A method of CFOP speedcubing training that yields systematic progress

I would really like a PLL trainer that gave me the right ratios of the various possible PLL situations and kept track of my times on each situation separately. I think this would give me much less error in my statistics.

I would think such a tool would be useful for getting a better estimate on how fast someone is with their PLLs for the purpose ofusing this speedcubing training method.

I'll do some split timings tommorow and post them for analysis. I'm sub-40 now; got 28 seconds earlier today!!! (28 is my PB!)

CarlBrannen said:

I would really like a PLL trainer that gave me the right ratios of the various possible PLL situations and kept track of my times on each situation separately. I think this would give me much less error in my statistics.

I would think such a tool would be useful for getting a better estimate on how fast someone is with their PLLs for the purpose ofusing this speedcubing training method.

Click to expand...

OK, here you go, an excel spreadsheet that takes your PLL times and provides two averages. The first is a regular average, the second is weighted for the correct ratios of how often each case comes up (PLL skip excepted). So if your H perms (uncommon, 1/71 cases excluding solved) are poor, but your Us (very common, 4/71 cases for each) are great, the weighted average will be lower than your regular average as the H-perms are given less weight in the average. Vice versa, and your weighted average will be higher than the regular. If you wish to use the weighted average in the main spreadsheet, just type over the "overall PLL average" field where the regular average is given. Hope this helps

P.S. Let me know if you need further data, like for example the specific weighted times of each case rather than the overall average.

Speedcuber50: I look forward to your splits, make sure you download the spreadsheet so you can keep track of your improvements in the future, and once you post I will be happy to offer as much help as I can.

Interesting difference between PLL recognition/AUF and PLL execution. I wonder if it is misleading since I didn´t use any G-perms. When they came up I did 2look PLL. Guess I should redo it when I have learned the G-perms decently.

Thanks speedpicker but I've already got that included in the software. I should have been more clear. What I need are the ratios of the times that humans get in the 22 cases.

For instance, I need a table that looks something like this:

-- 1_lk 2_look
Aa 1.20 1.20

Ab 1.20 1.20
Ua 1.40 1.40
Ub 1.40 1.40
Ra 2.00 3.22
Rb 2.00 3.22
...

Of course these numbers will depend on the cuber. I expect that the really fast cubers will have pretty similar numbers and that rank beginners will be a lot more spread out. So I'm guessing that the numbers from the hot shots are the ones I need.

Right now I've got software which takes into account the probabilities of the various PLL cases and computes the average. It also computes how much each of your PLL cases contributes to your average time.

For me, I use a 2-look algorithm for the T and Na/Nb. My longest time is on the Na/Nb so at first glance I should improve the Na/Nb. But the T is 4x more likely to show up so the T contributes much more to my average PLL time (about 6% of my time doing PLLs is spent doing Ts versus around 2% for Na and Nb). So on 2nd glance my times will improve more if I work on the T.

But the T is a fairly long algorithm. So on 3rd glance, maybe there's another algorithm I can learn/improve where there is more room for improvement. But to know how much room there is for improvement I need the times of the fast cubers.

When I say "room for improvement" I mean relative to my own finger speeds. So if my Ua is much slower than my Aa, while the winners get them done in about the same time, I'm assuming I need to improve my Ua more than my Aa.

OK, so I took the executions of a fast (almost sub-1 PLL) cuber and stuck them in a spreadsheet. These are no recognition, execution only times. I then figured out which combination of 2-look PLLs solved each one look case, and then got a time for the two look case by adding the total execution time of the relevant 1-look cases.

I am assuming we are talking about Badmephisto style two look, with Aa and E for corners, Uz, Ub, H and Z for edges. Aa is always executed headlights on B, and the E-perm can be executed from any orientation where you dont see headlights. I have assumed that for the actual E-perm case, the cuber has the correct orientation, so arent solving E-perm with E-perm followed by H perm...

For example, a T perm is solved in two look by executing an Aa (headlights at B), then following up with a Z perm, so the total time would be Aa + Z.

This suffers from a couple of crippling assumptions, which just go to show that learning one look is just better. It does not take into account the time taken for identifying the second case, or the time taken to AUF to get the second case into its correct orientation. This is an appreciable amount of time.

May I recommend just learning 1-look PLL? Even excluding the extra time required for AUF and identification, you are looking at an average saving of 0.42 seconds per case, and thats also assuming you can execute each of the six 2-look algs in a second or less.

i hope this is what you are after. None of these averages are weighted in any way, but you can see the relative execution times for 1 vs 2-look PLL for a very fast cuber with no AUF or identification time whatsoever.

Oh dear! All my split attemps today have resulted in nothing! How do you do a split timing without spoiling lookahead???

You can use Prisma Puzzle Timer to set up the cube directly for OLL/PLL cases.

I did all crosses, then all cross + one pair, all full F2L and so on.

speedpicker, that's awesome! Exactly what I want.

One of the things that bothers me about his style of 2-look is that I've already memorized Ab, so doing Ab = Aa+H seems like a waste.

And I am going to learn 1-look. Having these numbers will help the motivation.

speedcuber50 said:

Oh dear! All my split attemps today have resulted in nothing! How do you do a split timing without spoiling lookahead???

Click to expand...

TP is right, the splits should not interfere with look ahead. Cross, cross + 1 and full f2l should be done from a completely scrambled cube, you are just stopping at a given point in the solve and recording the time.

As regards full last layer, OLL and PLL, the best way is probably to get multiple cubes (say 5), do the ful f2l on each from a different scramble (recording split times for full f2l) and then place them U face down on the table, maybe even mix them round, find the lady style. When you then come to do your last layer times, just pick up a cube as you start the timer, then solve last layer, stop timer. Doing it in this way keeps the time taken for recognition and AUF as part of the overall time.

To do OLL and PLL singly, do the same thing (F2L complete, face down, mix round) then pick up and time OLL, replace cube U face down (PLL is incomplete). Do this for all 5 cubes and you have your OLL time. Then mix them around and do the same thing for PLL. This will include recognition and AUF also.

TP:- Your big problems are cross and PLL. For cross, try doing some work on solving the cross with your eyes closed after as much inspection as is required. Times of 4 seconds plus tell me that you are solving a couple of pieces, then finding the ones you missed then solving those on the fly, rather than having the whole thing planned out. If you cant plan all four and place them eyes closed, plan three and get good at that, so the only thing you need to do in addition is find the fourth piece and insert it. Then try to predict where that fourth piece will end up. Then its a short step to controlling where that piece ends up so its got good orientation and is easy to place, to eventually placing it as part of the planning. Executing eyes closed is essential though, as you dont want to maintain the crutch of needing to stare at the pieces.

You can then reduce your movecount by comparing your solutions to Jarcs or Prisma Timers cross solver. If you are say three or four moves higher than something more optimal, try and go through the optimal solution without physically moving the pieces, so that you can see the interaction mentally rather than passively witnessing the physical results. You should rapidly get to the point where you can plan all four pieces, execute the cross without looking, and use your eyes during cross execution to track your first corner edge pair.

If you are having difficulty executing the cross quickly once its planned out, then your issue is with non-standard move orders. By this I mean that 2-gen algs and well thought out OLLs/PLLs are learned since they are (relatively) easy for the fingers to accomplish. You may find that the cross solution you come up with is just not finger friendly at all, and you dont want this to hinder you (bear in mind a very fingertrickable cross solution that is one or two moves longer than a very unfriendly one is fine since you will execute it quicker. Try to use wide turns during cross, this will often eliminate extreme finger unfriendliness) . To train yourself to execute disconnected random moves, try timing how long it takes you to apply a scramble and improve that time. This will improve the ergonomic efficiency with which you handle the situations often being presented to you in cross, and you will find that it works wonders for training cross execution. Then your only barrier is how well you plan.

As regards PLL, you are right that using a 2-look alg set will seriously affect these numbers. If you are two-looking, then there is twice as much recognition and twice as much AUF time. If your PLL executions are 3 seconds, and you have to do two of these to get PLL done, thats six seconds already, add in a 1 second per alg for AUF and recognition and you end up with an 8 second PLL! I reckon that why your PLL recog and AUF is so high, its because you are essentially doing it twice.

On the flip side though, you are probably using two look OLL also, and your times for that are fine. I wonder if your PLL executions are a bit awkward? It is a common problem that people get much quicker times when just timing their PLLs in isolation than when they do them as part of a solve. There are a number of reasons for this:

1) Incorrect timing techniques
2) Taking the best rather than an average time
3) Not including time taken for regripping
4) Doing an alg once under pressure of a solve situation is very different to executing an alg for the 20th time in a row trying to get a good average execution time.

Examine the way in which you are executing these cases and compare your fingertricking with that of expert cubers. Faz, Cornelius, Chris Olson, Erik etc all have PLL execution videos out there. You dont have to copy them entirely, but perhaps they will show you ways in which you can reduce uneccessary cube rotations or regrips (probably regrips is your major issue).

And maybe its time you just learned full one look PLL? If its only the G perms, dont fear them, they are all related to one another and have a similar "feel" during execution. People seem to put learning these off, but really its not that bad. Headlights left, find the two by one block and start by splitting it up.

Learn them in pairs. Gb is the inverse of Ga, so you can apply Ga to a solved cube to end up with Gb (verifying you got it right), then solve Gb back to a solved state. Since each is the inverse of the other you will find they go together in your muscle memory, like Aa and Ab do. The other two are kind of the same moves but done to the opposite side of the cube (reflected in a line from UL to UR), so once you learn two, muscle memory helps with learning the others. You'll be amazed at how quickly you pick them up. Despite the rotation, they have nice flow.

Your F2L is taking a long time. Also work on that. Again, executing pairs with eyes closed is a great exercise. Spot the pair on the cube, close eyes, solve, open, repeat until all four done. This is an excellent way to internalise the relative movement of the pieces during F2L, and tremendously improves fluency. You are probably pausing multiple times during F2L, perhaps during the execution of an alg. If you can reduce the number of pauses your times will start to fall rapidly, and this exercise is a great way to start that.

Also consider metronome practice like I have discussed erlier in the thread. Start at 60 bpm and try to do one move per click. This helps you gain lookahead (further reducing pauses) and further internalises the movement of the pieces, since it forces you to go slow.

If you have any really long F2L cases (over 8 moves), find a better way from the wiki, and if you arent already, try to solve in other slots, further reducing cube rotations. Anything you can do in the right hand you should try to do in your left, which means you can insert into FL as well as FR, and some simple rear insertions (basic unmatched and matched pairs into BR and BL) will also save you a lot of rotations. Dont go trying to learn an alg for every case from every orientation at this stage, that would be counterproductive. Just try and expand your intuitive solving to include other slots than FR, make sure you are utilising empty slots where possible, and be conscious of how much you pause and rotate. Lay a solid foundation with metronome and eyes closed solving, and you will find that further improvement comes quickly, once you have gotten over this "brain active" phase. Its liek driving a car, at the moment you are going "check mirror, indicator, clutch, change gear, foot back to where it was, wheel turn, accelerator pedal.... etc" instead of just "turn left". Its the same with your pairs. Once you internalise the moves so its just "place that one, place that one, place that one" your only issue is finding them and movecount. Finding them = lookahead, which you will have started improvement on by metronome solving, and movecount is a check the wiki for difficult cases situation. May I offer a document in this respect which may assist? It can be found here:

http://www.speedsolving.com/forum/showthread.php?6085-How-to-Get-Faster-using-the-Fridrich-CFOP-Method&p=824176&viewfull=1#post824176



Hope this helps, any more questions about what I have written, dont hesitate to ask

I use similar splits for calculating progress, but starting at last pair. So last pair+OLL, last pair+OLL+PLL. Starting at last pair means I'm more accurately factoring in recognition time.

I interleave the measurements rather than do all of one type of split together, so will scramble a cube, time cross+1, solve the 2nd and 3rd pair, then time last pair+OLL. Basically so I'm not scrambling all day.

Brest includes LSLL data in his reconstructions. Its definitely an excellent idea.

can u make an OH spreadsheet too

timeless said:

can u make an OH spreadsheet too

Click to expand...

Spreadsheet for bigger cubes would as well would be interesting.

Are we asking for a OH 2-look vs 4-look spreadsheet?

And a second one for 2-look vs 4-look for larger cubes?

speedpicker said:

Are we asking for a OH 2-look vs 4-look spreadsheet?

And a second one for 2-look vs 4-look for larger cubes?

Click to expand...

i guess the optimal % time it takes to do each step

Oh right. That makes more sense. This will be more difficult as we have Redux vs Yau, different edge pairing methods, and so on. However, I will look into it and release something if it seems like a feasible project. OH at the top level is also not entirely straight forward, as there are some very fast CFOP, ZZ and Roux cubers, all of which will vary considerably, and edge control is a bigger factor. Also there is a considerably smaller pool of reconstructed solved to look at, so this may skew the data also.

Still, its by no means impossible, and Brests reconstructions allow me to go through all this, so maybe I'll get into it. Leave it with me, but it will have to wait until after my current OLL and fingertricking project comes out, and after that Im doing something similar for F2L and PLL. For those anticipating the OLL survey, its up to 120 pages and is looking really good.

can you share the WR spreadsheets using this formats

http://www.speedsolving.com/forum/attachment.php?attachmentid=2527&d=1354739412

i want to see their rec/ steps time

OK, heres an example. Felix Zemdegs average of 12 (comp averages of 5 are obviously not as good as a complete avg of 12), 7.23 seconds.

Theres a little bit of guesswork as regards PLL and OLL execution averages, as far as I can tell Felix has not published his execution times for all his algs, so I have arbitrarily given him a 1.1 second PLL average, which is arguably a little slower than reality. Also his X-cross skills made the cross time an educated guess also, but is probably pretty accurate, as I have observed similar cross times from Felix in other solves. Anyway, heres a first shot:



Note the absolutely ninja cross times (partially an illustration of the benefits of fluent colour neutrality in reducing cross movecount), the high level cross + 1 markup, indicating superior lookahead, and the almost non existent OLL recognition (same reason).

The PLL recog/AUF is possibly a little high (all I have is total last layer data, and the aforementioned educated PLL guess), but it does fit, as you can see that the overall last layer is within a hair of the predicted, the issue is with the precise splits due to lack of data. I could do some frame counting and get a better idea, maybe another time.

Note that this whole exercise is a little futile, as it was solves like Felix's that actually generated these figures in the first place, so we are in danger of comparing Felix with Felix, which is pointless.

The blank F2L layouts from a few pages back reminded me that I did a similar thing with the idea of collecting algorithms that top cubers have used. Even got inspiration for the layout from the same website lol.

If anyone is interested this is it. (Haven't worked on it in a while so not complete yet + right now it is mostly Feliks algs)
https://docs.google.com/document/d/1R84sciLA1ED2aOGHvQvpfTFSBJ0-DiZUfNeJ-xV6rZs/edit