L1meDaBestest
Member
Introduction
This is a post aimed at people who are interested in learning algs, specifically large subsets of 1 Look Last Layer (1LLL)(or anyone who just happens to be curious).
Needless to say, 1LLL is a lot of algs (3915) which makes it very challenging to learn and it doesn't help that there aren't a lot of resources for it. I would really love to see a lot more people learning 1LLL because it truly pushes the boundaries of what speedcubers consider possible and I do believe it'll become meta one day. That's why I've made this post, to get more people learning 1LLL by giving the learning process a little bit of structure. This is my attempt to break down the monster and make it seem much more achievable.
I would also like to credit @trangium who helped a lot with this.
Tools and method
This was all done using Trangium's Movecount Coefficient Calculator (MCC), which works by inputing an algorithm, then the program takes into account multiple factors such as move count, type of turn, overworking, regrips, ect and gives the algorithm a score, with a lower score being a better algorithm.
Link: https://trangium.github.io/MovecountCoefficient/
Using this, we can take the 72 (sometimes 40) 1LLL algs in each OLL set and get the average score of all of them which we can use to compare different 1LLL sets. This gives us a ranking of the 1LLL sets with the best algorithms.
However, we can take this further. To figure out which sets are most "worth it" to learn, we can take this average and compare it to the score of the standard OLL alg. By subtracting the standard OLL alg score from the OLL's average 1LLL score, we are given a difference that tells us on average how much slower the 1LLL is compared to the standard OLL alg, with a smaller number meaning the 1LLL alg is very close in speed to the standard OLL alg. After doing this for all algorithms we can rank them to figure out which sets are most worth it (that being the 1LLL sets with the smallest difference).
This can also be used to figure out which sets are not worth learning. If we use ZBLL as a reference, most people learn 5/7 sets, ignoring the sunes and anti-sunes because they are determined to be not worth it. By this logic, any set that scores worse than a sune is not worth it.
Ranking
Here is the sheet which ranks 1LLL sets by speed and the average difference between the OLL and the 1LLL: https://docs.google.com/spreadsheets/d/1pQjqOKxDGQldheg5PcAND7TvSpQb_hgl1rEzaFV88Rs/edit?usp=sharing
Analysis
You'll notice that the 7 sets with the best algs are all ZBLLs, with sune and anti-sune being the top 2 even though its recommended not to learn them. This is why its important to compare the average 1LLL with the standard alg. The 8 sets with the worst algs are all dots. For this reason I would suggest not learning the dots and instead opting for edge control during F2L, perhaps even learning VLS and HLS for the dots. This leaves 42 1LLL sets. 14 of these have a larger difference than anti-sune (the sune that has the lower difference) so they are determined to be not worth learning. This leaves 28 sets or 26*72+2*40=1952 algorithms. Including 5/7 ZBLL (328 algs), that's 2280 1LLL algorithms.
This is definitely the best place to start.
Learning order
There are a few ways to go about learning this. One thing that remains constant however is starting with 5/7 ZBLL. I would highly recommend learning these 328 algs first. For set order it doesn't really matter, if you want something to follow go U, T, L, Pi, H but once again if you want to swap some of those around thats fine (personally I learnt Pi, U, T, L, H so anything works).
Then there are 2 ways to learn the rest, firstly you can just go down the list learning each set in order, or alternatively you can group the 1LLL sets by similar OLL shapes and learn 2 sets at a time (not actually at the same time just one after another). I have made another tab on that spreadsheet which partially groups the 1LLL sets by OLL shape and arranged all of the cases in the best learning order (that being most worth it to least). Which of these you do is up to personal preference (personally I've opted for grouping). Learning the sets starting from the most worth it to least worth it means you'll start having the largest benefits in your solves in the shortest time.
When it comes to learning in general, always learn 1 OLL at a time (1 set of 72). Only learning cases with easy recognition or easy algs will put too much strain on your recognition during solves if you have to think about whether or not you know the 1LLL case for that OLL.
How to learn
There's kind of only one viable way to learn this many algorithms and that's through the use of a trainer, the following is sort of your only option:
https://tao-yu.github.io/Alg-Trainer/1LLL.html
Here's the link to train ZBLL:
tao-yu.github.io
Currently only one person to my knowledge has learnt full 1LLL, that being EDMARTER on Youtube. From what I've heard the system he used to memorise them was learn 144 algs on Sunday then spend the rest of the week reviewing. There are many ways to go about learning so it's up to you what you want to do and I'm sure many variations on this technique are viable. I think there's a large benefit to learning a lot of algs at one time because that way when the OLL case comes up in solves, you'll know that you know the alg for it as opposed to having to check and then not knowing the case.
Some examples include learning 12 or 24 algs per day with 1 review day per week, learning 24 algs per day with 4 review days a week, learning 72 algs over the weekend then reviewing over the week, ect and of course you can throw in some breaks here and there if you need to. When you're just starting out with ZBLL its definitely good to start with something smaller like 12 or 6 algs per day with 1 review day per week, but over time as algs become easier to learn you'll likely find yourself able to take on the more intense learning regiments (this is also dependent on how much time you're willing to dedicate to cubing).
Once again this is flexible so if you find something that works for you use it.
Possible intermediate 1LLL methods
If you wanted, you could decide to learn only the cases with adjacently oriented edges or oppositely oriented edges, then force such a case by inserting the last F2L pair with around 1-2 extra moves on average (this would be used in conjunction with ZBLL so if you had all oriented edges you wouldn't unforce that). You can either learn opp oriented edges for less algs or adj oriented edges for nicer set up moves. You could even do phasing to cut the number of cases in 3, but then you'd have to sort out your own algs since phased 1LLL doesn't exist anywhere to my knowledge, also it isn't as good unless you learn everything. Learning 1 set at a time is much easier to incorporate into your solves.
Extra things
For certain OLL cases its quite common for people to do EO>ZBLL such as F sexy F' for OLL 28 or R U R' U' M' U R U' r' for OLL 39, I haven't taken any of that into account. For now its up to personal preference whether you choose to do that or learn the 1LLL set.
You may want to learn some not worth it cases that have a similar OLL shape to one of the worth it cases to make recognition a little nicer, as you won't see the shape then pause think about whether its the one you learnt the 1LLL for or the mirror. I've marked these on the last tab of the sheet where applicable. Not too big of an issue either way, and its not something you have to decide early on.
On the topic of alg quality. Let's be honest, 1LLL is not optimised. Does that mean you shouldn't learn it? In my opinion, no. The hardest part about learning large alg sets isn't memorising the algs themselves, its the recognition. I think it's worth learning the algs as they are now and getting comfortable recognising them as it doesn't require as much effort to change algs as everyone makes it out to be. Interestingly, the unoptimised algs could be better to learn as they're often a lot simpler, which actually makes execution on larger cubes such as 4x4 and up a lot better, and you'll probably learn them a lot faster too.
Conclusion
1LLL is a huge commitment, and no matter what you do it will take hundreds of hours to learn, but I hope this post has at least given some good direction for anyone crazy enough to embark on this journey. At the very least I also hope to have shown that its not an all or nothing task, if you've finished 5/7 ZBLL, maybe consider picking up 2 or 3 of the sets that are the most worth it and stick with that. At the end of the day "worth it" not only means the speed of the algs themselves, but also the time you're willing to dedicate to learning them.
This is a post aimed at people who are interested in learning algs, specifically large subsets of 1 Look Last Layer (1LLL)(or anyone who just happens to be curious).
Needless to say, 1LLL is a lot of algs (3915) which makes it very challenging to learn and it doesn't help that there aren't a lot of resources for it. I would really love to see a lot more people learning 1LLL because it truly pushes the boundaries of what speedcubers consider possible and I do believe it'll become meta one day. That's why I've made this post, to get more people learning 1LLL by giving the learning process a little bit of structure. This is my attempt to break down the monster and make it seem much more achievable.
I would also like to credit @trangium who helped a lot with this.
Tools and method
This was all done using Trangium's Movecount Coefficient Calculator (MCC), which works by inputing an algorithm, then the program takes into account multiple factors such as move count, type of turn, overworking, regrips, ect and gives the algorithm a score, with a lower score being a better algorithm.
Link: https://trangium.github.io/MovecountCoefficient/
Using this, we can take the 72 (sometimes 40) 1LLL algs in each OLL set and get the average score of all of them which we can use to compare different 1LLL sets. This gives us a ranking of the 1LLL sets with the best algorithms.
However, we can take this further. To figure out which sets are most "worth it" to learn, we can take this average and compare it to the score of the standard OLL alg. By subtracting the standard OLL alg score from the OLL's average 1LLL score, we are given a difference that tells us on average how much slower the 1LLL is compared to the standard OLL alg, with a smaller number meaning the 1LLL alg is very close in speed to the standard OLL alg. After doing this for all algorithms we can rank them to figure out which sets are most worth it (that being the 1LLL sets with the smallest difference).
This can also be used to figure out which sets are not worth learning. If we use ZBLL as a reference, most people learn 5/7 sets, ignoring the sunes and anti-sunes because they are determined to be not worth it. By this logic, any set that scores worse than a sune is not worth it.
Ranking
Here is the sheet which ranks 1LLL sets by speed and the average difference between the OLL and the 1LLL: https://docs.google.com/spreadsheets/d/1pQjqOKxDGQldheg5PcAND7TvSpQb_hgl1rEzaFV88Rs/edit?usp=sharing
Analysis
You'll notice that the 7 sets with the best algs are all ZBLLs, with sune and anti-sune being the top 2 even though its recommended not to learn them. This is why its important to compare the average 1LLL with the standard alg. The 8 sets with the worst algs are all dots. For this reason I would suggest not learning the dots and instead opting for edge control during F2L, perhaps even learning VLS and HLS for the dots. This leaves 42 1LLL sets. 14 of these have a larger difference than anti-sune (the sune that has the lower difference) so they are determined to be not worth learning. This leaves 28 sets or 26*72+2*40=1952 algorithms. Including 5/7 ZBLL (328 algs), that's 2280 1LLL algorithms.
This is definitely the best place to start.
Learning order
There are a few ways to go about learning this. One thing that remains constant however is starting with 5/7 ZBLL. I would highly recommend learning these 328 algs first. For set order it doesn't really matter, if you want something to follow go U, T, L, Pi, H but once again if you want to swap some of those around thats fine (personally I learnt Pi, U, T, L, H so anything works).
Then there are 2 ways to learn the rest, firstly you can just go down the list learning each set in order, or alternatively you can group the 1LLL sets by similar OLL shapes and learn 2 sets at a time (not actually at the same time just one after another). I have made another tab on that spreadsheet which partially groups the 1LLL sets by OLL shape and arranged all of the cases in the best learning order (that being most worth it to least). Which of these you do is up to personal preference (personally I've opted for grouping). Learning the sets starting from the most worth it to least worth it means you'll start having the largest benefits in your solves in the shortest time.
When it comes to learning in general, always learn 1 OLL at a time (1 set of 72). Only learning cases with easy recognition or easy algs will put too much strain on your recognition during solves if you have to think about whether or not you know the 1LLL case for that OLL.
How to learn
There's kind of only one viable way to learn this many algorithms and that's through the use of a trainer, the following is sort of your only option:
https://tao-yu.github.io/Alg-Trainer/1LLL.html
Here's the link to train ZBLL:
Alg Trainer
tao-yu.github.io
Some examples include learning 12 or 24 algs per day with 1 review day per week, learning 24 algs per day with 4 review days a week, learning 72 algs over the weekend then reviewing over the week, ect and of course you can throw in some breaks here and there if you need to. When you're just starting out with ZBLL its definitely good to start with something smaller like 12 or 6 algs per day with 1 review day per week, but over time as algs become easier to learn you'll likely find yourself able to take on the more intense learning regiments (this is also dependent on how much time you're willing to dedicate to cubing).
Once again this is flexible so if you find something that works for you use it.
Possible intermediate 1LLL methods
If you wanted, you could decide to learn only the cases with adjacently oriented edges or oppositely oriented edges, then force such a case by inserting the last F2L pair with around 1-2 extra moves on average (this would be used in conjunction with ZBLL so if you had all oriented edges you wouldn't unforce that). You can either learn opp oriented edges for less algs or adj oriented edges for nicer set up moves. You could even do phasing to cut the number of cases in 3, but then you'd have to sort out your own algs since phased 1LLL doesn't exist anywhere to my knowledge, also it isn't as good unless you learn everything. Learning 1 set at a time is much easier to incorporate into your solves.
Extra things
For certain OLL cases its quite common for people to do EO>ZBLL such as F sexy F' for OLL 28 or R U R' U' M' U R U' r' for OLL 39, I haven't taken any of that into account. For now its up to personal preference whether you choose to do that or learn the 1LLL set.
You may want to learn some not worth it cases that have a similar OLL shape to one of the worth it cases to make recognition a little nicer, as you won't see the shape then pause think about whether its the one you learnt the 1LLL for or the mirror. I've marked these on the last tab of the sheet where applicable. Not too big of an issue either way, and its not something you have to decide early on.
On the topic of alg quality. Let's be honest, 1LLL is not optimised. Does that mean you shouldn't learn it? In my opinion, no. The hardest part about learning large alg sets isn't memorising the algs themselves, its the recognition. I think it's worth learning the algs as they are now and getting comfortable recognising them as it doesn't require as much effort to change algs as everyone makes it out to be. Interestingly, the unoptimised algs could be better to learn as they're often a lot simpler, which actually makes execution on larger cubes such as 4x4 and up a lot better, and you'll probably learn them a lot faster too.
Conclusion
1LLL is a huge commitment, and no matter what you do it will take hundreds of hours to learn, but I hope this post has at least given some good direction for anyone crazy enough to embark on this journey. At the very least I also hope to have shown that its not an all or nothing task, if you've finished 5/7 ZBLL, maybe consider picking up 2 or 3 of the sets that are the most worth it and stick with that. At the end of the day "worth it" not only means the speed of the algs themselves, but also the time you're willing to dedicate to learning them.
- The OLL algs used may not be the algs you use, some may be slower or faster which potentially changes the ranking. I think at the very least the ranking as it is gives a great starting point and is ok to follow.
- The MCC, while being very good, is not perfect. It may occasionally rate an alg better than another when that is not the case, although in my experience when it does it's usually a very slight difference. This will likely be due to the way different movements are weighted in the calculation which may cause it to prefer 1 type of alg over another.
- Using Sune as a baseline for how worth it an algorithm is is somewhat arbitrary. Sunes are also considered bad due to having the worst CP recognition out of all ZBLLs and this post does not take recognition into account, however I still think the primary reason is that the standard algs are so fast, and CP recognition whilst still being tricky is likely a skill issue.
- Technically I also could've decided anything with a score worse than H/Pi ZBLL is not worth it or something else similar. This post is merely one interpretation of this data. The important thing about this post is the ranking, so an individual can work down the list as far as they want to.
- The term "worth it" to begin with is a little strange. Sunes at the highest level are probably worth learning, however the only difference is that you need to put in a lot more effort to get them faster than OLL PLL, significantly more than their TULPiH counterparts. This means despite me saying 14 sets were not worth learning, if someone were to put in the required effort there is a good chance they would become faster than OLL PLL. (and hey if you're already learning 1LLL whats another 1000 hours?)
- Despite having a good amount of experience when it comes to learning algs, obviously I have not learnt 1LLL so take everything I'm saying with an appropriately sized grain of salt.
- The MCC, while being very good, is not perfect. It may occasionally rate an alg better than another when that is not the case, although in my experience when it does it's usually a very slight difference. This will likely be due to the way different movements are weighted in the calculation which may cause it to prefer 1 type of alg over another.
- Using Sune as a baseline for how worth it an algorithm is is somewhat arbitrary. Sunes are also considered bad due to having the worst CP recognition out of all ZBLLs and this post does not take recognition into account, however I still think the primary reason is that the standard algs are so fast, and CP recognition whilst still being tricky is likely a skill issue.
- Technically I also could've decided anything with a score worse than H/Pi ZBLL is not worth it or something else similar. This post is merely one interpretation of this data. The important thing about this post is the ranking, so an individual can work down the list as far as they want to.
- The term "worth it" to begin with is a little strange. Sunes at the highest level are probably worth learning, however the only difference is that you need to put in a lot more effort to get them faster than OLL PLL, significantly more than their TULPiH counterparts. This means despite me saying 14 sets were not worth learning, if someone were to put in the required effort there is a good chance they would become faster than OLL PLL. (and hey if you're already learning 1LLL whats another 1000 hours?)
- Despite having a good amount of experience when it comes to learning algs, obviously I have not learnt 1LLL so take everything I'm saying with an appropriately sized grain of salt.
Last edited:
