Developing a better LL system

[uberCuber]

Oh boy I cannot wait to see this so I can decide how I want to apply it to 4x4.

 

[Akash Rupela]

I m not really sure what i m saying, but rather than having most cases start with sune, wont it be better to group them such as the starting alg is same depending on certain common characteristics(i know its something tough and saying is easy but still)?

Also, Great job, i hope this is in learnable or even semi-learnable form(something more learnable than the previous one), cuz i have holidays till mid july, so would be good if i can get this soon

 

[Cubenovice]

I m not really sure what i m saying, but rather than having most cases start with sune, wont it be better to group them such as the starting alg is same depending on certain common characteristics(i know its something tough and saying is easy but still)?

Let's wait a little bit and see the final output and proposed recognition.
Once the full set is available* you're free to reorganise in any way you like.

Thom,
the times of your posts make me wonder: how much sleep do you get nowadays?

Anyway, keep up the good work but make sure you don't crash...
We can wait (but you probably cannot )

*
Thom, I propose you that you put a paypal "donate" button onto the page you're going to publish this

 

[LarsN]

3853/4128
Added M'UMU2M'UM, R'U'R'FRF'UR, RUR'U'M'URU'r'

Forget about the rest. Lets just make some 1LLL algs for those few cases

 

[Kirjava]

I m not really sure what i m saying, but rather than having most cases start with sune, wont it be better to group them such as the starting alg is same depending on certain common characteristics(i know its something tough and saying is easy but still)?

This is the idea I proposed most recently. I'm going to attempt to apply it to what I have once it's already organised. If that doesn't work, I'll finish documenting it in this style and look to do it the other way from scratch, although that will take a lot longer than this way did. (Maybe, since I thought it'd take months to get to the stage I am at now.)

Let's wait a little bit and see the final output and proposed recognition.
Once the full set is available* you're free to reorganise in any way you like.

People are already free to reorganise it however they like, since the full table is available.

Maybe I'll put my organised version on the wiki once it's finished so people can constantly improve it.

the times of your posts make me wonder: how much sleep do you get nowadays?

5 or so hours a night during the week.

Anyway, keep up the good work but make sure you don't crash...
We can wait (but you probably cannot )

Thanks! It's actually quite enjoyable to do this, so I don't mind all that much at the moment. The fact that I might have something tangible soon is great motivation.

Thom, I propose you that you put a paypal "donate" button onto the page you're going to publish this

I appreciate the sentiment, but I do it for the love

Forget about the rest. Lets just make some 1LLL algs for those few cases

275 is way too many, I'd prefer to only have to do 10 or so.

One thing that's bugging me is that the more I document, the fewer cases each alg covers. I think once I hit about 100 left I'll organise them by hand.

 

[Cubenovice]

One thing that's bugging me is that the more I document, the fewer cases each alg covers. I think once I hit about 100 left I'll organise them by hand.

Or perhaps you throw in a couple more nice algs and run the solver again, just on those remaining cases.
Maybe now is already the right moment for such an excersize?

 

[bobthegiraffemonkey]

Even though I will likely only use a small subset of these I would probably send a small token of appreciation via that donate button if you decide to have one, mostly out of kudos for pulling this craziness off (when/if it gets organised into a good format), and I'm sure quite a few others would too. Just sayin' .

 

[asportking]

Or perhaps you throw in a couple more nice algs and run the solver again, just on those remaining cases.
Maybe now is already the right moment for such an excersize?

Which ones is he using now? I agree, using a few more would probably help.

This method sounds like it'll be really fast and useful. One thing I don't quite get is this:

Each case only needs one solution, and the number of cases you need to memorise will end up being quite a bit smaller than the number of actual LL cases.

I can understand why you'd need to memorize significantly less algs than there are cases, but wouldn't you still have to memorize which two algs you use for every case?

 

[Kirjava]

I can understand why you'd need to memorize significantly less algs than there are cases, but wouldn't you still have to memorize which two algs you use for every case?

No. You only need to learn the first alg for each case, and my case reduction idea combines recognition from multiple cases into one.

 

[Petro Leum]

so would be sune/antisune and niklas /antiniklas (lol) as first alg be an option for methods like ZZ were teh edges are already oriented?

 

[AbstractAlg]

This sounds amazing. Can't wait for it. *_*
Idea is amazing, good luck!!

 

[Kirjava]

so would be sune/antisune and niklas /antiniklas (lol) as first alg be an option for methods like ZZ were teh edges are already oriented?

I might make a ZBLL version, but I don't really want to because this idea doesn't work as well if you're doing EOLS aswell.

 

[mullemeckmannen]

how´s it going Kirjava?

 

[Dacuba]

how´s it going Kirjava?

Be patient. Good stuff needs time.
He will bump the thread by himself when he is ready for more updates.

 

[Kirjava]

Wow that was a lot of work. Updates!

Sorry this took so long, the more cases you do, the harder it is to do more at once, since less cases are available to be covered by a single alg.

So I've organised all 4128 cases by algorithm. At the moment I've just got it organised loosely in a bunch of files, I'll convert this to a sexy HTML format for people to oogle at sometime soon.

That'll end up being the final product eventually. However, it still won't be very friendly initially, but will be somewhat learnable! A lot of work will still need to be done, but all the initial grind it out of the way - all that's rest is fun discovery

I need to do case reduction, find patterns and discover other shortcuts. Please help once I've posted the neat version if you have any idea what I'm talking about

I've found a few patterns and little quirks already, but it's kind of hard to spot with how the data is now. Once it looks pretty it should be a lot easier to improve :3 Just need to keep throwing man hours at the damn thing until it's playing nice >

 

[sa11297]

Wow that was a lot of work. Updates!

Sorry this took so long, the more cases you do, the harder it is to do more at once, since less cases are available to be covered by a single alg.

So I've organised all 4128 cases by algorithm. At the moment I've just got it organised loosely in a bunch of files, I'll convert this to a sexy HTML format for people to oogle at sometime soon.

That'll end up being the final product eventually. However, it still won't be very friendly initially, but will be somewhat learnable! A lot of work will still need to be done, but all the initial grind it out of the way - all that's rest is fun discovery

I need to do case reduction, find patterns and discover other shortcuts. Please help once I've posted the neat version if you have any idea what I'm talking about

I've found a few patterns and little quirks already, but it's kind of hard to spot with how the data is now. Once it looks pretty it should be a lot easier to improve :3 Just need to keep throwing man hours at the damn thing until it's playing nice >

awesome, your work is much appreciated

 

[jeff081692]

So I started my feeble attempt at learning this and I wonder is the number of cases really 4128? For example I counted at least 26 cases that were actually PLL cases. Also When practicing recognition I noticed that image 464 and 553 are the same case just a different AUF which leads me to believe that more cases could be over counted.
example: case 464 solution is U Inv(Sune) Inv(Sune) and case 553 solution is U' Inv(Sune) Inv(Sune).

 

[Kirjava]

The site's starting to come together.

It's taking forever because there are thousands of cases. I've been writing little scripts and stuff to speed things up, but it's still a lot of grinding.

 

[Christopher Mowla]

Thom,

This is the first time I looked at this thread (sorry!). I have read all of it, and I have a question.

Maybe I missed something, but, by looking at the raw data table, you seem to have dozens of algorithms for each case. How are you (or anyone else) supposed to organize theses cases by algorithms if there is more than one alg listed for each case? I might be interested in helping, but I don't know where to start because of this.

Guess: Is it to choose the algorithm for each case so that we have the minimum number of subsets? If so, how are you separating subsets? I've read that you are gearing to have algorithms start with sune, antisune, etc. Is that your approach? If so, what is the "second best" starts to continue with after that? And after that?...

If I didn't guess right that time, is it based on speed (I am aware that you have chosen fast algs to generate all cases, but I am referring to when they are combined)? If so, how is anyone ever going to agree on this (since there are so many alg combinations for each case)?

A lot of LL situations are just a 3-cycle of corners and a 3-cycle of edges. It would only be like 20 algs to give a lot of 2-alg 1LLLs.

EDIT: 12 edge cycles and 12 corner cycles, and not too bad algs either.

I don't know what the "actual" number is, but from the raw number of cases (with mathematics) for 3-cycles of edges and corners, if I did my calculations right, is 22.2% of the raw number of LL cases:

\( \frac{\left( 8\times 3^{3} \right)\left( 8\times 2^{3} \right)}{\frac{1}{2}\left( 4! \right)\left[ 2^{3} \right]\left( 4! \right)\left[ 3^{3} \right]}=\frac{2}{9}. \) But this is all irrelevant to this thread anyway.

 

[LarsN]

The site's starting to come together.

Sounds great. Are there any news on a potential recognition method?