[Feature] Opening Trainer

[pnodet]

Hi !

A few days ago I stumbled on this open-source opening trainer: studyopenings.com The source code is hosted on GitHub: github.com/jven/studyopenings

This tool as some flaws but I believe the idea is great! The goal is to help people memorizing their openings.

The studies on Lichess are an incredible tool and we can see that many of them are dedicated to opening theory. So I feel like an opening trainer would be a nice feature that people would use.

It can be useful for beginners focused on one opening with only the main lines or intermediate players with a few more variations, etc.

How I think (imho) it should work:

Build one repertoire:

• Let people choose their color.
• Let people choose one or different moves for the opposite side. Preparing different variations.
• Only allow on move for the player.
• Handle transposition.
• Let people add a line of move from a study or a game.

Study your repertoire:

• Computer plays randomly one line.
• Player has to play the 'right' move : ie. the move they chose in the repertoire.
• If player tries another move it could display hints.

EDIT:

• Forum post where this idea is thoroughly discussed: lichess/forum/suggestion-opening-trainer

[pedro-pedrosa]

I'd reply in the forum thread but it looks like it has been archived?

anyway, I like this idea, and to some extent this can be achieved through studies using interactive lessons, although obviously it's not very user friendly for this purpose, ie you'd have to create a chapter for each variation and you would have to click a different chapter to get the next line to practice.

I think if we want to make this feature a reality we could try to extract smaller, more reusable features from it and slowly implement them until we can use interactive lesson studies for this purpose.

example set of features:

• have the computer follow variant lines randomly instead of only following the main line
• improve the above by following lines depending on % of games in the db
• improve the above by storing failed attempts and implementing spaced repetition
• implement flashcards from random positions found in the study
• implement PGN import into an existing move tree

the above alone would be amazing not only for this feature but for other uses as well like endgames or middlegame theory

[YodaEmbedding]

Other useful things:

• Learning how to punish non-standard moves or opening blunders. These moves could be pulled from the Lichess opening book for some rating range.
• Similar to above, finding good traps or suboptimal lines that players commonly fall into by looking at %es in the openings database.

[pnodet]

I'd reply in the forum thread but it looks like it has been archived?

Yes indeed.

I think if we want to make this feature a reality we could try to extract smaller, more reusable features from it and slowly implement them until we can use interactive lesson studies for this purpose.

So would it be better to build the feature one by one and then use some of the to make an opening trainer ?

[pedro-pedrosa]

I think "opening trainer" is a use case, not exactly a feature. you can practice openings using studies + a new set of functionalities that would benefit not only the "opening trainer" use case but also others.

I believe dividing the use case into a set of smaller bite-sized features that appeal a broader audience will increase the chances of it getting implemented.

if you have a look at this other issue I have created (in which I mention this issue) it requests a feature that would make practicing openings a possibility. from there it's just improvements to it, no need to create a special "opening trainer" functionality that works a lot like studies and interactive lessons.

[ni784]

Simillar to #8582

[aimorris]

From #8582

[ghost]

Basically https://www.chessable.com/movetrainer but on lichess

[Alix-Martin]

I think a spaced repetition mechanism is necessary but we can do it different/better than chessable. When using chessable I feel I spend too much time learning moves that are so deep in the tree that it is unlikely to reach them, either because they are unlikely in a game or because I go wrong before. So we should spend time learning the "trunk" moves before the leaves.

Mathematically we should order repertoire moves by probability of occurring in a game times probability of getting them wrong.

And the second term depends on history of attempts on the move and time since last successful attempt.

[purefan]

I just made a new issue related to this topic but more focused on the implementation, please feel free to comment https://github.com/lichess-org/lila/issues/14619