CFOP-Breaker? Mehta method

[Devagio]

Finally came round to write code for simulating solves with Mehta, accounting for everything that HARCS could not. Here are the results for a 1000 solve simulation -

Step Mean StD. Median Min Max
FB (CN, unfixed): 5.158 0.730 5.0 2 7
3QB (unfixed): 4.988 0.968 5.0 1 7
EOLE (speed-optimal): 7.278 1.547 7.0 3 11
Finish (speed-optimal): 26.274 3.321 27.0 15 35
Total: 43.698 3.863 44.0 30 56

Out of 1000 solves, the shortest solutions were using:
CDRLL: 446
JTLE : 129
6CP : 467
APDR : 200

Here is the movecount histogram


The basic take-away is that the method averages ~43.7 moves; and with the intuitive portion of the solve being only 10.1 (which is the only place where humans will be inefficient compared to computers), the practical / human movecount average for speedsolves should be 45-46 moves at worst (assuming 20% inefficiency for intuitive steps). This keeps the method up there with Roux in terms of efficiency (said to be ~48 moves in human speedsolves), but with an algorithmic approach.

More detailed stats and the code are put up in discord. I will soon do a simulation for the TDR variation as well.

 

[DNF_Cuber]

Have you thought about after 6CO, instead of permuting all the corners, you could just do the D layer ones, and then go into HKPLL? (or PLL+1 I think it's the same.)
Just an idea.

 

[Jam88]

Have you thought about after 6CO, instead of permuting all the corners, you could just do the D layer ones, and then go into HKPLL? (or PLL+1 I think it's the same.)
Just an idea.

Could make a nice variant

 

[Devagio]

Have you thought about after 6CO, instead of permuting all the corners, you could just do the D layer ones, and then go into HKPLL? (or PLL+1 I think it's the same.)
Just an idea.

We did consider HKPLL and TTLL variations, but compared to the sets currently in use these have pretty bad algorithms on average. What these could be useful for are as trick subsets. Basically, the solver learns the better algs in the algset. Then, for example if 2 of the 3 bottom pieces happen to be solved after 6CO (would happen about once every 10 solves), the solver would check to see if it is one of the case they know; if they do know it then they would in essence be skipping a step.

 

[DNF_Cuber]

We did consider HKPLL and TTLL variations, but compared to the sets currently in use these have pretty bad algorithms on average. What these could be useful for are as trick subsets. Basically, the solver learns the better algs in the algset. Then, for example if 2 of the 3 bottom pieces happen to be solved after 6CO (would happen about once every 10 solves), the solver would check to see if it is one of the case they know; if they do know it then they would in essence be skipping a step.

Okay, I was just wondering if you had thought of it.

 

[Devagio]

16000 Solves were simulated using each of the 4 paths individually. The entire data file and code is available on discord, here is the summary of the data.
Note, FB and 3QB are move-optimal CN solutions, and the rest is speed-optimal algorithms; thus the human move-count during a speedsolve would be 1-3 moves higher than the indicated values.

With full option-select (which is basically the theme of the method) the human move-count could very well average around 45 moves as indicated in one of the previous posts. For a high-TPS algorithmic method, this should be plenty to at least compete at the highest level.

 

[Cubing Forever]

16000 Solves were simulated using each of the 4 paths individually. The entire data file and code is available on discord, here is the summary of the data.
Note, FB and 3QB are move-optimal CN solutions, and the rest is speed-optimal algorithms; thus the human move-count during a speedsolve would be 1-3 moves higher than the indicated values.

View attachment 14308

View attachment 14309

View attachment 14310

View attachment 14311

With full option-select (which is basically the theme of the method) the human move-count could very well average around 45 moves as indicated in one of the previous posts. For a high-TPS algorithmic method, this should be plenty to at least compete at the highest level.

What about TDR?

 

[Devagio]

What about TDR?

It is not a part of the option-select variant, it is a method by itself. I will generate the statistics once the final algorithms are ready. Currently we are done with about 200/350 of the algorithms.

 

[Spencer131]

The 'problem' with a 500 algorithm method isn't learning the algorithms-- it is spending the HUGE amount of time generating-testing-regenerating algorithms for each case.

Have you seen AlgExplorer? It doesn't always produce optimal solutions but this could be a good starting place.

Original post: https://www.speedsolving.com/thread...ine-utility-to-assist-in-alg-searching.65189/
Github: https://github.com/john-ml/AlgExplorer

 

[crystalcuber]

Here's my video tutorial for the beginners' variant of Mehta:

I also worked with Devagio and Andreas to design a logo for Mehta:

 

[OreKehStrah]

Here's my video tutorial for the beginners' variant of Mehta:

I also worked with Devagio and Andreas to design a logo for Mehta:
View attachment 14363

The logo looks sick! Good job!

 

[crystalcuber]

The logo looks sick! Good job!

Thanks!

I wonder if Mehta is the first method to have a logo
That would be cool

 

[DNF_Cuber]

Thanks!

I wonder if Mehta is the first method to have a logo
That would be cool

well CFOP has the first cross, roux has Blocks+Corners, and ZZ has line and those are kinda like logos.

 

[Devagio]

The Mehta Website is finally complete! This is meant to be a one-stop place for the method; it has EVERYTHING that there is to know about the method in one place including algs, stats, relevant links, etc. I've tried to make it as concise and user-friendly as possible.

Mehta

3x3x3 Speedsolving Method - Mehta

devagio.github.io

 

[Silky]

The Mehta Website is finally complete! This is meant to be a one-stop place for the method; it has EVERYTHING that there is to know about the method in one place including algs, stats, relevant links, etc. I've tried to make it as concise and user-friendly as possible.

Mehta

3x3x3 Speedsolving Method - Mehta

devagio.github.io

Niiiice dude. Are you going to put up YruRU as well ?

 

[Cubing Forever]

The Mehta Website is finally complete! This is meant to be a one-stop place for the method; it has EVERYTHING that there is to know about the method in one place including algs, stats, relevant links, etc. I've tried to make it as concise and user-friendly as possible.

Mehta

3x3x3 Speedsolving Method - Mehta

devagio.github.io

It's soooo cool

Niiiice dude. Are you going to put up YruRU as well ?

I think there's one already. I can't find it

 

[Cuberstache]

Niiiice dude. Are you going to put up YruRU as well ?

YruRU

One-handed speedsolving method YruRU

devagio.github.io

 

[ProStar]

I'm finally taking an actual look at this method, and it looks really cool. However, I'm having trouble deciphering your alg organization system. Could someone please give an explanation of it?

 

[Cuberstache]

I'm finally taking an actual look at this method, and it looks really cool. However, I'm having trouble deciphering your alg organization system. Could someone please give an explanation of it?

Check out the github page on OS: https://devagio.github.io/Mehta/mehta_os.html
The image on the front page is also helpful: https://devagio.github.io/Mehta/

 

[Devagio]

Here is a leaderboard of the top averages and singles of users with Mehta.
To submit your times, join the discord and look for the announcement dated 21 January (or 22 January for Eastern countries).