• 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 40,000+ people from around the world today!

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

New Speedcubing Training Software

Goldsmith

Member
Joined
Apr 8, 2019
Messages
10
https://www.sc-trainer.com

Hi everyone!

I'm a long time casual speedcuber (averaging a bit over 20 seconds), and have always really enjoyed the hobby. I also loved seeing this really awesome community expand from afar, with the amazing Youtube channels that have come up over the last years and cool videos sharing the awesome atmosphere at competitions. Now there's even a Netflix documentary!

I had an idea for a new approach to an algorithm trainer a few years ago though, and being a software engineer by trade I tinkered a bit with a low quality prototype for my own use in my spare-time. I have recently been blessed with the chance to work full time on whatever projects I'd like for some time though, and decided to give this idea a shot, and I just finished a high quality minimum version of what I'd envisioned! It's available at https://www.sc-trainer.com, and is designed to work well on all devices, computers/tablets/smartphones.

A little preview:

Screenshot from 2021-04-15 11-40-59.png

Before I explain a bit more about the idea behind it, I just wanted to ask you if you would take the time to try it out and provide me with some feedback? I have both included "Give feedback" buttons in between algorithm tests which link to a Google Form, but I'm also happy to receive any feedback or discussions about the potential of the app here in replies. I'd also love to hear if I've missed similar tools that already exist that do what I want to create, as I don't know of any right now!

So my overall goals with this web app is to make algorithm practice have more flow, be more efficient, maybe also a bit more gamified/fun, and also give you more insights into your strengths and weaknesses, through better data than current tools allow. As of this early version it is mostly (hopefully) satisfying the flow part, and is basically a full revision tool for PLL with 2-sided recognition, as it only has the single mode of reviewing all algorithms from all angles.

My main innovation (at least as far as I know it's an innovation), is to attempt to remove the scrambling phase from algorithm trainers which is the only method I currently know of. Instead of scrambling before every new algorithm to make your cube show the case, you instead start with a solved cube, see the case on screen, apply the moves to the cube that would solve the case on the screen, and then verify that your cube looks as it should (the expected look is shown to you). If you got it correct you can just continue straight on without applying any moves to your cube. If you got it wrong, you will have to apply moves though, to solve your cube again, as the app no longer knows the state of your cube. I don't see a good solution to this without integrating with bluetooth cubes which I don't imagine supporting anytime soon. However, I do have soon coming solutions for intelligently ordering algorithms that should avoid giving cases you're not ready to try yet, which I'll explain more about in the next paragraphs. This should help minimize the amount of wrongs. In addition, when you know the whole alg set, and are just practicing to improve your recognition + execution time, then incorrect applications should be infrequent, thereby allowing for much more flow than the usual method of scrambling before every case.

I believe this new approach allows for several nice improvements to algorithm practice:
  1. As already mentioned it allows for a lot more flow when practicing alg sets you infrequently get wrong, with the goal of gaining speed
  2. It accurately times your recognition time, which is a more complete data point as this is an important factor in a real solve. You can of course look away while scrambling with a traditional alg trainer, and start the time before looking, but here it is forced
  3. It forces you to include pre-AUF and post-AUF in your time, and the trainer deliberately randomizes between all the AUF combinations, this will currently allow you to notice any outliers that need more practice, and has big future potential for data-driven insights.
  4. It incentivizes you to improve at quickly predicting all the AUFs without any U-moves or cube rotations, and executing them fluently together with the algorithm as it impacts your time
As I still can't actually do full 2-sided PLL recognition, the app is funnily rather unusable for me at the moment as I'm lost on about 2/3 of the cases haha, but that also incentivizes me heavily to get my next planned feature completed, which is very important to me:

I want to add a spaced repetition algorithm to the app. This will basically allow the software to track your current knowledge and skill at all the cases, and intelligently choose which ones to show you, to optimize your learning. When you still don't know all the cases, it'll make sure to limit how many weak cases are in the rotation at any given time, so you don't get overwhelmed with new cases, and automatically add in new ones as you learn more, while intelligently giving you old ones to revise when it's optimal for retention. When you know all the cases it will then help you practice your weaker/slower cases more often, to boost your overall ability at the algorithm set.

This is the feature that will make it a web app I'm really excited to use, and will hopefully be very helpful to myself and many others. The reason I'm already sharing it here now though, is to get some feedback in case I made some wrong assumptions about how helpful this type of tool could be, as I believe the current state of it makes the potential quite clear, even if it isn't fully featured yet. With this I hopefully won't go blindly down the wrong path of features.

Lastly, some other features I dream of making in the future are:
  1. More algorithm sets should be easy. Including OLL, maybe COLL/EG-1/EG-2 for 2x2, or different Roux sets. One could also do advanced sets like Winter Variation / ZBLL etc.
  2. I am quite interested in improving from my current very rookie level in blindfolded solving, so am excited to hopefully make tools for that to both practice and really improve recall on letter pairs, improving tracing, and also a 3-style trainer
  3. Provide detailed statistics of your strengths and weaknesses in different algorithm sets. Highlighting things like weak algorithms, but also weak AUF combinations etc.
  4. Otherwise I'm super excited, though it may be a bit of time before I can get there, to do some quite advanced practice tools, surrounding F2L and Cross(+1). I have some vague ideas for how one could make an F2L trainer, to not only practice cases, but also different tools to help with improving lookahead. For the cross I'm really excited to create a cube solver that takes into account ergonomics (how easy the execution/finger tricks is) kind of similar to the algorithm JPerm created to evaluate how good an algorithm is. So it wouldn't just solve for shortest move sequence, but it would be able to intelligently guide you to speed-optimal solutions, and maybe also be able to take into account human-spottable patterns in inspection, such as blocks, and teaching you how to take advantage of those. In all these tools I'm very motivated to find more human-friendly software-assisted practice methods.
  5. All of the above is a massive amount of work, but in a dream scenario where this gets a lot of traction and I can maybe also get help from other software engineers, I also have fun ideas like creating badges/achievements (you learned full PLL / You average under 2 seconds for PLL! etc.), creating some social features, that could maybe mimick what Strava is to runners/cyclists/triathletes, maybe create leaderboards for the different training tools, maybe crunch the algorithmic data from users to calculate what the best algorithms / finger-tricks are for different algorithm sets, and many other fun ideas :). That's all very far fetched though haha!

If you made it this far then thank you a lot for reading through all my thoughts and dreams for what speedcubing practice tools could achieve, and I hope you'll give my app a test-drive and provide me some good feedback!

And here it is once again: https://www.sc-trainer.com
 

abunickabhi

Member
Joined
Jan 9, 2014
Messages
2,564
Location
Yo
WCA
2013GHOD01
YouTube
Visit Channel
Hello,

Just tried out your application. It was quite good for the first version. Good work!
I also filled up the google feedback form. I hope more people try out the tool you have made.

I would love if you make a tool on 3-style trainer or letter pair trainer in the future. You have already mentioned it on your to-do list. Also blind algorithms are a lot, and even if they are intuitive they take time to get used to. So, I think the spaced repetition feature that is on your to-do list will be most useful for blind alg feature tools.

Social feature for cubers is still not out there. With the advent of bluetooth cubes a few years back, there are some leaderboard where you can post your time directly, and also you can share your recon via the app instantly because the bluetooth sensor records the moves done. It will take a lot of commercial and technological effort to reach the level of Strava app.

Even I do not like the scrambling phase while drilling algs, thats why I use keyboard cubes to train, I have tried qcube and tao yu's trainer.

My personal feature request will be a tool, to share algs and reconstruction with others. Right now, we just message a person the alg.cubing or cubedb.net link, but it will be better if it can be done insight, and without the added process of copying and pasting links.

Good luck!
 

Goldsmith

Member
Joined
Apr 8, 2019
Messages
10
Hello,

Just tried out your application. It was quite good for the first version. Good work!
I also filled up the google feedback form. I hope more people try out the tool you have made.

I would love if you make a tool on 3-style trainer or letter pair trainer in the future. You have already mentioned it on your to-do list. Also blind algorithms are a lot, and even if they are intuitive they take time to get used to. So, I think the spaced repetition feature that is on your to-do list will be most useful for blind alg feature tools.

Social feature for cubers is still not out there. With the advent of bluetooth cubes a few years back, there are some leaderboard where you can post your time directly, and also you can share your recon via the app instantly because the bluetooth sensor records the moves done. It will take a lot of commercial and technological effort to reach the level of Strava app.

Even I do not like the scrambling phase while drilling algs, thats why I use keyboard cubes to train, I have tried qcube and tao yu's trainer.

My personal feature request will be a tool, to share algs and reconstruction with others. Right now, we just message a person the alg.cubing or cubedb.net link, but it will be better if it can be done insight, and without the added process of copying and pasting links.

Good luck!
That's really good feedback! Thank you @abunickabhi! I especially love hearing that this tool could possibly allow you to switch away from keyboard cubes to physical cubes for alg practice, that would be a huge win if it could solve such a significant pain point for you!

To respond to a few of your points:

  • Yes I'm also really excited about the BLD features. As you say 3-style is intuitive, but because you're holding so much in your head, remembering letter pairs, and then having to figure out 3-style algs, anything that you can build more into muscle memory instead of having to use intuition will benefit all the other parts too as you can spend more of your effort there. And definitely spaced repetition would be super helpful for such a big alg set. I know some people are interested in learning 5-style algorithms too, and there it definitely would be haha, as there's several thousands of algorithms if I remember correctly? :p
  • Yep would no doubt take a humongous effort to add in significant social features, and would also increase server costs by a lot I'm sure. That would need a team and not just a single person like me, so not at all sure that will happen, and if it does definitely not soon, but let's see!
  • Interesting feature request. What I have already thought about, is that when I get to the stage where there's login and user features, the app will be gathering information about what algorithms people use, also just for the sake of making some tools work at all they would need that information (for example to be able to show the expected state after an OLL algorithm, I need to know which algorithm they use as there are many different OLL algorithms that result in different permutations, as one could say they are just different ZBLLs). In that case one could have a profile page that displays the algorithms one uses for example. Would that for example be something you'd be excited about?

    For reconstructions and displaying moves in general I must say I feel like alg.cubing and cubedb.net already solve that quite well don't you think? I'm not sure exactly what it is you're looking for by having it inside, is it that you'd be able to have a list of reconstructions in your user page that people can go see from some of your best solves for example? Because if it's specific reconstructions from something you just did, I feel like you'd have to share a link through a chat if you're sharing it with someone no matter what right? Or maybe I completely misunderstood you, I'd love to hear you clarify in that case
 

abunickabhi

Member
Joined
Jan 9, 2014
Messages
2,564
Location
Yo
WCA
2013GHOD01
YouTube
Visit Channel
I know some people are interested in learning 5-style algorithms too, and there it definitely would be haha, as there's several thousands of algorithms if I remember correctly? :p
Yes there are about 126,720 cases for edges on a 3x3. For such a large number spaced repetition will be required. The second biggest algset in cubing right now after that is 1LLL which is about 4k cases. Megaminx OLLs, PLLs, and full ZBLL follow after that.

In that case one could have a profile page that displays the algorithms one uses for example. Would that for example be something you'd be excited about?

For reconstructions and displaying moves in general I must say I feel like alg.cubing and cubedb.net already solve that quite well don't you think? I'm not sure exactly what it is you're looking for by having it inside, is it that you'd be able to have a list of reconstructions in your user page that people can go see from some of your best solves for example? Because if it's specific reconstructions from something you just did, I feel like you'd have to share a link through a chat if you're sharing it with someone no matter what right? Or maybe I completely misunderstood you, I'd love to hear you clarify in that case

Yes that would be something I would be excited about. Recently I made a thread on this particular topic, https://www.speedsolving.com/threads/better-way-of-maintaining-personal-reconstructions.79746/, the main improvement I seek, is that we do not have to copy paste links, and all the alg cubing and cubedb solutions will be stored in one repo. My idea was to have this in a site like speedcubedb.
Yeah but having such features require login and sharing, and it will cost a bit of server space.
 

Goldsmith

Member
Joined
Apr 8, 2019
Messages
10
Yes there are about 126,720 cases for edges on a 3x3

Haha that's just wild, that would take a lifetime to learn, no? :p

... A google search later I can confirm it would take approximately 4 lifetimes if one learned one a day from birth xD

Yes that would be something I would be excited about. Recently I made a thread on this particular topic, https://www.speedsolving.com/threads/better-way-of-maintaining-personal-reconstructions.79746/, the main improvement I seek, is that we do not have to copy paste links, and all the alg cubing and cubedb solutions will be stored in one repo. My idea was to have this in a site like speedcubedb.

Okay great! That gives me a lot of understanding into what you were looking for. And I see you're a fellow Chess player from your Lichess reference too :D.

So as I mentioned above I think that the profile that includes which algorithms you use seems to definitely fit within what my app is trying to achieve, though it's not a top priority feature, but I think it could happen as a medium priority feature as I mentioned above that there are other reasons to gather the data of which algorithms people use for the trainers.

The reconstructions I don't see as being as much within the scope, unless at least we can reuse the software from one of the tools you mentioned, so maybe one could on the profile have a series of links to reconstructions displayed, and one could even embed the reconstruction software through an iframe (basically just means displaying a website inside another website, that's for example how the youtube video player shows up on other websites). It probably isn't something I would prioritize implementing myself as it seems somewhat tangential to the idea of the app, but it's definitely something I would be open to other programmers to contribute to the site if they wanted to implement it and we found a good way to include it from a user experience design point of view :)
 

abunickabhi

Member
Joined
Jan 9, 2014
Messages
2,564
Location
Yo
WCA
2013GHOD01
YouTube
Visit Channel
A google search later I can confirm it would take approximately 4 lifetimes if one learned one a day from birth xD
It will take a bit lesser time, as the algset becomes only 30k after mirrors and inverses reduction. Also, its possible to learn a lot of algs at once using memory techniques, or just drilling. My guess is a couple of years. I started in 2017, and right now I am nearing completion. Although fluency and recall of these algorithm is still something that needs to be trained, and its an ongoing process.

Okay great! That gives me a lot of understanding into what you were looking for. And I see you're a fellow Chess player from your Lichess reference too.

Yes I am chess player. Actually it was bad chess results in 2013 over the board tournament, that prompted me to take a break from chess, and pursue speedcubing. Since the pandemic, I have become a lichess addict, playing all time controls and all variants.

I agree with your point about personal reconstructions. It is a feature that is not on high priority and using iframe, might be the solution for viewing recons on profile page.
 
Top