OC(P)ELL/CPLL for OH Discussion

[Silky]

Recently found a thread on OC(P)ELL and CPLL as an alternative for LL with oriented edges. Was curious it this could be viable for OH. It's 22 move count between the two of them, one less efficient than OLL/PLL. OC(P)ELL is also completely 2-gen and seems to be better than COLL OH algorithms. Similar to COLL there is a 1/12 chance for CPLL skip which is better than the 1/72 for PLL. I'm currently switching to LLOB for OH and it seems to be very tempting since recognition is easier during EODFDB. Plus I can combine this with 2GLL which would be quite excellent.

Love to heard your thoughts.

 

[xyzzy]

OC(P)ELL is also completely 2-gen and seems to be better than COLL OH algorithms

If we're comparing OCELL+CPLL against COLL+EPLL, it might be useful to compare the 2-gen-able steps with each other instead of comparing them in the progressive order.

OCELL versus EPLL: I think it's clear that EPLL is a solid winner. Z perms and H perms suck if you don't table (I almost never), but they still seem better than most OCELL algs regardless.
CPLL versus COLL: Imo A perms and E perms are kinda disgusting; A perms are maybe slightly worse than the median COLL case, but E perms are basically as bad as the worst COLL cases. That said, CPLL and COLL probably aren't that different (there are awful COLL cases too), but if I had to choose, I'd say COLL is better on the whole.

Similar to COLL there is a 1/12 chance for CPLL skip which is better than the 1/72 for PLL.

Skips are only noteworthy if you didn't have to work extra to get them. The more likely a skip is in an LL/LSLL method, the less a skip is worth.

Plus I can combine this with 2GLL which would be quite excellent.

Could you elaborate?

---
One more thing to watch out for is that OCELL requires you to do one of these two things:
(i) CP recognition (even though you're not solving it in this step!) or
(ii) looking at three sides (either by AUFing, tilting your hand, or tilting your head).

It's impossible to uniquely determine the OCELL case just from corner orientation and two edges. This is unlike how COLL (or even ZBLL) can always be recognised from just the three visible faces. I don't think this is a dealbreaker for OCELL, but it's something you have to keep in mind.

 

[Silky]

Could you elaborate?

This is separate from the OCELL/CPLL discussion. This is just my plan to incorporate 2GLL into my LLOB solves. Since corner recognition is free during EODFDB You only have to pay attention to edges, which should be similar to OCELL.

If we're comparing OCELL+CPLL against COLL+EPLL, it might be useful to compare the 2-gen-able steps with each other instead of comparing them in the progressive order.

OCELL versus EPLL: I think it's clear that EPLL is a solid winner. Z perms and H perms suck if you don't table (I almost never), but they still seem better than most OCELL algs regardless.
CPLL versus COLL: Imo A perms and E perms are kinda disgusting; A perms are maybe slightly worse than the median COLL case, but E perms are basically as bad as the worst COLL cases. That said, CPLL and COLL probably aren't that different (there are awful COLL cases too), but if I had to choose, I'd say COLL is better on the whole.

I think you should keep in mind the E-Perm and Z-perm have the same probability of showing up, respectively. The thing with COLL is that you're far more likely to get a bad case and there are equal chances of getting a bad case with EPLL/CPLL ( depending on whether or not you consider A-perms good ).

If we're comparing OCELL+CPLL against COLL+EPLL, it might be useful to compare the 2-gen-able steps with each other instead of comparing them in the progressive order.

OCELL versus EPLL:

I don't agree with this. In the case of OCELL vs EPLL, EPLL wins because you have 67% for U-perm where OCELL you'll have, on average, slightly worse cases simply because you have more cases. I don't think OCELL algorithms are really that bad, 2-gen is 2-gen even if they're not 100% the best.

I'd say on average you're getting better OCELL/CPLL cases than COLL/EPLL cases. I'd rather have one really bad case than a bunch of kinda of bad cases because I can just grind out the one until I'm really good at it. Maybe they'll just equal out though.

Skips are only noteworthy if you didn't have to work extra to get them. The more likely a skip is in an LL/LSLL method, the less a skip is worth.

I don't really understand this. You're not working any harder between OLL/PLL, COLL/EPLL, or OCELL/CPLL. More PLL skips is always better than less PLL skips.

And just for comparisons sake OCELL/CPLL has 41/46 2-gen algs and COLL/EPLL 12/46 2-gen algs ( I think. Could be closer to 15-16 but that isn't really the point ). For me, bad 2-gen algs are better than bad non-2-gen algs

 

[Tao Yu]

I think A perm is not absolutely disgusting if you use the new alg: x' [R' U': [R U R', D']] and the inverse - not sure if these are widely known yet. I know that Feliks uses it, and Antonio Kam told me he prefers it to the D2 alg. Antonio Kam's E perm alg is also nice and maybe not so well known: F R' F' r U R U' r2 U' R U r f' U' f

It feels rather counter-intuitive to choose LLOB + this method instead of just Roux. Roux's movecount is pretty similar to LLOB + Full ZBLL from my experiments, and you also get a 2gen moveset. I guess you just want to avoid mainstream methods though?

 

[Silky]

I think A perm is not absolutely disgusting if you use the new alg: x' [R' U': [R U R', D']] and the inverse - not sure if these are widely known yet. I know that Feliks uses it, and Antonio Kam told me he prefers it to the D2 alg. Antonio Kam's E perm alg is also nice and maybe not so well known: F R' F' r U R U' r2 U' R U r f' U' f

It feels rather counter-intuitive to choose LLOB + this method instead of just Roux. Roux's movecount is pretty similar to LLOB + Full ZBLL from my experiments, and you also get a 2gen moveset. I guess you just want to avoid mainstream methods though?

Well to be honest I've had a bunch of trouble figuring out which method to use for OH. I'm a Roux solver so Roux is the natural fit, however I really don't want to relearn CMLL ( Maybe I'm just stubborn ). I recently did some LLOB with 3LLL and enjoyed it much more than I thought ( at first glance I too though it was a worse version of Roux ). For me its a good mix between Roux stuff and last layer stuff ( I really wanted to learn some 2GLL for OH ).

I think yes, its kind of wanting to use something less 'mainstream' but also just wanting variety between events. I kind of just stumbled onto this algset and thought, 'Wow ! This would be good for OH', hence the thread. My current OH average of 5 is 35 with Roux and 38 with LLOB ( after only 25 solves ). Originally, I was just going to go with OLL/PLL but after finding these algs I'm not so sure. I'm okay at MU stuff but I think RU just may be better.

Roux's movecount is pretty similar to LLOB + Full ZBLL from my experiments,

With full ZBLL it supposed to be ~5 moves more efficient but I'm not going to learn full ZBLL just to save 5 moves. I think with the 2-gen-ness of this algset the TPS should make up for the worse efficiency.

 

[Tao Yu]

With full ZBLL it supposed to be ~5 moves more efficient but I'm not going to learn full ZBLL just to save 5 moves. I think with the 2-gen-ness of this algset the TPS should make up for the worse efficiency.

Pretty sure this is not true. @AtlasTheGarbage used HARCs to generate 100000 computer solves with each method, and Roux + EOLRb came out to 41.04 moves while LLOB+ZBLL came out to 40.78 moves. However Roux is 2gen while LLOB+ZBLL is not.

 

[Silky]

Pretty sure this is not true. @AtlasTheGarbage used HARCs to generate 100000 computer solves with each method, and Roux + EOLRb came out to 41.04 moves while LLOB+ZBLL came out to 40.78 moves. However Roux is 2gen while LLOB+ZBLL is not.

I was just going off of the Wiki. LLOB is 41-45, Roux 48.

 

[xyzzy]

And just for comparisons sake OCELL/CPLL has 41/46 2-gen algs and COLL/EPLL 12/46 2-gen algs ( I think. Could be closer to 15-16 but that isn't really the point ). For me, bad 2-gen algs are better than bad non-2-gen algs

I agree with this, but it's also… irrelevant? How does it matter how many of the algs are 2-gen? You should be focusing on how frequently you get a 2-gen-able case! It's the same between both methods: OCELL is 100% 2-gen-able and CPLL is 17% 2-gen-able (skip and H perm), versus COLL is 17% 2-gen-able and EPLL is 100% 2-gen-able.

If you want to dig into how much of each alg is 2-gen (as in, the longest 2-gen substring), this starts to depend on the exact algs you use. Something like F R U' R' U R U2 R' U' R U R' U' F' is obviously not 2-gen, but it's still mostly 2-gen.

I don't agree with this. In the case of OCELL vs EPLL, EPLL wins because you have 67% for U-perm where OCELL you'll have, on average, slightly worse cases simply because you have more cases. I don't think OCELL algorithms are really that bad, 2-gen is 2-gen even if they're not 100% the best.

When comparing a two-step method to another two-step method, you have to compare both steps!

If you directly compare OCELL versus COLL and CPLL versus EPLL, you find that OCELL is better than COLL (arguable, actually, but let's just grant this for now) and EPLL is better than CPLL, and obviously you can't just add inequalities that point in opposite directions.

I'd say on average you're getting better OCELL/CPLL cases than COLL/EPLL cases. I'd rather have one really bad case than a bunch of kinda of bad cases because I can just grind out the one until I'm really good at it. Maybe they'll just equal out though.

That's fair. Hadn't really thought about that.

I don't really understand this. You're not working any harder between OLL/PLL, COLL/EPLL, or OCELL/CPLL. More PLL skips is always better than less PLL skips.

You were comparing the 1/72 PLL skip rate to 1/12 CPLL skip rate; you're doing more work by moving from OCLL (corner orientation only) to OCELL (corner orientation and edge permutation). All else equal, more skips are better, but my point is that this "all else equal" premise almost never holds. There's a sort of conservation law here: if you want more skips, you have to do more work elsewhere, partially (or sometimes fully) offsetting the benefit of having increased skip rates.

(This isn't a mathematical theorem or anything. It's just what I've observed from playing with different ways of splitting up steps. It's not literally always true and there are a few counterexamples, e.g. EO+ZBLL for a normal 2-look last layer has the lowest average move count among all classical 2LLL methods, and also has a really high 1/8 skip rate—but you still pay for it by needing nearly five hundred algs.)

 

[Tao Yu]

I was just going off of the Wiki. LLOB is 41-45, Roux 48.

Never trust the wiki for these things. Especially since the numbers can be added by the creators themselves and movecount estimation is often done using different assumptions. The 41-45 on the wiki seems to be based on pretty optimistic block movecounts, plus a 13-14 ZBLL movecount when it's actually closer to 14.5. In turn, the movecount on the Roux page might have been based on more realistic assumptions since we have more reconstructions of actual Roux solves. Remember too that Roux users can get better movecounts when doing OH since they don't turn as fast.

Just comparing the L10P I think the following are realistic movecounts:

ZBroux:
AUF: 0.75
EODFDB: 7.78 (@AtlasTheGarbage's algs on my trainer)
AUF: 0.75
ZBLL: 14.5
AUF: 0.75
= 24.53

Roux:

AUF: 0.75
CMLL: 10
LSE: 13-14 (According to Antonio Kam, a fast Roux OH user)
= 24.75

So there's no way there's a 5 move difference. I would say this probably guarantees Roux to be faster since LSE is 2gen.

 

[antonio19kamh3]

I think A perm is not absolutely disgusting if you use the new alg: x' [R' U': [R U R', D']] and the inverse - not sure if these are widely known yet. I know that Feliks uses it, and Antonio Kam told me he prefers it to the D2 alg. Antonio Kam's E perm alg is also nice and maybe not so well known: F R' F' r U R U' r2 U' R U r f' U' f

It feels rather counter-intuitive to choose LLOB + this method instead of just Roux. Roux's movecount is pretty similar to LLOB + Full ZBLL from my experiments, and you also get a 2gen moveset. I guess you just want to avoid mainstream methods though?

thank you for the e perm alg

 

[Silky]

So I did some number crunching to see the average LL times between COLL/EPLL and OCELL/CPLL.

Preliminary notes. When averaging 2-Gen algs I considered 4.8 TPS and for 3-Gen and up I assumed 3.4 TPS. This is based off of my times which are probably a very bad measure since I'm not great at OH. I also considered rotations as an extra half turn.

OCELL => ~2.6 seconds CPLL => ~3 seconds Total => 5.6 seconds
COLL => ~2.5 seconds EPLL => ~2.6 seconds Total => 5.1 seconds

So at face value COLL is a half second faster.

However, for arguments sake, consider this.

Using ZZ the Gens per step are this =>

ZZ with OCELL/CPLL
EOline/Cross: Multi-Gen
ZZ-F2L: 3-Gen
OCELL => 2-Gen
CPLL => 3-Gen

ZZ with COLL/EPLL
EOline/Cross: Multi-Gen
ZZ-F2L: 3-Gen
COLL => 3-Gen
EPLL => 2-Gen

With LLOB =>

LLOB with OCELL/CPLL
FB: Mutli-Gen
SB: Psuedo-2-Gen/3-Gen
EODB/FD: 2-Gen
OCELL: 2-Gen
CPLL: 3-Gen

LLOB with COLL/EPLL
FB: Mutli-Gen
SB: Psuedo-2-Gen/3-Gen
EODB/FD: 2-Gen
COLL: 3-Gen
EPLL: 2-Gen

So for consideration, the flow of methods with TPS. With OCELL/CPLL you can flow more smoothly from Multi-Gen to Gen-Reduction and then to 3-Gen. This should mean that steps 2/3/4 will have better flow and higher TPS than with COLL/EPLL since this creates more of an awkward gen transition between steps.

But obviously there are a lot more variables to consider and better assumption when considering things like TPS.