The fastest possible internal times program

[dawid]

What about program that will calculate the fastest time in each int lev? U know 'code' of elma physic (u have it in elma 2). So all u have to do is a little change code of elma 2. My tips:
-program should calculate all possible combinations of rides except some cases* (up to a appointed time of ride):
#for example: in int 10 program should stops calculation when time is more than 10:50 (or we can use SL cheats to decrease the shortest possible time) and it calculate the next combination. When program will find shorter time, it should shorts the time up to a shortest found time to optymalize calculations.
-it will be hard to do but if we use something like CUDA (nvidia) the calculations will be much faster than CPU.
-program should save about 10 fastest recs.
-first lev which should be calculate is int10 because it has the shortest time (the least calculations)
-the combinations shouldn't contain avolt because it appear when there is combination of left and right arrows at the same time.
!the combinations of pressed volt keys should be calculate only every 0.8 sec (to avoid calculating pressed volt keys when kuski cant volts)
-OFC program should works very fast, not in real time (the physic and the timer should works the fastest possible speed of CPU (program without timers))
-we should be able to set the simulating fps of physic (maybe 60?). The less fps, the faster calculations.
!there shouldn't be the combinations where the throttle and brake keys are pressed at the same time (because then only breake key works) so program should choose from brake and throttle state from one variable.
-I should learn more english
*vide ! points

[Tigro]

get some supercomputer, or some grid, and lets get on with it.

[Kopaka]

and then we all quit elma because there is nothing to strive for anymore

[dawid]

and then we all quit elma because there is nothing to strive for anymore

WTF? the times made by program will not be in wr...

[dawid]

get some supercomputer, or some grid, and lets get on with it.

U need supercomputer for elma? I think that on the PC it can take about 5-10 days to calculate int 10.

[jonsykkel]

to put things in a litel perspective: there are 5 keys but the volts are a bit triky so lets say 3 keys, gas brake turn

Code: Select all

gas  -x--x-
brake--x--x
turn ---xxx

6 states per frame
all posible cumbinations for 10 seconds at the lowest allowed framerate(30): 6^(30*10)=2.789*10^233
an average computer can calc phisics 50 000 times per sec, lets say you improve that to 500000000 per sec
(2.789*10^233)/500000000/31536000(seconds in year)=
1.769*10^217 years to calc int10 excluding volting

[analcactus]

jonsykkel best sig

[dawid]

to put things in a litel perspective: there are 5 keys but the volts are a bit triky so lets say 3 keys, gas brake turn
all posible cumbinations for 10 seconds at the lowest allowed framerate(30): 3^(30*10)=136891479058588375991326027382088315966463695625337436471480190078368997177499076593800206155688941388250484440597994042813512732765695774566001

gas and brake is one key (one variable, 3 states:gas, barke, neutral)! and volt keys works about every 0.8 sec so u cant use ur formula. We can apply second variable to volts: 4 states-right, left, right and left (avolt), neutral. So there is only 2 varabiles and one of these variables works only every about 0.8 sec. elma can works much faster without displaying graphic and without timers in code (100% CPU).

[jonsykkel]

to put things in a litel perspective: there are 5 keys but the volts are a bit triky so lets say 3 keys, gas brake turn
all posible cumbinations for 10 seconds at the lowest allowed framerate(30): 3^(30*10)=136891479058588375991326027382088315966463695625337436471480190078368997177499076593800206155688941388250484440597994042813512732765695774566001

gas and brake is one key (one variable, 3 states:gas, barke, neutral)! and volt keys works about every 0.8 sec so u cant use ur formula. We can apply second variable to volts: 4 states-right, left, right and left (avolt), neutral. So there is only 2 varabiles and one of these variables works only every about 0.8 sec. elma can works much faster without displaying graphic and without timers in code (100% CPU).

ops my math was retarded fixed a litle, also im said lets exclude volts from that so duno what you mean by "so u cant use ur formula"
when im said average pcs can do about 50000 phisics frames im meant without graphics

[culinko]

i remember mila wanted to do something like teh, but it's impsy with so many calculations imho. not sure how he did his 32tt project, afaik with recs merging and some tuning and whatnot.

[Mawane]

I like this topic, wouldnt mind getting spoiled about best Warm up possible time at least with like 10 digits

[Ropelli]

Well, jonsykkel pretty much proved it's impsy to just bruteforce the ride. /thread

[Mawane]

http://www.iflscience.com/technology/ne ... cond-could
wonder how fast it would be with this über computer

[Chris]

http://www.iflscience.com/technology/ne ... cond-could
wonder how fast it would be with this über computer

I think memristor tech is mostly useful for working with huge data sets, what is not case in elma.

[kuchitsu]

-program should calculate all possible combinations of rides

See you in 10000000000000... years!

[dawid]

to put things in a litel perspective: there are 5 keys but the volts are a bit triky so lets say 3 keys, gas brake turn

Code: Select all

gas  -x--x-
brake--x--x
turn ---xxx

6 states per frame

U still dont understand me. In program we use variable, not keys. Gas and brake is in one variable

-program should calculate all possible combinations of rides

See you in 10000000000000... years!

CUDA, cloud computing...
and the most important: Moore's law. Maybe we dont have to wait too long time...

[kuchitsu]

dreaming dreaming alice in wonderland

[jonsykkel]

to put things in a litel perspective: there are 5 keys but the volts are a bit triky so lets say 3 keys, gas brake turn

Code: Select all

gas  -x--x-
brake--x--x
turn ---xxx

6 states per frame

U still dont understand me. In program we use variable, not keys. Gas and brake is in one variable

i did that, notice gas and brake never on at the same time

CUDA, cloud computing...
and the most important: Moore's law. Maybe we dont have to wait too long time...

lal what planet are yu living on, nat sure you understod that number i gave
if you made a supercomputer with 1 000 000 000 000 times the procesing power i puted in my post (which is alredy ridiculously exaggerated) and gave 1 000 000 000 000 of those to every person on the planet (~7 000 000 000) and then copied all that another 1 000 000 000 000 000 000 000 000 000 times you would decrease the time it takes to get optimal int10 run (still without volting) to:
2527000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 years
(obviusly a very rough estimate, should be noted that you could stop procesing branches where the palyer dies etc)

[dawid]

to put things in a litel perspective: there are 5 keys but the volts are a bit triky so lets say 3 keys, gas brake turn

Code: Select all

gas  -x--x-
brake--x--x
turn ---xxx

6 states per frame

U still dont understand me. In program we use variable, not keys. Gas and brake is in one variable

i did that, notice gas and brake never on at the same time

CUDA, cloud computing...
and the most important: Moore's law. Maybe we dont have to wait too long time...

lal what planet are yu living on, nat sure you understod that number i gave
if you made a supercomputer with 1 000 000 000 000 times the procesing power i puted in my post (which is alredy ridiculously exaggerated) and gave 1 000 000 000 000 of those to every person on the planet (~7 000 000 000) and then copied all that another 1 000 000 000 000 000 000 000 000 000 times you would decrease the time it takes to get optimal int10 run (still without volting) to:
2527000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 years
(obviusly a very rough estimate, should be noted that you could stop procesing branches where the palyer dies etc)

dunno. But isnt it 3 states per frame, not 6?

[jonsykkel]

no, only gas and brake would be 3, with turning its 6

Code: Select all

state  123456
gas    -x--x-
brake  --x--x
turn   ---xxx

even if it was 3 it would take 1240000000000000000000000000000000000000000000000000000000000000000 years with the specs from my last post

[FinMan]

dawid u a stupid XD wtf XD

[ville_j]

I scent an litel Koopa2 here. However, would be really interesting to play around with some .exe which could be seeded with a rec and then apply some variations at certain point of the run... someone make that!

[dawid]

I would do it right that (there should be 3rd variable for volt):
And look here:
http://webdocs.cs.ualberta.ca/~chinook/
http://www.sciencemag.org/content/317/5 ... 0Schaeffer

[pawq]

Come on dawid, stop embarassing yourself and trying to prove the impossible. No one has even mentioned the random element of elma, which would make any such calculations pretty much worthless. Just trust the old community here and stop boggling yourself about how to do it.

[dawid]

Come on dawid, stop embarassing yourself and trying to prove the impossible. No one has even mentioned the random element of elma, which would make any such calculations pretty much worthless. Just trust the old community here and stop boggling yourself about how to do it.

calm down bro, It was just my crazy idea I thought it could be possible...

[pawq]

Plenty of people tried explaining it to you before me, but you seemed to go on hard about it ;)

[Chris]

I would do it right that (there should be 3rd variable for volt):
And look here:
http://webdocs.cs.ualberta.ca/~chinook/
http://www.sciencemag.org/content/317/5 ... 0Schaeffer

So for 60 frames it will be 36^30 combinations? Do you realize how many combinations is it just for 1 second of 60 fps gameplay?

[analcactus]

ez calculate any chris level in a couple of hours

[dawid]

hmm. maybe the quantum computers will handle it

[Tigro]

dawid u a stupid XD wtf XD

^ Very this.
No kind of BOINC is gonna do this. Even if you somehow manage to reduce time of computing to 1/1000th of what jon already said it would take, there are still enough zeros...
And I still sense some kind of entropy in the whole elma universe... And considering 1000 fps playstyle: GG.

[A.K.B.]

Zweq is supercomputer imo.

[VALDO]

Zweq is supercomputer imo.

and did all the calculations with pen and paper

[kuchitsu]

I still would be interested in seeing such a bruteforcer done, for example to use it on these extremely short battles that Nekit likes. Maybe pos calculate under 2-3 sec levels in a realistic time? chnek04.lev for example.

[dawid]

I still would be interested in seeing such a bruteforcer done, for example to use it on these extremely short battles that Nekit likes. Maybe pos calculate under 2-3 sec levels in a realistic time? chnek04.lev for example.

I think that it will be boring because it won't be much difference between bruteforce and normal recs.

[pawq]

My next year uni project actually involves something that could probably be used here, genetic algorithms... Not like it can be done anyway, but it would probably be the way to go :p

[Stini]

I'd probably approach this as a reinforcement learning problem, I guess this way you could actually make an AI that can finish levels. Also the bruteforce way could be easily improved by dynamic programming, although it still wouldn't make the problem feasible.

[anpdad]

Just find style (i.e. all moves and approximate timings of brakes/gas/volts) and optimize with genetic/swarm algorithms part by part to cut 0.0x everywhere. Quick and dirty. There's really no other way unless you're megapr0 and have shitton of time on your hands. But then i'd choose to earn 10k/month instead of wasting time on elma coding lel xDxDxD

[dawid]

Just find style (i.e. all moves and approximate timings of brakes/gas/volts) and optimize with genetic/swarm algorithms part by part to cut 0.0x everywhere. Quick and dirty. There's really no other way unless you're megapr0 and have shitton of time on your hands. But then i'd choose to earn 10k/month instead of wasting time on elma coding lel xDxDxD

I think you are wrong because of http://en.wikipedia.org/wiki/Chaos_theory
look at chaos in attractors.
If u cut 0.0x at start, u can lose pro run at middle of lev.

[anpdad]

Apparently either someone lacks reading comprehension skills or is just trolling big time. Anyway though, I specifically said you need to already know all the moves exactly and just be willing to perfect the ride further, by a pretty low margin (it can be substantial in small levs and can add up to couple seconds in labpro at best).

There's no magical algorithm that would find styles/voltings for you, not even close.

[dawid]

There's no magical algorithm that would find styles/voltings for you, not even close.

I never said that there was that algoritm.

[bene]

But what if you cutting 0.02s in the first second of the level cause you to get 0.09 too much speed at 25.534s in the level so the push from the vertex at coordinates 243,542 makes you lose 0.46 speed or 0.09s. CHAOS THEORY BITCH.

I found this thread very amusing yesterday, then anpdad came and was all russian and serious.

[anpdad]

oke i turn ov brane for b0ne. ye that is obvius and is already popping up all the fukin time while saveloadink. some levs are norm, like fat trak, warmup, frefal and stuff, so can go all out in those and not carE xD. Pretty sure you'd be able optimize int34 mid and end well enough with you're knowledge tho xDxD.

[Thundr-]

i dont even know why i read this now that i have quitted elma and deleted all unsent int wr recs and state.dats i have, but i also find this thread amusink very good read after work each day helps me relax my ass muscles so i can take a nice big shite and lose 1kg weight every day.

sorry for trollpost but all elmaquitters is doing it and i just want to be popular and trendy like Madness

[Memphis]

I do enjoy this post as well and I don't play much. I find it quite sad that everyone is jumping on dawid for trying to think outside the box a bit. Yes, some of the counter points brought up good information but other stuff was just unessicary. Yes it is a far fetch dream but most great inventions start out that way. This is typical society jumping on inventors telling them they are crazy. Anyway, keep doing what your doing dawid.

[dawid]

I do enjoy this post as well and I don't play much. I find it quite sad that everyone is jumping on dawid for trying to think outside the box a bit. Yes, some of the counter points brought up good information but other stuff was just unessicary. Yes it is a far fetch dream but most great inventions start out that way. This is typical society jumping on inventors telling them they are crazy. Anyway, keep doing what your doing dawid.

Thx. And remember that there are less combinations than formula shows because a lot of combinations will fail after 1 sec,3 sec so the calculations will be faster.

[Lousku]

I do enjoy this post as well and I don't play much. I find it quite sad that everyone is jumping on dawid for trying to think outside the box a bit. Yes, some of the counter points brought up good information but other stuff was just unessicary. Yes it is a far fetch dream but most great inventions start out that way. This is typical society jumping on inventors telling them they are crazy. Anyway, keep doing what your doing dawid.

It's nat outside the box. It's an old idea and has been discussed before with teh same conclusion. Geniuses often mistaken for nutjobs, but that doesn't mean every nutjob is a genius. :P

[kuchitsu]

It was discussed probably like 1000 years ago when someone thought that he could "solve" chess by trying all possible moves.

[dawid]

It was discussed probably like 1000 years ago when someone thought that he could "solve" chess by trying all possible moves.

http://webdocs.cs.ualberta.ca/~chinook/
http://www.sciencemag.org/content/317/5 ... 0Schaeffer

Problem?

[kuchitsu]

ye the problem is that chess is not checkers i gues. a lot more complex game.

[dawid]

ye the problem is that chess is not checkers i gues. a lot more complex game.

And the elma is not chess. But as I said earlier, we have to wait for quantum computers