or Rare)*E(# of drops to get Rare) + P(Rare|Common or Rare)*E(# of drops to get Common)}
+ P(Rare)*{E(# of drops to get Common or Uncommon) + P(Common|Common or Uncommon)*E(# of drops to get Uncommon) + P(Uncommon|Common or Uncommon)*E(# of drops to get Common)}
=
1 + (1/2)*{2 + (2/3)*6 + (1/3)*3} + (1/3)*{3/2 + (3/4)*6 + (1/4)*2} + (1/6)*{6/5 + (3/5)*3 + (2/5)*2}
=
1 + (1/2)*7 + (1/3)*(13/2) + (1/6)*(19/5) = 1 + 7/2 + 13/6 + 19/30 = 30/30 + 105/30 + 65/30 + 19/30 = 219/30 = 7.3
Since the ratio of the types of gems tends to 3:2:1, the expected number of common, uncommon, and rare gems, respectively is 7.3/2 = 3.65, 7.3/3 = 2.43333... , and 7.3/6 = 1.21666... . This is consistent with the numbers given by OmnipotentEntity's program.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Now that I got around to watching, this TAS was merely "OK" for me. I wish it wasn't possible to skip so much.
Lord Tom wrote:
Kaizo, a Japanese word that translates as "reorganize" or "restructure"
The word that comes to mind for me is "remodel". Or maybe just "modified" or "changed" since all the other terms sound too glorified for things like ROM hacks. :D
Lord Tom wrote:
Bisqwit wrote:
The submission text speaks of closed captions, but the video does not have any.
Anyone else having this problem? They're published for English, and I see them when I'm not logged into my YT account as long as I press the button to enable cc...
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
BizHawk 1.11.6 has been released! Updates include latest mGBA core and savestate fix (old mGBA savestates do not work in this version). For more details, see http://tasvideos.org/BizHawk.html.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
hidaigai wrote:
>FractalFusion
Hi, thanks for encoding Lua.
My TAS's rerecords is 13261 because my bk2 is generated by Tastudio (BizHawk-1.10.0). Tastudio seems to save rerecords. (Is my usage wrong?)
I don't know if you used other additional files, but when I do a TAS of something like Mega Man X2, it takes me around 60000 rerecords. Mega Man X2 is not easy to TAS.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Hi hidaigai.
Nice run, and thanks for the Lua output. Did it really only take 13261 rerecords?
Also, I'm sorry that I missed your e-mail from one month ago. The run looks good.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
I heard that Spikestuff was interested in doing an all-bombs TAS (though I don't know how true that is). I'll leave the route here just in case.
The time on the clock for each level is:
Starting from Cellphone and up until Doorknob, the player is given a clock screen and chooses one of two stages to do next. Selecting the next stage requires time for the clock to turn; so the transitions should be minimized. The stage transitions that minimize the time are:
To 10:29 cassette tape:
- From 14:15 rail crossing
To 23:58 light bulb and 15:24 kawai rika:
- From 10:12 ice coffee and 11:42 pinball machine
To 15:16 lunch:
- From 15:24 kawai rika
To 13:18 ocean and 19:43 engine:
- From 11:39 gunder and 21:23 jukebox
with all other transitions forced.
Thus the optimal route which has the fewest saves/loads (3 each, the minimum) is:
Pairs of >-type bullet points mark save (first) and load (second) points. Saves and loads should be done on the second slot of whichever memory card(s); using the second slot is faster than using any other slot. The Moon branch and the Rail Crossing branch can be switched if necessary (unless only one memory card is used).
Contact me if anyone is interested in doing more with Suzuki Bakuhatsu.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Spikestuff wrote:
Finally, the Moon. This bomb is a 50/50 too,
Spikestuff wrote:
You guessed it another run killer.
Just want to get this out of the way before I move to my thoughts on Suzuki Bakuhatsu.
The bomb is only 50/50 if you cut it blindly. In each level (except the last one on Hard), there are hints as to which wire to cut, which range from finding notes to read, to investigating wires for information, to using visual cues of the layout to determine which one to cut. Some of the notes necessary to determine which one is buried deep into dead-end tunnels. So therein lies the dilemma for an RTA; waste time to confirm the wire to cut, or don't waste time and take the 50/50 chance.
----
OK, now my thoughts on Suzuki Bakuhatsu.
Suzuki Bakuhatsu is one of those really weird Japanese games with a cult following (but not in the west; there isn't even an English Wikipedia page for it). Basically the premise is that someone with a twisted sense of humor is sending bombs to Miss Suzuki, and she has to disarm them. All the bombs are puzzles. Said bomb-sender is also mentally ill and the notes on the bomb are mostly messages professing his love for Suzuki, if they aren't hints on how to solve the puzzle.
There are 19 bombs in the game (of which this TAS only does 10). The route map is as follows:
I'm guessing that the route in this TAS is probably the fastest? Moon and Rail Crossing are a toss-up, but you'd have to do Moon because Ice Coffee is so much faster than the other two on its tier. Kawai Rika seems faster than Light Bulb, and Gunder is faster than the other two for a TAS because of luck manipulation (for RTA, who knows). Both Ocean and Engine are long and complex but Engine I think is a bit faster. Note that the 9th tier is entirely optional.
Anyway, this game has you unscrewing a lot of stuff. You get a time limit to disarm the bomb; you get less time on Hard. Sometimes, unscrewing something will trigger an emergency explosion timer for which you have to unscrew the entire set in the allotted time. Not often seen on Easy or Normal, Hard sprays this feature everywhere and gives really short time limits (some of them test your button-mashing). Also, some aspects were changed from Easy and Normal to make it that much harder in Hard. Basically, Hard is insanely hard and not for casuals or new players.
Regardless of difficulty, the game also plays mind games with you.
----
If you want to see someone attempt and fail at playing Hard, Nicovideo has this Japanese person:
(account) http://www.nicovideo.jp/mylist/11420012
(no account) http://www.nicozon.net/mylist/11420012
----
By the way, my thoughts on each level from what I can remember about them:
Orange:
An introductory level. Certainly easy on Easy and Normal but on Hard? Let's just say you have only one minute... NOPE YOU HAVE 32 SECONDS LOL.
Pistol:
Another introductory level. This level has a trap to punish you if you unscrew immediately without looking carefully. The other thing is that the existence of scissors that threaten to cut wires indicates which wire to cut.
Cellphone:
The first wandering-only level. There are "phones" giving conflicting information to cut red or blue. Only by seeing the vision of a person do you know which wire to cut. (Of course, in an RTA or TAS, you'd just skip directly to the bomb; the wire to cut is always the same.)
Moon:
It's a moon! No wait, it's a missile! The puzzle of unscrewing the moon's head is tricky (hint is that the whole thing looks like a screw at that point). This level has a dead end, which tells you to the wire to cut on Hard. On easier difficulties, that might not be the case; I can't remember. Also, there's a nasty trap here where screwing a gear wrong kills you; it's deliberately set up so that it's very easy to do by accident.
Rail Crossing (not in the TAS):
You have to examine the wires to determine which one makes the sound; combined with one of the notes it should tell you which wire to cut. Also the dreaded pendulum makes an appearance here; you have to unscrew slowly or else it will set off the bomb. This level is murder on Hard.
Ice Coffee:
Another wandering-only level. The state switches between EXPLODE and NOT EXPLODE each time you move. It's set up so that if you go straight forward, you will always reach the bomb in the EXPLODE state. You have to triangulate your movement (as the TAS does here) to reach the bomb in the NOT EXPLODE state.
Cassette Tape (not in the TAS):
Honestly this level isn't even that hard compared to Moon and Rail Crossing. I think this one has the vibrating wire; I don't remember how the notes in this level logically point to the right wire to cut because it was rather vague.
Pinball Machine (not in the TAS):
This one has a hidden note telling you which wire to cut (at least on Hard). Also, the pendulums make a return, with a ball acting as a timer. While most of the level is doable, that ball-and-pendulum section is insane on Hard. You are forced to switch between screws repeatedly while unscrewing them.
Kawai Rika:
Kawai Rika is the name of the pop idol in the last cutscene, but she gets turned into a box because... whatever. While this bomb is rather easy, there is a multi-part note hint that involves digging into a dead end; on Hard, this comes as a statement as to which wire to cut, but then another statement saying that it was a lie.
Light Bulb (not in the TAS):
This bomb is intermediate in difficulty. By the way, there is a special room in this level with like 20 notes that serves solely to indicate how messed up the person sending you the bomb is. Somewhere in the level there is a hint as to cut the wire that glows in the dark (or the wire that doesn't; this is random); this comes into play when you switch off the light. Strangely it doesn't glow when you are right up against it in complete darkness trying to decide which one to cut.
Gunder:
This is just rock paper scissors, and it's all luck.
Lunch (not in the TAS):
This is a forward-only level. The game tries to throw a bunch of misleading information before giving you the choice of wire to cut which should be obvious if you pay attention (btw it's the same every time).
Jukebox (not in the TAS):
Another wandering-only level. You have to go to the right spot, then logically deduce the wire to cut (again, it's the same every time).
Ocean (not in the TAS):
Only in this game can an ocean be a bomb. You actually start in this laggy area and use torpedoes to destroy some submarines. When you get to the submarine-bomb thing, you unscrew. A lot. There is a note that says which wire to cut, and another note that basically goes like this: "Take what I said and do the reverse of the reverse of the... of the reverse". Yeah, the wire to cut is determined by the number of 逆 in the note. The message has like 50 of these 逆 in it on Hard (iirc in Hard it always ends up equivalent to a single reverse). Also Hard throws in some really nasty emergency timers, and you can't even see the "reverse" note without setting off a timer.
Engine:
This is hard like Ocean. First off the game baits you into taping the side broken wire then unscrewing the bolts; if you did that without taping the bottom broken wire, you die. After that it's puzzle after puzzle. On Hard, the game even throws in some really bad emergency timers as well. The note telling you which wire to cut is buried in some dead end, and you even have to realize to turn the gear the right way as well to see it.
Shadow (not in the TAS):
A short and intense level. You have to knock down the pointy hammers but guard the balloon from popping (which results in death); on Hard, this is basically trial and error since the game is trolling you badly. After unscrewing everything, the game throws the bomb up so it lands in front of you in any orientation; which side the bomb comes down determines which wire to cut.
Heating Table (not in the TAS):
A forward-only level that takes you back to school. Doppler effect, blah blah blah. The game then quizzes you for the correct wire to cut (it's the same every time).
Doorknob:
Two words: more puzzles. As if it weren't hard enough, Hard puts emergency timers on them as well.
Ito:
Climactic showdown with a person-bomb. One thing I like about this level is the reactor-like thing. You have to unscrew the screws but doing so "unbalances" the reactor and if it is too unbalanced, you die. That's why the TAS unscrews part of one, then part of another, and so on. Also, as I said before, you hear "cut red" or "cut blue" but on Hard this is independent of which wire to cut.
----
About RTA: I haven't thought much about how an RTA would work in this game, but I suspect that no one is going to bother. The reason I think so is because of both luck and simplistic design. On Hard, you'd have the final wire, then the rock-paper-scissors, then at least a couple other bombs where guessing on the last wire as opposed to finding the hints saves like 45s each. Getting a perfect rock-paper-scissors by the way is 1/27. So not a lot of people want to attempt that luck. On the other hand, the simplistic design discourages people from making concessions to increase luck. So we have a situation where nobody will concede on luck since "any determined person can beat it with perfect luck", but nobody wants to suffer to do it with perfect luck.
Final note: I like this game. However, I understand that it doesn't really make that good of a TAS, and the game having Japanese cult status doesn't really change that (see also http://tasvideos.org/2489M.html , a game with a similar cult status but an arguably worse reaction from this community).
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Hourglass doesn't support every PC game. Furthermore, I can't determine any way of knowing whether any given game will work with it, other than trying all the various options.
Furthermore, even if it did work and you were able to use it to make a TAS, there is no guarantee that anybody else will be able to play it back correctly.
In short, Hourglass is terrible. Don't use it.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Alyosha wrote:
What is being selected is the hard AI in Stella (the binary one) but the easy AI in Bizhawk (the kind of dog face looking one.) This is because hard AI happens to be 5 (mod 6) and Stella ends up with $A2=23. Easy AI is 3 (mod 6) and Bizhawk gives $A2=21.
I've played the TAS on emulator, and judging from the speed of the CPU paddle, it looks like BizHawk is using hard AI. Same as the encode earlier in this thread.
To others, please be more specific than "Hard mode". For some reason the author of this game decided to have multiple layers of so-called difficulty:
In particular:
● COM vs. EASY COM vs. 2P
COM means CPU goes fast ("hard AI"); EASY COM means CPU goes slow ("easy AI"), 2P is just 2-player mode. This selection is part of the menu.
● expert - FAST vs. novice - SLOW
expert - FAST means ball goes fast; novice - SLOW means ball goes slow. This is controlled by the left difficulty switch. Left difficulty false means expert - FAST; left difficulty true means novice - SLOW.
● CLASSIC vs POORLORDS
CLASSIC means without walls and first to 10 wins; POORLORDS means with one wall and first to 3 wins. This selection is part of the menu.
This submission uses, or at least is supposed to use, COM, expert - FAST, CLASSIC, though Alyosha says that BizHawk is using EASY COM mode somehow (but I don't understand because it isn't happening in my case, and not for the person who did the encode either).
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Just to be clear, there are two forms of difficulty switches in this game. It's in the manual up there.
- "Expert" (fast ball) vs. "Novice" (slow ball). This is controlled by the left difficulty switch.
- "COM" vs. "EASY COM". This is part of the menu selection.
Please be clear which switches you are referring to when you say "easy difficulty", "hard difficulty", etc.
Also, an "Expert" (fast ball) run does not necessarily finish faster than a "Novice" (slow ball) run. Sure the ball goes faster, but that doesn't mean the game ends faster.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Using EASY COM mode now are we? Why bother with CPU at all? Might as well go straight to 2 player.
http://tasvideos.org/userfiles/info/28759645062563661
So-called "best ending" achieved in 7 frames.
(Thanks BrunoVisnadi for the idea.)
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Flip wrote:
If I have the first n natural numbers arranged in a random order, how many would you expect to be in their correct locations?
The expected value of the number of numbers is their correct locations (1 in 1st, 2 in 2nd, etc.) is the sum of the expected values of the functions "1 if number i is in correct location, 0 otherwise" over all i from 1 to n. Since (n-1)! of the n possible orders have number i in the correct location, the expected value is n*(n-1)!/n! = n*(1/n) = 1.
The orders are commonly known as "permutations" and a number in its correct location is known as a "fixed point". As an aside, note that the number of permutations with exactly k fixed points is C(n,k) dn-k, where di is the ith derangement number. So the expected value is Sum{0≤k≤n} (k C(n,k) dn-k / n!). This gives an identity:
Sum{0≤k≤n} (k C(n,k) dn-k / n!) = 1.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
49 frames on Expert mode (fast ball, left difficulty = "false")
I was actually hoping that simply starting the game would have been enough (at most 11 frames). However, there are only four starting directions for the ball and no RNG. So there are only 8 possibilities (four starting directions on two difficulties) and none of them give a win without a lost ball.
So the next best thing is to play the ball once. The fastest way to do this is on Expert mode (fast ball) and start with the ball going down-left. Technically the ball going up-left reaches the wall a couple frames faster, but the problem is that it is too high for the strat used.
The horizontal speed of the ball is a single value that depends only on difficulty, and the point of contact with this strat is always the same (it is possible to contact the ball behind the paddle face, but not with this strat). The vertical speed after contact depends only on the vertical speed of your paddle. (Vertical speed of the ball in sub-units is RAM address 0x0E, two bytes signed little endian.) By allowing the paddle to bounce off the ceiling and then the floor, I can get a ball vertical speed of -141, which results in a win without a lost ball.
49 frames was the best I could find; I tried a few other things but they were slower. I tried not hitting the ceiling; this allows for earlier end of input while still contacting the ball, but the lower height and near-zero speed on contact means the ball only gets a vertical speed of -32 or 32 (default values when too close to 0), neither of which result in a win without a lost ball.
dwangoAC wrote:
In BizHawk, the default Atari 2600 sync settings are configured to have both players set with the difficulty switch set to true, i.e. "Expert" mode as it's described for this game.
Actually (according to BizHawk) when left difficulty is "true", the mode is "Novice" (slow ball). To get "Expert" (fast ball), the left difficulty should be set to "false".
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
I sent Hetfield90 the following script which he said worked.
This script uses a special key to skip forward two frames. The space key is used as the special key; you can edit it to whatever you normally use as frame advance. For this to work, the actual frame advance key has to be set to something other than space. Note that the special key cannot be held down for key repeat.
Language: lua
local key="Space"
local t={}
local s=emu.framecount()
local monitor=false
local buttonheld=false
while true do
t=input.get()
if t[key] and not buttonheld then
s=emu.framecount()
client.unpause()
buttonheld=true
monitor=true
end
if not t[key] and buttonheld then
buttonheld=false
end
if emu.framecount()>=s+2 and monitor then
client.pause()
monitor=false
end
emu.yield()
end
Note that emu.yield() instead of emu.frameadvance() allows the while loop to run even when emulation is paused.
Please also remember that emu.frameadvance() does not advance emulation by one frame; it halts lua script execution until the next frame. This is an easy misunderstanding to make.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
Under Config -> Hotkey Config... there are options to set keys for Auto-Hold and Auto-Hold Clear.
To set a controller button to be held down, hold the Auto-Hold key and then press the key for the controller button. To clear everything so that nothing is held down, press the Auto-Hold Clear key.
Editor, Experienced Forum User, Published Author, Skilled player
(1941)
Joined: 6/15/2005
Posts: 3247
OK, I looked at choosing the next guess based on the Bayesian probability distribution.
That is, if I understand it right, we make a guess with probability Prob(guess is correct|evidence).
Since
Prob(guess is correct|evidence) = Prob(evidence|guess is correct) * Prob(guess is correct) / Prob(evidence)
and a priori Prob(guess is correct) is always 1/10000 and at each stage Prob(evidence) is a fixed number, the probability distribution is proportional to likelihood.
So we just select the next guess proportionally according to
prA qrB pwC qwD,
or, after dividing by pwA+C qwB+D, which is a fixed number at each stage,
(pr/pw)A (qr/qw)B.
I modified the previous C++ program to get a new program that uses this method, and here are the results. The average number of tries is larger than for the maximum likelihood strategy by up to 50% but otherwise shows similar patterns like the other results.
Click to expand.
In particular:
● COM vs. EASY COM vs. 2P
COM means CPU goes fast ("hard AI"); EASY COM means CPU goes slow ("easy AI"), 2P is just 2-player mode. This selection is part of the menu.
● expert - FAST vs. novice - SLOW
expert - FAST means ball goes fast; novice - SLOW means ball goes slow. This is controlled by the left difficulty switch. Left difficulty false means expert - FAST; left difficulty true means novice - SLOW.
● CLASSIC vs POORLORDS
CLASSIC means without walls and first to 10 wins; POORLORDS means with one wall and first to 3 wins. This selection is part of the menu.
This submission uses, or at least is supposed to use, COM, expert - FAST, CLASSIC, though Alyosha says that BizHawk is using EASY COM mode somehow (but I don't understand because it isn't happening in my case, and not for the person who did the encode either).