Let's make alternatives to speedcubedb and cubedb

GPT-4 shows some promise for analysing 3x3 solves, although it is currently not fine-tuneable. I think that we should make a database of solve critiqes, because it likely will gain that feature in the future. GPT-3 is also fine tuneable, however it is less advanced, but could still work.
 
the open source alg database now has a visualization for each alg (activate by pressing the play button next to the alg) 👀

1681452880023.png
 
GPT-4 shows some promise for analysing 3x3 solves, although it is currently not fine-tuneable. I think that we should make a database of solve critiqes, because it likely will gain that feature in the future. GPT-3 is also fine tuneable, however it is less advanced, but could still work.
How I remember cubedb/speedcubedb (relatively new to cubing so not sure which is which) is it showed how it would solve a certain part of the cube. I don’t know if there are any web-based APIs that we could make requests to for an optimal solver to solve one position into another specific position, we could use that.
 
FYI the site is on a different repo than the database.
Some thoughts:
Maybe YAML is easier to human-write than JSON. I don't think you are writing it manually, right?
The website or the database should list HTM, STM (or even my wacky 1.5HTM). If this is included in the database, data integrity checks need to be in place.
The most important measure for an alg should be how many upvotes it has, so good algs are rewarded. Or you can manually set algs as good or bad.
 
How I remember cubedb/speedcubedb (relatively new to cubing so not sure which is which) is it showed how it would solve a certain part of the cube. I don’t know if there are any web-based APIs that we could make requests to for an optimal solver to solve one position into another specific position, we could use that.
I'm not sure if those exist. But maybe the algorithm database for better F2L/LL algs/solutions?
 
Some thoughts:
Maybe YAML is easier to human-write than JSON. I don't think you are writing it manually, right?
The website or the database should list HTM, STM (or even my wacky 1.5HTM). If this is included in the database, data integrity checks need to be in place.
The most important measure for an alg should be how many upvotes it has, so good algs are rewarded. Or you can manually set algs as good or bad.
1. I chose to use json so the code would be easy to write. Javascript can handle json very easily, and yaml not so much
2. Thank you for the suggestion to add movecounts, I can definitely put it on my todo list. I can write a script that computes the movecounts
3. A voting system is a relatively complex feature, and I would consider it if enough people were interested
 
  • Like
Reactions: qwr
Are there plans to add a setup alg or the cases? I can try to help with that maybe, but in the meantime, I've opened an issue in the repo for that.
 
The algorithm trainer has setup algs. Does that cover what you had in mind?

If I'm understanding your question correctly, cubingapp has it but in a different area: cubingapp.com/train.html
Well I was just thinking for learning algs maybe there could be a setup alg above like this:
1692596358796.png

Excuse the really bad diagram, but above the algs would be the alg that would make the case, is that doable?
 
Well I was just thinking for learning algs maybe there could be a setup alg above like this:
View attachment 26017
Excuse the really bad diagram, but above the algs would be the alg that would make the case, is that doable?
The best way to memorize and learn algs is the trainer 😊

The algorithm database is meant to be a quick way to find any algorithm
 
Back
Top