[Help Thread] Square-1 Discussion

[Sam N]

since I simply can't keep up with the tps of you young whippersnappers.

haha.

Looks like EO+CP too me not PBL (Permute Both Layers). Also check my webpage for more Sq stuff.

yeah, I don't know what I was thinking. Multitasking and answering a question are not a good combination. I was referring to EO+CP not PBL.

With regards to PBL, here is a link to all the PLL- style cases on square-1.

http://www.mediafire.com/view/hwurvnnp6dh7na5/Square-1+Full+PLL+-+Raul+Low+Beattie.pdf

 

[bobthegiraffemonkey]

I was hoping people would be developing on top of CS+P by now, but nobody else has even learned it as far as I know. Is there anything I could do to excourage more people to try? I've already got plans for better videos talking over the cases as well as some example solves, but actually finding the time is difficult.

 

[TMarshall]

I was hoping people would be developing on top of CS+P by now, but nobody else has even learned it as far as I know. Is there anything I could do to excourage more people to try? I've already got plans for better videos talking over the cases as well as some example solves, but actually finding the time is difficult.

I would definitely be interested in learning some of it. Correct me if I'm wrong, but you'd have to have somewhat good memo techniques to do it in under 15 seconds right? I haven't looked into your method a whole lot, but I seem to remember seeing something on counting the number of cycles you have. I would definitely have to work on that a lot to get it under 15 seconds, considering how bad I am at 3bld memo.

It's nice to see that my post caused a good bit of discussion, even though I had a lot of incorrect information in my original post. Hopefully someone would be willing to learn this in full (I'll definitely talk to Sophie Chan about this next comp)

 

[Sam N]

I was hoping people would be developing on top of CS+P by now, but nobody else has even learned it as far as I know. Is there anything I could do to excourage more people to try? I've already got plans for better videos talking over the cases as well as some example solves, but actually finding the time is difficult.

The biggest barrier seems to be understanding how to identify parity in cube shape and the methodology behind that. If you could create a video series that explains that in a clear fashion, I think that would help a lot. Your previous video series was okay, but CSP is not the easiest topic to learn starting off. If there is a way you can introduce it in a clearer fashion, that might help a lot. That was the hardest part for me when I learned it. I found that it was difficult understanding how it worked, and how you identify CSP. The algorithms behind it are quite easy once you understand the basic idea of how parity is made, so I never found those to be intimidating. Once you understand the ways parity can be made, and map out the shapes that can create it, you basically get all the algs for free.

 

[bobthegiraffemonkey]

I would definitely be interested in learning some of it. Correct me if I'm wrong, but you'd have to have somewhat good memo techniques to do it in under 15 seconds right? I haven't looked into your method a whole lot, but I seem to remember seeing something on counting the number of cycles you have. I would definitely have to work on that a lot to get it under 15 seconds, considering how bad I am at 3bld memo.

It's important to realise that you don't need to memo, you just need to know if the number or targets is even or odd. This is much easier than actually doing BLD memo, and several people can do that sub-10, though it still takes some practice to sub-15 CS+P inspection.

I'd develop more stuff built around this myself, but I'm just not interested enough. I'd maybe be tempted to figure out OBL control, but I don't ever see myself using OBL since I don't even use Vandenbergh.

 

[bobthegiraffemonkey]

The biggest barrier seems to be understanding how to identify parity in cube shape and the methodology behind that. If you could create a video series that explains that in a clear fashion, I think that would help a lot.

Noted, I thought this might be part of the issue. I'll try to get around to this in the next few weeks.

 

[willtri4]

I was hoping people would be developing on top of CS+P by now, but nobody else has even learned it as far as I know. Is there anything I could do to excourage more people to try? I've already got plans for better videos talking over the cases as well as some example solves, but actually finding the time is difficult.

I've tried learning, but the spreadsheet makes no sense without the videos, and your accent is incomprehensible.

 

[TMarshall]

It's important to realise that you don't need to memo, you just need to know if the number or targets is even or odd. This is much easier than actually doing BLD memo, and several people can do that sub-10, though it still takes some practice to sub-15 CS+P inspection.

I'd develop more stuff built around this myself, but I'm just not interested enough. I'd maybe be tempted to figure out OBL control, but I don't ever see myself using OBL since I don't even use Vandenbergh.

Thank you for the clarification. I'll see if I can grasp the concept of it by next week, so that I can start learning the cases. I just skimmed through some of the videos and the document, but I think the document is a bit hard to decipher. I'll try to look through it a bit more in depth later.

 

[blade740]

Thoughts on the vandenbergh method

While I think vandenbergh method is capable of sub 10 global averages, I do think it will have its limitations for times closer to sub 9. Here are some thoughts on how it could be improved. First off, the next reasonable step would be to combine two of the steps (CS, CO, EO, CP, EP) while EP and CP would probably be the best place to combine, the number of algs makes this close to impossible in my mind. The next place to combine, EO and CP would be a lot of algs, but nowhere near as many algs as PBL (I think). Next, combining CO and EO is just known as AO, and I think that is the most reasonable place to combine. While it is ~180 algs, most of the algs are very short. However, from my experience with this, it can be somewhat hard to recognize. So, I think it would be possible to do CS+AO in one look, by using something similar to what Mike Hughey did for sq1 bld, just knowing where the pieces end up and so being able to do CS,Co, and EO in one look. Back to EP and CP, one thing that I noticed I was doing intuitively was after seeing a certain PLL case so many times, you start to generally know what EP case you'll have (I'm sure all of the faster sq1 solvers already do this) Therefore, I would just finish this off and learn what PLL case yields what EP case (and the angle at which you get the PLL so that you know how to AUF/ADF) so that you can one look CP and EP. If you were able to do all of the things mentioned, you could solve the sq1 in 2 looks every time. However, I think this can still be improved. By using CS+parity, you could probably drop ~1 second on your times. However, you would have to know how 180 (plus mirrors) CS algs affect the corners and edges so you could do AO. However, even if you eliminate parity that early, there are still some PBL cases that if you just do normal CP+EP, are downright nasty (stuff like M-Ca, which I think will give you W-O for EP) Therefore, for the really nasty cases, you could learn some PBL algs for it. This would probably only be about 20 PBLs, but you could always learn more.

In conclusion, I think the best vandenbergh variant would be CS+parity and AO in one look, and then CP+EP or PBL. However, this would also be quite a lot of algs and you would have to put a massive amount of time and effort just to be able to learn all the algs/cases for this method. I think the alg count would be somewhere around 260 algs, plus figuring out how 180 CS cases effect your pieces. On top of that, you'd have to learn how to tell if you have parity during CS, but I don't think that is too bad.(I would ask Jabari if he'd be willing to learn all these algs, but he already told me that he hates Sq1 ) I think this is doable, though because if you learned one alg or CS case a day, which is very doable, you would learn this full method in about 16 months. Also, a lot of you probably know lots of non-parity EPs already, and so you could reduce the alg count greatly.

I think you're seriously underestimating how difficult it is to track permutation through Cubeshape. Compared to CS+OBL, learning PBL is a walk in the park. And the way you simply brush off CS-parity recognition as "not too bad" is insane.

I think the "ideal" method would be CS+parity -> OBL -> PBL. With those three steps (so you're only worried about even parity PBL), I'd consider CS+parity to be by far the most difficult.

 

[TMarshall]

I think you're seriously underestimating how difficult it is to track permutation through Cubeshape. Compared to CS+OBL, learning PBL is a walk in the park. And the way you simply brush off CS-parity recognition as "not too bad" is insane.

I think the "ideal" method would be CS+parity -> OBL -> PBL. With those three steps (so you're only worried about even parity PBL), I'd consider CS+parity to be by far the most difficult.

Im not saying that you track the permutation: I'm saying you use a system like Mike Hughey's sq1 bld system so that you know where the pieces end up after cubeshape. It seems like you read it as you learn algs for all the cs+OBL cases, but what I'm saying is that you learn 2 algs for every cubeshape case (1 for parity and 1 for no parity) and then you figure out how each of those algs effects the rest of the pieces. Then, once you get to your solve, you would inspect to see if you have parity, figure out which of your 2 cubeshape algs you're gonna use, and then recall how that alg affects you're pieces so that you can know the OBL case.

And yes, I agree that I may have made cs+parity seem too easy. However,this is supposed to be the ' ultimate ' variation of the Vandenbergh method. It's supposed to push the limits of how fast you can solve it in every aspect (inspection, alg learning, etc.). The only reason I didn't just do cs, OBL, and PBL is because of the alg count for it. Correct me if I'm wrong, but i think PBL is well over 1000 algs. Plus, if you were ever able to fully use the variation that I proposed, you could solve the cube in 2 looks as opposed to 3. All in all, I highly doubt anyone will ever use my method in full, just because it would take so much time and effort to get good with it.

 

[Berd]

I was hoping people would be developing on top of CS+P by now, but nobody else has even learned it as far as I know. Is there anything I could do to excourage more people to try? I've already got plans for better videos talking over the cases as well as some example solves, but actually finding the time is difficult.

I was looking at it today, I still need to learn full cubeshape haha.

 

[bobthegiraffemonkey]

I've tried learning, but the spreadsheet makes no sense without the videos, and your accent is incomprehensible.

I thought I had a link in my sig to a pastebin which is better as a standalone reference, but it isn't there. Will fix later, for now clicky. It can still be improved I think, feedback/questions would be good. I can't do much about having a Scottish accent though ...

 

[TMarshall]

I thought I had a link in my sig to a pastebin which is better as a standalone reference, but it isn't there. Will fix later, for now clicky. It can still be improved I think, feedback/questions would be good. I can't do much about having a Scottish accent though ...

I'll start looking at the videos this week,so I'll give you feedback sometime soon.

 

[not_kevin]

Just idly doing some thinking solves - for easy cubeshapes (<= 3 twists), I can generally trace cubeshape + obl in around 30 seconds. Because of this, I think that CS+OBL is actually something doable in inspection, given quite a bit of time investment. My thoughts here are that one can (similar to Ty's suggestion on using something like Mike's sq1 bld cubeshape idea) fairly easily learn how to track what pieces go where for cubeshape - this is even "easier" than a fullblown version, since learning all the <=4 twist cases is only something like 30 total cases, and tracking a single slice or two should be easy enough to go with [given my above experiments]. Since you only need to care about OBL, you end up ignoring a lot of information that would otherwise be necessary for CSP.

Given this, I think a reasonable 2.5/3-look Square-1 method is CS+OBL -> PBL (CPP -> EP). I think this is good, because in my experience, the OBL -> CP transition is inherently worse than the EO -> CP transition, so the generally-cited downside of "extra time of inspection" for CPP is actually minimized a good deal. Plus, since you have to slow down anyway for the PBL recognition, like with CPP before, you get the easy benefit of having immediate use for any PBLs you _do_ learn - like learning COLLs when you already predict the CPLL after doing your normal OLL alg.

As confirmation of a previous point, full PBL does indeed take thousands of algs, 'tho only a small number of thousands - since there are 43 PLL cases with mirrors/inverses excluding the solved case, there are 44^2 - 1 = 1935 total PBLs. If you can guarantee that it's even parity, then it drops down by half, to 967, but that's still a bunch - somewhere between full ZBLL and full 1LLL, as a point of comparison (mirrors/inverses are generally much easier to learn on 3x3, because they flow better; Square-1 doesn't have similar luxuries, a lot of the time).

However, I think that CSP will become a much more attractive method given more research into it, and I hope that someone can figure out a good way to do that sub-12. I think CSP+OBL -> PBL is basically only an FMC technique at this point, and will remain so for at least a couple years, but I hope I'm proven wrong. Being able to 2-look a solve definitely feels pretty awesome, from experience

 

[Hssandwich]

If you can guarantee that it's even parity, then it drops down by half, to 967, but that's still a bunch - somewhere between full ZBLL and full 1LLL, as a point of comparison (mirrors/inverses are generally much easier to learn on 3x3, because they flow better; Square-1 doesn't have similar luxuries, a lot of the time).

I would say that square-1 does have 'mirrors' though they are just executed from the back, similarly to sune and back sune. E.G. / 3 / -3,-3 / 0,3 / and / -3 / 3,3 / 0,-3 /. Those two are mirror algs, but one is from the front and the other from the back. These algs, where possibly less fingertrickable, are just as easy to learn.

 

[not_kevin]

I would say that square-1 does have 'mirrors' though they are just executed from the back, similarly to sune and back sune. E.G. / 3 / -3,-3 / 0,3 / and / -3 / 3,3 / 0,-3 /. Those two are mirror algs, but one is from the front and the other from the back. These algs, where possibly less fingertrickable, are just as easy to learn.

I'll agree that there are definitely mirrors (I wouldn't use that particular case, since they really solve the same PBL [like, I wouldn't consider R' U R' U' R' U' R' U R U R2 a different case from R2 U R U R' U' R' U' R' U R', since they're both the same U perm] - I'd compare things like / -3,0 / 3,3 / 0,-3 / with 1,-1 / -3,0 / 3,3 / 0,-3 / -1,1), but in general, there are much fewer mirrors because of the fixed orientation that Square-1 forces.

 

[Sam N]

Just idly doing some thinking solves - for easy cubeshapes (<= 3 twists), I can generally trace cubeshape + obl in around 30 seconds. Because of this, I think that CS+OBL is actually something doable in inspection, given quite a bit of time investment. My thoughts here are that one can (similar to Ty's suggestion on using something like Mike's sq1 bld cubeshape idea) fairly easily learn how to track what pieces go where for cubeshape - this is even "easier" than a fullblown version, since learning all the <=4 twist cases is only something like 30 total cases, and tracking a single slice or two should be easy enough to go with [given my above experiments]. Since you only need to care about OBL, you end up ignoring a lot of information that would otherwise be necessary for CSP.

Given this, I think a reasonable 2.5/3-look Square-1 method is CS+OBL -> PBL (CPP -> EP). I think this is good, because in my experience, the OBL -> CP transition is inherently worse than the EO -> CP transition, so the generally-cited downside of "extra time of inspection" for CPP is actually minimized a good deal. Plus, since you have to slow down anyway for the PBL recognition, like with CPP before, you get the easy benefit of having immediate use for any PBLs you _do_ learn - like learning COLLs when you already predict the CPLL after doing your normal OLL alg.

As confirmation of a previous point, full PBL does indeed take thousands of algs, 'tho only a small number of thousands - since there are 43 PLL cases with mirrors/inverses excluding the solved case, there are 44^2 - 1 = 1935 total PBLs. If you can guarantee that it's even parity, then it drops down by half, to 967, but that's still a bunch - somewhere between full ZBLL and full 1LLL, as a point of comparison (mirrors/inverses are generally much easier to learn on 3x3, because they flow better; Square-1 doesn't have similar luxuries, a lot of the time).

However, I think that CSP will become a much more attractive method given more research into it, and I hope that someone can figure out a good way to do that sub-12. I think CSP+OBL -> PBL is basically only an FMC technique at this point, and will remain so for at least a couple years, but I hope I'm proven wrong. Being able to 2-look a solve definitely feels pretty awesome, from experience

CSP is a very solid method. Given unlimited inspection I've been able to get sub 12 averages with it. The main issue is getting inspection down from personal experience. I'm not a great blind solver so that might attribute to this, but it seems possible to get it under 15 seconds. I feel that in the future someone will be able to get it down and use it in a competition and be quite successful with it. I know Bobthegirraffemonkey has been able to do this in competition before.

 

[Hssandwich]

I'll agree that there are definitely mirrors (I wouldn't use that particular case, since they really solve the same PBL [like, I wouldn't consider R' U R' U' R' U' R' U R U R2 a different case from R2 U R U R' U' R' U' R' U R', since they're both the same U perm] - I'd compare things like / -3,0 / 3,3 / 0,-3 / with 1,-1 / -3,0 / 3,3 / 0,-3 / -1,1), but in general, there are much fewer mirrors because of the fixed orientation that Square-1 forces.

Oh yeah, I didn't really think before posting that example I forgot what the alg did.

 

[not_kevin]

CSP is a very solid method. Given unlimited inspection I've been able to get sub 12 averages with it. The main issue is getting inspection down from personal experience. I'm not a great blind solver so that might attribute to this, but it seems possible to get it under 15 seconds. I feel that in the future someone will be able to get it down and use it in a competition and be quite successful with it. I know Bobthegirraffemonkey has been able to do this in competition before.

Sorry, I wasn't clear in my post - I'm in no way doubting the power of CSP. Of all of the places parity is dealt with, it's clear (at least, in my head) that CSP is by far the best place to do it. However, I'm still skeptical in its consistent sub-15 ability in inspection, especially since the biggest payoff, I imagine, would be to guarantee all bad cubeshapes to have no parity (eg, if it's a cubeshape skip, I'd probably just use my 15 seconds to get OBL + figure parity, and solve parity during CPP, rather than doing 6 twists just to deal with parity, and then have relatively little info on how to continue the solve). Plus, I'm imagining that CSP crowds out the ability to really use anything else in inspection, since simply tracing the cycles, without thinking about what that does to the solve other than parity, likely takes a good chunk of time.

To clarify, then, I think there are 2 strong/doable/etc. 3-look Square-1 methods: CSP -> OBL -> PBL and CS+OBL -> CPP -> EP. Of these two, I think the latter is easier, and (very vitally, I think) is easier to transition into a 2-look method - based on my musings above, I think combining CPP with EP into full PBL is easier than combining CSP with OBL recognition. But again, I hope I'm proven wrong, because as I said before, CSP is by far the most interesting and high-potential still-in-development method, to me, even more than PBL (sorry Lid). I'd love to see a day where people can actually realistically 2-look solves in competition, unlike my current ~0.3% ability to do that (a cubeshape case I can predict to OBL, an OBL case I know [aka, not one of the 4 opp/opp cases I don't know], a PBL I know [I know about 10% or so]).

Oh yeah, I didn't really think before posting that example I forgot what the alg did.

Don't worry - you made your point

 

[bobthegiraffemonkey]

However, I'm still skeptical in its consistent sub-15 ability in inspection

Timed some inspection just to see how well it would go, 11.88 AO12:
1. 10.04
2. 12.89
3. 13.38
4. 12.10
5. (DNF(11.83))
6. 12.74
7. 11.48
8. 12.67
9. (8.78)
10. 11.59
11. 9.34
12. 12.56

Cubeshape wasn't always particularly fast because I suck at some cases, the DNF is when I messed up parity from rushing too much, and obviously it will take a little longer to do in official solves. However, I can definitely do this faster with more practice, so overall I'm convinced it can be done. Whether you can recognise anything else at the some time I don't know for certain - because I've never tried looking for anything else in inspection, but I'd guess at least CO prediction would be doable.