• Welcome to the Speedsolving.com, home of the web's largest puzzle community!
    You are currently viewing our forum as a guest which gives you limited access to join discussions and access our other features.

    Registration is fast, simple and absolutely free so please, join our community of 35,000+ people from around the world today!

    If you are already a member, simply login to hide this message and begin participating in the community!

A method of CFOP speedcubing training that yields systematic progress

Joined
Jul 3, 2011
Messages
184
Likes
10
Location
Wales
Thread starter #1
A method of CFOP speedcubing training that yields systematic progress - downloadable

I noticed that the time proportions thread had been bumped, and thought it was time I posted my "speedcubing training system" which is based around a similar idea.

I took hundreds of solves from sub 10 cubers (all world class CFOP people, I'm making the assumption that they are doing it right!), and averaged all the proportions of each, to come up with an overall baseline to shoot for with regard to splits. I then made an excel spreadsheet which dissects your solves and the relative proportions of each phase, then establishes how far from the "ideal" you are in each area.

Im hoping to develop some sort of app or software (help wanted, anyone?) which takes your split times and then gives you specific advice on your weaknesses, hopefully helping to a) lead people toward the type of mindful practice recommended by Escher and 5BLD and b) somewhat curtailing the incessant "what should I work on? "How can I get sub xx?" type threads which clutter the forum.

The "idealised" relative proportions are: cross 12%, cross+1 24.5%, F2L (inc cross) 62%, OLL 16.5% (inc recog and AUF), PLL 21.5% (inc recognition and AUF), Total last layer 38%.

To figure out your own weakness simply multiply your global average (or todays average of 12 if you are inconsistent day to day) by 0.12, 0.245, 0.62 etc, and then do an average of 5 or 12 (more accurate), just doing that specific part of the solve. If your cross (for example) is appreciably more (say 10% more) than 0.12 * your global average then that is the area you need to work on. If your f2l is less than 62% of your solve total, then this is an area of strength. You can work out as a percentage how much out of the "ideal" you are for each phase, and thus decide on your biggest problem.

If you systematically work on the area you are weakest on until it is no longer your weakest area, then move on to practicing your new weakest area, your times plummet. Believe me, I've been training to this system for a couple of weeks now and I have shaved 10% off my solve time, which is a real shot in the arm for my progress, which appeared to have come to a shuddering halt, until this.

The point is it forces you to do the self aware practice that really achieves results, instead of just doing lots of solves and calling it "practice" (there are loads of threads on the folly of this approach.... some wise words on this forum if you look)

The idea of figuring out your worst area and working on it comes from my approach to OLL/PLL training, which is basically:
1) Time all your PLLs,
2) Which ever is the slowest, drill it until its not your slowest
3) Repeat

If this is consistently not working over a long period, review your fingertricks/regrips or get better algs.

The spreadsheet takes your PLL times and highlights which need work (and which are good) using the above system, and extrapolates an estimate of your OLL time, also figures out TPS for each phase of the solve and (most illuminatingly) tells you the metronome setting which corresponds to your F2L time. This then gives you a target for metronome solving since if you can solve to a metronome faster than this number, you are effectively solving faster than your regular F2L. Unless you already have excellent look ahead, I think you will be shocked by how slow this actually is, and if you can't keep up with it, this highlights lookahead as an area to work on. BTW, if you are already solving in perfect proportions and nothing is identified as a weakness, you should work on F2L as it always has the most scope for improvement being the longest part of the solve.

From all this we can also extrapolate sensible estimates of OLL recog, PLL recog, individual pair insertion, first pair identification time etc etc, and I am devising a 4 look, 3 look and Roux versions (although a much smaller sample will be a factor there, basically it will train you to solve with 5BLD proportions, but no bad thing right?)

Top tip - for accuracy on the last layer phases, do your f2l average on 5 or 12 different cubes, leaving each U face down at the end, then to test your last layer, pick them up and turn them over as you start the timer, ensuring that you get no advantage in recognition.

I haven't posted this spreadsheet yet as it still needs tweaking (if enough people want it I'm happy to post it as is), and of course adapting for other methods as I already mentioned. If anyone wants to help me code this into an app or something that would be great. Imagine a touch phone app called "solve analyser" that got you to do a bunch of split times and alg executions and then spit out a prescription for improvement based on a mathematical analysis of your specific weaknesses. It could also have a built in metronome, alg list, 2-look PLL recog trainer, produce graphs of your progress in different areas, etc etc. This would be cool.


To summarise basically all of the generic advice handed out, heres an abbreviated guide to how to improve each given area once you have identified it: (anyone who has been on this forum for a decent length of time has heard all this a thousand times, but here we go again...)

0) Practice

1) Cross - ensure you are doing it on the D face, make sure you know your colour scheme cold, utilise your 15 seconds effectively, make sure plan all four pieces in advance, practice solving cross eyes closed, if 15 seconds isnt long enough, take as long as you like to plan it (this time rapidly comes down with practice), if you can't do it consistently in 8 moves or less, go on JARCS and compare your solutions to a given cross with its optimal solutions, if you still suck at crosses just spend an entire day figuring out cross solutions and executing them with your eyes closed, watch badmephistos cross vids.

2) Cross + 1 - you need to identify your first pair either during the solution of the cross, or ideally during inspection (more advanced, basically to get this down take as long as you need for inspection as before then track the pieces whilst you cross, your times will come down and 6 pieces isnt too much for your brain even if it feels like it at first) so you can immediately go on to insert your first pair after the cross, if you have to watch the cross pieces as you solve them you can't do this so eyes closed cross solving practice removes the need to look. Work on look ahead in general (see F2L). For learning X-Cross check out Chris Hardwicks tutorial and Lars Petrus' block building stuff on their websites, once again taking as long as you like to plan the whole X-Cross and executing eyes closed as before, your required inspection time will rapidly fall, an X-cross of over 11-12 moves is probably not as efficient as it could be.

3) F2L - SLOW DOWN AND LOOK AHEAD, make sure you dont need to look at the pieces when solving, if you do have this problem: find the pieces, close your eyes and solve them, as you get more advanced try identifying two pairs and solving them both eyes closed, try solving to a metronome (after around 200 bpm this is less productive), learn to solve in all slots (Macky's site if you are stuck, I have written about this before: http://www.speedsolving.com/forum/showthread.php?8710-The-One-Answer-Question-Thread&p=720540&viewfull=1#post720540), make sure you have decent algs (8 moves or less) for each case, the only cases which should take more moves should be tricks that can be executed v. fast (RUR'U' * 3 for example), long winded intuitive solutions can usually be optimised by checking the wiki, if you struggle to spot pieces try finding a corner first, minimise cube rotations to 2 or 3 per solve, never do y2, look into edge orientation to reduce this, try forcing yourself to solve slowly without any cube rotations to figure out creative ways to deal with different cases (or check Macky's site), do lots of slow solving trying to minimise move count and keep continuously turning (U,U,U, doesn't count!), practice at a speed that is as fast as you can go without pausing, try not to learn crazy trick ways to solve cases if they interfere with lookahead, go through good cubers reconstructions continuously asking yourself what you would do here and figuring out why they have done what they have done, some people find it it helpful to practice f2l using a cube with no stickers on the U layer (if you are into this I advise getting a coloured plastic Guhong/Zhanchi and an unstickered black on and then you can make a cross only cube, and f2l cube or whatever variation you like by interchanging pieces), watch badmephistos F2L vids.

4) OLL - Learn 1 look OLL, AUF as opposed to rotating the cube, get good algs that suit your turning style, algs that minimise regrips during execution, drill drill drill, maybe learn multiple versions to reduce AUF (if you are crazy into learning algs, or absolutely have to shave that last 0.2 seconds off), learn a bit of basic edge control (I have super simple edge control system tutorial on this to follow at some point), try to predict your OLL case during the last pair insertion, drill some more, work on fingertricking the individual elements of the alg, try OLL recog software if recog is an issue, drill drill drill, watch badmephistos OLL vids, (then investigate alternative algs from the Wiki or Breandans videos or whoever)

5) PLL - Learn 1 look PLL, AUF as opposed to rotating the cube, get good algs that suit your turning style, algs that minimise regrips during execution, drill drill drill, maybe learn multiple versions to reduce AUF, learn to predict what colour will be at the front at the end of the alg so you know what your AUF will be, learn 2-look PLL recog to reduce time identifying the case (some cool new software is floating about on this at the mo), drill some more, fingertrick the algs as much as possible, try PLL recog software if recog is an issue, drill drill drill, watch badmephistos PLL vids, (then investigate alternative algs from the Wiki or Breandans videos or whoever)

6) Practice a lot, specifically targetting your weaknesses. Go through reconstructions to find new ideas, film yourself solving and examine it critically, read the How to Get Faster with the Fridrich Method and How to Practice threads. If you aren't colour neutral it really helps with X-crosses and such, but if you aren't and you are already pretty quick its probably not worth switching as the time could be better used elsewhere, try solving opposite colour cross (e.g. yellow if you are white) as this doubles your options and isnt as difficult as full colour neutrality, but if its not for you dont worry about it, learn a bit of COLL, WV, VHLS, RLS etc, some cases are really worth knowing, play with different methods as there is a lot of helpful crossover (Petrus/ZZ/Roux teaches you a lot about edge orientation, block building, using restricted movesets etc), continuously set specific achievable goals, and work to achieve them.

7) Practice.


I think thats basically everything, but Im in the process of writing a book on this stuff (200 pages in..... ) so further contributions are welcome, and I expect theres going to be a lot of TLDR here......

Spreadsheet available further in the thread, along with blank f2l case documents:
http://www.speedsolving.com/forum/showthread.php?39406-A-method-of-CFOP-speedcubing-training-that-yields-systematic-progress&p=805512&viewfull=1#post805512

There has been some discussion of having a weighted PLL calculation based on frequency of cases occurring, a spreadsheet which calculates this can be found here: http://www.speedsolving.com/forum/showthread.php?39406-A-method-of-CFOP-speedcubing-training-that-yields-systematic-progress&p=824093&viewfull=1#post824093
 
Last edited:
Joined
May 6, 2012
Messages
367
Likes
12
Location
Pullman, WA
WCA
2013BRAN01
#4
I'd also like to encourage this. Of course people with different finger skills will end up with different ratios. And if you're trying to improve your *average* solve time, it makes sense to spend more time practicing the OLL and PLLs that you need more often. But just reducing the longest times is a start.
 
Joined
Feb 28, 2011
Messages
336
Likes
1
Location
Over Here
WCA
2011UNKR01
#6
You should be hired for writing those insanely long web pages about products where they are like BUT THATS NOT ALL. Just saying.

But seriously, this could be potentially quite helpful, any new cubers asking how to lower their times could be sent to a more organized thread/website detailing the final version of what you are working on. I'm all for seeing the documents as soon as you can.
 
Joined
Jul 23, 2008
Messages
3,373
Likes
72
WCA
2008KINN01
YouTube
RowanKinneavy
#8
I've got to say I would like to an 'improvement' sub-forum, where the stickies would be composed of essentially theory posts like this one (haven't thought enough about what you've said to comment properly just yet), and people can post their breakdowns and videos asking for specific advice. Although the crux of this post seems to be about analysing those breakdowns and setting targets, times will always just be a symptom of what you're doing wrong, separate from underlying causes; specific advice will always be needed.
 
Joined
Nov 28, 2009
Messages
1,795
Likes
2
WCA
2010LACH01
YouTube
fatboyxpc
#9
specific advice will always be needed.
Yep, this is my exact issue. I've known for quite some time that my times obviously aren't in order, however, I'm not always sure how to improve on certain things (learning lookahead just seems so difficult for me). To be fair, I don't really practice near as much as I should, but I think a lot of that is due to not being able to improve.
 
Joined
Mar 24, 2012
Messages
1,543
Likes
25
Location
Vancouver, Canada
WCA
2011HUTN01
YouTube
yoshinator879
#10
Two things: first of all, this is amazing, excellent concept, good writing, awesome. Second, there will never cease to be people who start their own threads about how they can improve, no matter how easily accessible you make threads like these.

Thank you!
 
Joined
Jul 3, 2011
Messages
184
Likes
10
Location
Wales
Thread starter #11
Thank you all for your kind comments. I would like to address the issue of people requiring specific advice. I am not in any way suggesting that any automated system could possibly rival the extraordinarily helpful advice people are occasionally getting here, such as: http://www.speedsolving.com/forum/showthread.php?39351-Cross-or-F2L-Help Dominate, this is some truly outstanding work.

However, I do feel that a big problem for a lot of people is identifying where their weaknesses actually lie. A great number of mediocre speedsolvers just solve over and over, and then get confused when their progress stalls. They are then in the position of having to identify precisely what their difficulty is without having the tools to do this. I have hopefully provided a methodology which will help people to overcome this obstacle.

A huge number of "how can I improve?" type threads are posted, and many are responded to in the same basic way, "Practice", "Go slow and look ahead", "Plan your full cross and execute it with eyes closed" etc etc. Although the uninitiated need this information (as we all did at one point) it does become repetitive noise to the more experienced cuber. If people used a tool such as the one I have proposed, these threads would be uneccessary, a simple link to the appropriate place would suffice. When they are really stuck (sub-15 stuck), posters such as Dominate can step in with the really useful nuts and bolts stuff. My list 0-7 is basically a reduction of the cookie cutter responses that people tend to receive when they ask for help, especially when they provide inadequate information for people to be truly helpful and clearly haven't done sufficient independent work. An example: http://www.speedsolving.com/forum/showthread.php?37173-Help-getting-faster-please . Even when they provide splits and a video the answers seem to fall into the same basic concepts. I defy you to find an over 25 second solve submitted for community analysis where the biggest issue isn't lack of lookahead or poor cross planning or similar basic skills. People post because there isn't a framework or methodology that they can turn to, the closest thing to this being the excellent How to Get Faster using the Fridrich CFOP Method (I found this extremely helpful starting out). I hope to move this situation forward by encouraging people to follow a similar progression, but allowing for the fact that different people progress at different rates in different areas. The skill people must learn is to be aware of their own limitations and weaknesses and thus how to address them. As Escher et al have explained many times before, this is the key to effective, rapid progress.

If we can get people into good practice habits from the point where they transition from the beginner method into a true speedsolving method, all the way through past the sub-20ish barrier (which I firmly believe is the point at which a lack of a practice methodology stalls progress) and through to sub-15 (and beyond) in a systematic fashion, then surely this enhances the whole community WITHOUT replacing concrete specific advice as exemplified by Dominate?

In conclusion, yes there is no substitute for real specific advice, but there are a whole lot of people who haven't got past the point where they need generalised, cookie cutter advice. My proposition is simply another tool to help people improve, but (if I may be so bold) one I feel could (if systematically followed) lead to extremely rapid gains, the development of good habits and the reduction in pointless threads where the only appropriate response is "do a search" and "practice". Which annoy me.

Take from it what you will, and thanks again for all your kind words. I have more stuff like this. I have been systematically reading and digesting this forum and not really posting much for years now, its time I gave something back (before any pedants jump in, I was reading it long before I actually created an account!). More to follow.
 
Last edited:
Joined
Mar 23, 2011
Messages
243
Likes
1
Location
Belfast
YouTube
MalusDB91
#12
I like your outlook on all of this speedpicker. Professional. It's encouraging to see people addressing developmental issues for cubers instead of falling back on the old shout outs like you mentioned. Keep her lit! lol
 
Joined
Oct 6, 2012
Messages
77
Likes
0
Location
Belgium
#13
I think that I first going to learn al PLL's and then practise this methode...
I find it a good idea to analyse your weakness on this way :cool:
 
Joined
Apr 14, 2011
Messages
3,179
Likes
16
Location
England
WCA
2011LAUA01
YouTube
5BLD
#14
Very good ideas. The one quibble I have is the fact that the breakdown percentage of fast cubers may not e the best for you personally-- as you get faster and faster you have less and less people to serve as examples and you think more personally (this is a good thing).

I think the most important thing you mention is finding specific faults and drilling them until they are gone. Personally I can't really be bothered anymore but if I wanted to still have extremely fast progress I'd as you say look at each alg and if its weak drill it until it's not. However for me, I've worked on technique for such a while that my major bad habits are gone and so simply solving casually to get more comfortable with the cube generally suffices as good practice for me. I know my CMLL is awful though so when I have the motivation that is my priority of practice.

When i say "casual solving" though i mean just maintaining speed. Any solve slower than average or not faster than average, from experience I can reflect on it and tell exactly what to avoid doing or what technique to change next solve.
 
Last edited:
Joined
Jul 3, 2011
Messages
184
Likes
10
Location
Wales
Thread starter #15
Very good ideas. The one quibble I have is the fact that the breakdown percentage of fast cubers may not e the best for you personally-- as you get faster and faster you have less and less people to serve as examples and you think more personally (this is a good thing).
This is a very good point. At your level (and method), there are few role models to serve as examples. However I would hold that for anyone to get to (say) sub10, maybe sub-12, they must have the ability to recognise their own flaws and systematically address them. You made this point here http://www.speedsolving.com/forum/showthread.php?39242-Please-help-me-become-sub-10&p=798614&viewfull=1#post798614 .

I want to help people get this skill. My methodology isnt for people trying to get from sub 9 to sub 8 (although I dont doubt it would work better than no methodology at all), but rather for people who haven't yet developed the ability to "think more personally" with the necessary realism to effectively self-critique, so basically anyone who doesn't know what practice to do to improve (or thinks their cross/f2l/whatever isn't a problem when in fact it is).

Unless someone is already awesome, they can do a lot worse than try to model solves on Felix, Matts, Cornelius et al. If someone is already world class, then they have already found their own path.

And I really doubt your CMLL is "awful" ;)
 
Last edited:
Joined
Sep 5, 2009
Messages
278
Likes
3
Location
Seattle
WCA
2007VUNI01
#17
As a behavior analyst, I approve of and am very excited about this approach!

I would love to have something like this for ZZ, but I realize the sample size is small and there is a wide variety of last layer methods.

For the app, it certainly wouldn't be the first feature you add, but it would be interesting to collect data from the users. You might discover unexpected trends that could further optimize the learning process. For example, maybe it's more efficient to use 4LLL until your sub-30; maybe if you favor both hands equally (as opposed to right or left dominate) your solve proportions would be different, maybe getting super good at cross would help improve the other steps faster. Rich data collection would begin to address the personalization problem, but also guide the general learning procedure. Again, just brainstorming; definitely not what needs to be focused on to get the app off the ground.

Prisma Timer is open source, and he already has LL scramblers and the like. Perhaps a java guru could hack on that instead of starting from scratch.
 
Joined
Jan 6, 2010
Messages
540
Likes
0
Location
Phoenix
#18
i think that this is a very effective method. its one that naturally one would take after a while to become faster. the problem that i have is that percentages would work for someone who already knew 2LLL. a lot of people who are trying to break that sub30 or sub20 or even some sub15 barrier dont have all those algs memorized. therefore those percentages would be skewed. but, none the less the best way to get better at CFOP is to memorize your OLL and PLL and then optimize your individual parts of the solve. i like how you introduced the cross+1. this came to me after i was already sub 20, but helped me break sub15. this is very underrated part of the solve in which a lot of time is wasted. the others being cross movecount/speed optimal and PLL recognition. this is just from my personal experience.
 
Joined
Jul 3, 2011
Messages
184
Likes
10
Location
Wales
Thread starter #19
rickvu2: The idea of collecting data and continuous refinement of the "improvement curve" is very attractive. I can imagine some difficulties with filtering the data but fundamentally a great idea. From previous experience of training people, it is a roughly exponential decay of solve time vs how long you have been cubing for. A good example is the graph on BadMephistos site of his own improvement. One would expect a couple of blips (first learning f2l for example), but the issue I see coming up is confirming how much practice people are actually doing. Very few people do a set amount per day/week/month, and since the practice time (and quality of practice time) is a key variable, I think the data would be useful only from truly dedicated cubers, who it would be tricky to identify in advance.

Prisma idea is interesting. I use prisma myself and really rate it. I like the idea of "solve analysis" being another option along with f2l/oll/pll training. The user is then taken through a sequence of averages of different phases of the solve, and results are presented for interpretation. This seems eminently doable. Any java gurus want to lend a hand with this?


maggot: Thank you for your positive feedback. The point about 2 look OLL or PLL is a good one, and I see two solutions. I could have a different splits for 2 look / 3 look / 4 look / maybe even beginner method, and thus separate analyses for each. The user would input the method they are using and it would have splits for the method in question.

However, in trials with real people (thank you guys at WSMO 2012 who were kind enough to subject themselves to this) I found the interesting side effect of the system that it does give you an indication when you should consider learning more algs. One person in particular (I forget his name), received his results (reasonable cross, good f2l, weakish OLL, very poor PLL) and simply said: "I suppose I'd better get round to learning 1 look PLL then. Thought as much.". The fact was, he had done enough work on his f2l to bring his times down, that further efforts in this area would be less productive than learning some new algs. Slower 4 look cubers had their f2l highlighted as a weakness, indicating that they would get better results from working on their f2l rather than learning all the 3 or 2 look algs at that time since their last layer was acceptably quick (if not move efficient). I am all for the "learn what you like whenever you fancy it approach", but I also like the idea that the system indicates to people when in their development they would most benefit from learning algs rather than other skills. Just helps with the whole "I average 32 seconds, should I learn 3 look last layer, or wait until Im sub-30" type questions.

I throw it open to the community then. Would you prefer a separate 2/3/4 look version of this thing, or do you like the idea that it indicates when would be a good idea to change up to the next level?
 
Last edited:
Top