When will the WR TT be under 35?!

culinko · Post by **culinko** » 30 Nov 2008, 23:39

no new styles has been done since the wr tt break. let's hope some guys will succeed in doing some other new styles

pawq · Post by **pawq** » 30 Nov 2008, 23:44

CuLiNo, wat are u talking about?? what with ziggy, steep, bb??

western · Post by **western** » 30 Nov 2008, 23:52

isn't there a thread about potential new styles? e.g. pipe bounce, crazy's warm up -- that one could save 0,02! that's almost half of a tenth of a percent of what's needed to reach the next minute break. aren't you beginning to feel a lil' bit depressed too or that's just me?

that's the realism's hitting, folks -- way to swallow truth in all its harshness. you're welcome for doing so.

teajay · Post by **teajay** » 1 Dec 2008, 01:04

no fucking chance western, we're the last to rule out certain uncertainties and the neverending story of improvements in elma. Not until the fat lady sings. But it's not like people are hÃ¶yling their asses off on warm up at the moment. They're eager to get some ez wrs yes indeed, but there's no such thing as a wr spree going on. TT has been dropping like a snail on a gravel trail this year, and this is going to be the same for a long while. When we hit some <35:30, improvements might knock on the door more easily, but for now and the year to come, I'd say 10 secs max improvement, and that's when considering revival when EOL final comes out.

western · Post by **western** » 1 Dec 2008, 01:47

interesting perspective. the more it approaches down to 34:xx, the more people will spend their time (even their dreamtime!) searching for new styles. the paradox is the more they want it, the more difficult it will become. whatever happens, you've got me addicted for the next twelve years...

culinko · Post by **culinko** » 1 Dec 2008, 15:11

Pawq wrote:CuLiNo, wat are u talking about?? what with ziggy, steep, bb??

sorry, forget teh steep. but never seen recs in zig and bb so dunno if they are new styles

Grace · Post by **Grace** » 2 Dec 2008, 06:28

zigzag is new style, not sure about bounce back though?

pawq · Post by **pawq** » 2 Dec 2008, 18:31

hey, zweq showed rec from 51.xx run with impsy-to-find style, and he said that he wont share 49,78 rec until next year so it must be new style

body25 · Post by **body25** » 3 Dec 2008, 02:48

tijsjoris wrote:no fucking chance western, we're the last to rule out certain uncertainties and the neverending story of improvements in elma. Not until the fat lady sings.

Youre wrong here and Western is right. There might be some new styles but its not gonna cut time dramatically.
Its called math, and go check exponential functions and/or strd deviation.

tijsjoris wrote:But it's not like people are hÃ¶yling their asses off on warm up at the moment.

tijsjoris wrote: They're eager to get some ez wrs yes indeed, but there's no such thing as a wr spree going on. TT has been dropping like a snail on a gravel trail this year, and this is going to be the same for a long while. When we hit some <35:30, improvements might knock on the door more easily, but for now and the year to come, I'd say 10 secs max improvement, and that's when considering revival when EOL final comes out.

Makes no sence and its inconsitent.
//
WR tt will never get <35. Internals are hoyled to teh max

western · Post by **western** » 3 Dec 2008, 04:58

I kind of get the feeling of what tijsjoris is saying. no wonder each time I write tijsjoris it's still just pure old copy 'n paste.

Zweq · Post by **Zweq** » 3 Dec 2008, 06:19

body25 wrote: WR tt will never get <35. Internals are hoyled to teh max

how can they be hoyled to teh max when nobody has hoyled them to teh max?

Grace · Post by **Grace** » 3 Dec 2008, 07:16

[quote="body25]WR tt will never get <35. Internals are hoyled to teh max

[/quote]

i suggest that you wait until you are informed before you make decisions like this, people thought it would never go under 39.

Memphis · Post by **Memphis** » 3 Dec 2008, 07:54

ya it's the most noob comment someone can make saying that the wr tt will not go under a certain limit. Unless you have holyed each level going for wr you don't really know how much time cut can be done on each of them still. I know speaking for new wave that with perfect run of some 1 in 1000 style you could probably be looking at 1:20:xx and I am sure there are a lot of levels like this. WR TT will go under 35min but it will take hella long time. As long as people still play internals wrs will be beaten.

Lukazz · Post by **Lukazz** » 3 Dec 2008, 10:38

i totally agree with memphis. i tried some tag styles and merged recs and stuff and even in this short level you can ez cut of 0,3 sec with lots of hoyling!

Grace · Post by **Grace** » 3 Dec 2008, 10:40

Body25 -> you are nab.

culinko · Post by **culinko** » 3 Dec 2008, 15:30

yep, zweq, memphis, lukazz and smaxa are right. there is old golden rule: nabs can't see the way of improving times and think internals are hoyled to teh max. i know there won't be 54 new styles, but always someone can cutt some time from teh.

The_BoneLESS · Post by **The_BoneLESS** » 3 Dec 2008, 16:17

Still, i can't help but thinking otherwise

Mathematicly speaking, there is a limit. And saying otherwise is simply unrealistic.

Sure perfection is hard (impossible) to reach, but we sure are getting closer in a few levels. (warm up, flat track, hang tight, etc., and please don't say "ez Wr" because first, this sentence has been overused recently and i am tired of hearing it and second, these levels can only be improved by a few tenth of a second, this will hardly get us through 35 so, imagine 34.)

I say, there is a limit. And i can imagine 35 being beaten. 34, i am having a hard time seeing that happen.

welle · Post by **welle** » 3 Dec 2008, 16:43

of course there's a limit, since the wr tt is monotonically decreasing and non-negative

western · Post by **western** » 3 Dec 2008, 16:44

that's understandable. would be interesting to see it graphically, though.

Zweq · Post by **Zweq** » 3 Dec 2008, 16:57

i've said this before and am willing to say it again:

- theoretical limit
- human limit

theoretical limit has always been the same value, if you first set rules for it(bugs). we wont really ever know the theoretical limit. even if the computer can hoyl your style, it wont find the routes for you. perhaps you can teach computer to find the max grip in levs like bumpy or haircut, but im very doubtful of any computer being capable of finding routes, in let's say apple harvest, without any breakthrough in the AI.

whatever that value might be then, my guess is 33tt.

human limit is another thing, which is just totally unpredictable. surely if you hoyl the next 10 years 15h a day you will have a 34tt but definitely never 33tt (assuming that's the theoretical limit).

ive brought this up too before: let's say theoretical best tt is 33.30. let's just take any stupid example, let's take internal 43: He He. the world record is 56.61, all my best merges would go down to 55.50 at least, have a computer ride it and it might as well be 54.91. that is 3% improvement to the 56.61. then let's say computer can make 13.94 warm up, now that's approx. 0,5% improvement. either of these times would be absolutely impossible to achieve(with current styles!:wink:). you could say the longer the lev, the more error there is in the human time. OK OK it totally depends on the nature and hardness of the tricks of the level, but this is why i said it's a stupid example.

since we're already far in fantasy at this point, i might as well invent more numbers in my head. i might as well say lab pro is improvable by 5% by the computer. lab pro is a 134.9 secs lev, improve it by 5% and you have 128.155 secs which is 2.08.15

if human is always(let's invent more numbers!) let's say 3% off from the theoretical tt limit and the human has (woop!) 35.xx tt the theoretical tt will be a tad over 1 minute away

buaah all the mathematicians are probably crying :)

culinko · Post by **culinko** » 3 Dec 2008, 17:33

hey zweq! mila thought about computer program that tries every single combinations of volts, gas, brake, alo etc. in level and calculate the best time. imho the program finds the new styles... but he did nat tell me if he did teh and what's the tt :X

Lukazz · Post by **Lukazz** » 3 Dec 2008, 17:44

LOL that would take pretty long for levels like labby pro

teajay · Post by **teajay** » 3 Dec 2008, 17:45

oh hey western, I kind of misread your post up there, I read too fast and took it as you were saying ez under 36 any time soon. So in relation to yours, my reply was rather inconsistent.

welle · Post by **welle** » 3 Dec 2008, 17:58

CuLiNo wrote:mila thought about computer program that tries every single combinations of volts, gas, brake, alo etc. in level and calculate the best time. imho the program finds the new styles... but he did nat tell me if he did teh and what's the tt :X

take warm up for example. if you stop every run after 14 seconds and only play with 30 fps, you still have 420 frames. Possible keypress-combinations are at least
{gas, break, none}x{left, right, alo, none} + turn
of course theres no need to volt only few frames after the last one, but if there were only three possibilities per frame, you get 3^420>10^200 different runs. Since some of them will fail before 14 seconds pass, lets say we have like 10^100 which take the 14 seconds. Even if you could simulate it 1000 times faster than real-time, this would take more than 10^90 years, which is far more than 10^70 times the age of our universe...

Memphis · Post by **Memphis** » 3 Dec 2008, 18:13

ya but thing is it's just like world records in real life really. They are broken all the time. Just look at Usain Bolts taking off a bunch of time in the 100 meter sprint. 10 years ago if you told some runner that time they would say impossible but it has been done. I understand elma you can't use human ability as much per say but if you try hard enough you can set wr. But I agree with boneless with noobs coming on these forms and saying ez wr. Unless you have already driven it don't say ez wr and even if you have you realize it wasn't all that easy and it took a fair bit of time to holya.

I think why noobs say ez wr is to make themselves feel tougher. It's like saying they could drive it easily but don't want to bother.

Lukazz · Post by **Lukazz** » 3 Dec 2008, 18:15

welle wrote:
CuLiNo wrote:mila thought about computer program that tries every single combinations of volts, gas, brake, alo etc. in level and calculate the best time. imho the program finds the new styles... but he did nat tell me if he did teh and what's the tt :X
take warm up for example. if you stop every run after 14 seconds and only play with 30 fps, you still have 420 frames. Possible keypress-combinations are at least
{gas, break, none}x{left, right, alo, none} + turn
of course theres no need to volt only few frames after the last one, but if there were only three possibilities per frame, you get 3^420>10^200 different runs. Since some of them will fail before 14 seconds pass, lets say we have like 10^100 which take the 14 seconds. Even if you could simulate it 1000 times faster than real-time, this would take more than 10^90 years, which is far more than 10^70 times the age of our universe...

HAHA love that mathematics shit

body25 · Post by **body25** » 3 Dec 2008, 18:25

Nice post welle

So we would have to wait (with this method) many many many years in the afterlife, only to find out warm up WR still would be Cruyffs number

Memphis · Post by **Memphis** » 3 Dec 2008, 19:59

ya very cool to think about it like that welle. Someone should make it now though so that maybe in 4000 years our great x 15 grand kids could see the best wrs for 54 levels. I am thinking that Artificial Inteligence would be a better method though. Just buy some robot to play elasto all day long 24/7.

milagros · Post by **milagros** » 3 Dec 2008, 20:37

1. you dont need to go through all possibilities
2. its standard optimization problem where decent local minimum is sufficient

welle · Post by **welle** » 3 Dec 2008, 21:00

milagros wrote:1. you dont need to go through all possibilities

I know, but I was referring to "every single combinations".

milagros wrote:2. its standard optimization problem where decent local minimum is sufficient

But I'm pretty sure that it could only find basic styles. If it passes steppes the long way, it eventually hoyls it quite nice, but I doubt it would ever do it over the pole then. In fact I think most of the current wr-styles are impossible to find this way, not to talk about those we don't even know. And even if you let it start with a decent style, I think it will most probably break it, because this problem is very ill-conditioned. Consider LabPro: If it varies the start just a little bit and blindly presses the keys like it had worked before, I can't imagine it finishes the lev. To make the bounce again there's significant adjustment needed, which can't be done by optimization.

culinko · Post by **culinko** » 3 Dec 2008, 21:11

and there's another theory. you have 100 replays of enigma for example, and it "extracts only the best moves" (the keypresses that makes your run the fastest) and make one rec (or calculates time).

Lukazz · Post by **Lukazz** » 3 Dec 2008, 21:18

CuLiNo wrote:and there's another theory. you have 100 replays of enigma for example, and it "extracts only the best moves" (the keypresses that makes your run the fastest) and make one rec (or calculates time).

well, that would be nice of course, but it wouldn't find the theoretical limit just the "best done by human yet"-time

welle · Post by **welle** » 3 Dec 2008, 21:28

And it doesn't work in reality. Maybe your best bounce is part of a run with a bad start. so you have different speed and angle at the bouncing spot than you have in the rec with the best start. now it's impossible to combine the best of both, because they simply don't fit together. only taking the keypresses into consideration makes it even less reliable.

So you have to create your 'best' bounce after making the 'best' start, to make it fit. This concept is called 'Save-Load' and considered cheating for a good reason...

milagros · Post by **milagros** » 3 Dec 2008, 22:07

you can give teh some initial solutions.. like human driven recs

Memphis · Post by **Memphis** » 3 Dec 2008, 22:42

I am still liking the consideration of Artificial Inteligence. I mean if you didn't show it any replays or anything off the start it would end up making new routes to beat wrs and it would have zweq skill after like 2 weeks of holyaing 24/7. (sorry zweg if you found it offensive comparing you to some robot)

Lukazz · Post by **Lukazz** » 3 Dec 2008, 22:48

actually it would scare me if there is such a bot.

welle · Post by **welle** » 3 Dec 2008, 22:54

milagros wrote:you can give teh some initial solutions.. like human driven recs

Yes, I already mentioned that and why I think this doesn't work. Finding a run (on a given lev) as short as possible, is a high dimensioal und very ill-conditioned problem. Its not suited for optimization algorithms.

Mempis wrote:I am still liking the consideration of Artificial Inteligence. I mean if you didn't show it any replays or anything off the start it would end up making new routes to beat wrs and it would have zweq skill after like 2 weeks of holyaing 24/7.

What kind of artificial intelligence are you thinking of? It's not like you can say "finish as fast as possible" and it will automatically find the 'right' routes and drive some wrs. You'll get the same problems as in the optimization approach.

milagros · Post by **milagros** » 3 Dec 2008, 23:26

welle wrote:
milagros wrote:you can give teh some initial solutions.. like human driven recs
Yes, I already mentioned that and why I think this doesn't work. Finding a run (on a given lev) as short as possible, is a high dimensioal und very ill-conditioned problem. Its not suited for optimization algorithms.

pls..

Memphis · Post by **Memphis** » 3 Dec 2008, 23:59

What kind of artificial intelligence are you thinking of? It's not like you can say "finish as fast as possible" and it will automatically find the 'right' routes and drive some wrs. You'll get the same problems as in the optimization approach

no but you program it to just play that all day with the objective of lowest time. Artificial intelegence meaning it's constantly learning from mistakes and improving upon them.

welle · Post by **welle** » 4 Dec 2008, 00:09

milagros wrote:pls..

Could you be a little more explicit there please?

Memphis wrote:no but you program it to just play that all day with the objective of lowest time. Artificial intelegence meaning it's constantly learning from mistakes and improving upon them.

Yes I know what artificial intelligence is, but you don't get for free. You need to specify how some sort of learn rate develops in progress. And this is where the trouble starts...

milagros · Post by **milagros** » 4 Dec 2008, 00:43

welle wrote:Could you be a little more explicit there please?

ez 54 wrs

Memphis · Post by **Memphis** » 4 Dec 2008, 00:49

welle as far as I understand artificial intellegence there is only so much you can code in. after a while there is no predicting how fast the robot learns or improves it is all up to it's intelegince which is artificial and the stuff that impacts him.

western · Post by **western** » 4 Dec 2008, 01:46

Abula wrote:A russian Elma player, CandErz, has programmed a software which calculates the theoretical minimum times of the levels. It uses some kind of Euler algorithms which are very efficient in this kind of processing. (April Fools)

This, of course, would be a dream come true, but I say we need an AI expert with a Ph.D. because admittedly none of us really has the necessary knowledge to talk about such an AI (I'm not even sure if someone on the planet could implement it right now). /2 cents

welle · Post by **welle** » 4 Dec 2008, 02:09

April Fool!

Zweq · Post by **Zweq** » 4 Dec 2008, 12:16

rofl did western get fooled by 4 years old april's or what's up with the edit and welle's post

owned

Uncle Milty · Post by **Uncle Milty** » 4 Dec 2008, 23:20

set, that limitexploration is irrelevant and referring to the fact ( i call it fact), that all wrs can be beaten, lukazz' theory, what i share, is, that way more than the "usual" wr-owners can make wrs, but for not-Johns or -Zweqs it would take even for guys close to some wrs nowadays an enormous amount of time to do so; maybe 2-3 weeks of pure one-lev-hoyling to the max, maybe less, but more likely it would take longer. <- sorry for that sentence

But taking this and expecting those guys to bring up the necessary motivation to hoyl "their" level, current styles can really get far more hoyled than they are at the moment. Then taking some guys like anpdad, Johann, Lukazz or Labs, who develop new styles, sharing them, get them being hoyled even more.

Well :s
Zweq and John don't have directly to do with this above, so all those others can drop tt "significantly" all using current styles. Then there's the thing: personal success / overall success. worldrecordowners rather don't share there recs, so their times won't be improved as fast as they could be. And they won't loose them as fast as they could :>

and here back to topic: Limitbreak comes the early, the personal hamstering stops.

Again: Well :s
Just a reflection.

Lukazz · Post by **Lukazz** » 5 Dec 2008, 08:29

Uncle Milty wrote:Then taking some guys like anpdad, Johann, Lukazz or Labs, who develop new styles, sharing them, get them being hoyled even more.

neither did labs share any new style nor did I!
top secret EF styles! don't talk bout them!

Zweq · Post by **Zweq** » 5 Dec 2008, 12:04

uncle milty: so basically you said wrholders should publish their recs/rides(as in video) so anyone else might see flaws in it and improve the wr -> improve the wrtt? in some cases this works, for example what talli did with hi flyer, xiphias with hill legend and so on. then some cases it doesnt work at all like animal farm, which is one of the crappiest wrs on the table

nick-o-matic · Post by **nick-o-matic** » 6 Dec 2008, 11:22

The WR TT can be taken under 35 if we arranged some computers with elma to some homes of autistic children.

western · Post by **western** » 6 Dec 2008, 15:56

though unethical (unless we could arrange them to be paid), autistic children somehow sometimes show shoes remarkable abilities in one particular field (though lacking in others). these are known as the "dumb geniuses."