lichess.org
Donate

Suggestions for improving Tactics Training

I shall pose three suggestions for improvement of Tactics Training: one which I think would be awesome but a bit laborious to implement, and two that are easy to implement but maybe not so important.

Sometimes, the best response to my move is not the one that is hardest for me to answer. Look here:
en.lichess.org/training/61293
In this position, I spotted Nb5, but then the computer answered e5 and it was easy to get Bb4.
On the other hand, the most challenging line (to me) would be a simple Nxf1. After Nd6, Kc7, I would blunder with Rxb7 instead of e5, which is (for me) hard to spot.

I know that lichess administrators can't spend their lives curating tactics training and it is done automatically, but maybe we could do this?

Suggestion 1: Add an option to suggest a response that is harder to spot.

Or:
Suggestion 1': Try to make the computer find the longest line until clear advantage instead of the best line.

Or:
Suggestion 1'': Introduce a "study mode" - the player has to answer all possible responses (without trivial ones, i.e. requiring just one more move to clarify the situation?), up to the point when the advantage is clear.

Two other issues was already addressed in the forums, I think.

Suggestion 2: Do not penalize winning moves, even if they are much worse than the best move. Tell the player to find a better move.

Suggestion 3: Indicate length of a shortest mating sequence. Maybe make it an option.
I understand what you mean that the response suggested as best by the computer is not necessarily the most challenging for the puzzle solver, but I think it's hard to implement since there is lots of subjectivity involved and each puzzle has to be examined in detail by the creator making it a very tedious work.

Todo_pro: I know it often says "it is a good move, find a better one", but I anyway found problems when winning moves were penalized (I analyzed the poitions with stockfish later). Don't have an example noted down though... Of course, there was quite a significant difference between the outcomes of my solution and the best solution - but it is good to know when one should finish seeking the solution, and usually one just seeks for a "winning position".

Blackzombie: Exactly, and I am aware of that. This is why I suggested three solutions that would make a computer (1', via "challenging line" ~ "long line") or users (1) do the thing, or would abandon making this choice at all (1'' - maybe not as a default, but as an option).

By the way, isn't 'the creator' now an algorithm? If it is so, then e.g. Suggestion 1' could be easily implemented, because it is already capable of determining what a "clear advantage" (that is, the end of the exercise) is.
#4 It's not that easy as to say a long line is more challenging always. A long line of natural moves can be easier to find than a short line of tricky moves.

What's difficult to find for a human is not necessarily what's difficult to find for a computer so every problem has to be manually revised then.

Also a problem is that there is subjectivity invoved in what's the hardest solution for a human to spot.

A computer alhorithm for deciding what's hard to spot for a human eye, I assume is very difficult to construct.

Since the tacrics trainer were revised there is only one solution that is best. Other will lead to equal or lost positions. Winning moves are therefore no longer penalized.
"It's not that easy as to say a long line is more challenging always. A long line of natural moves can be easier to find than a short line of tricky moves."

Surely, I don't claim that it is always the case. But I claim that in most cases, the longest line is quite tricky, and that it is very often trickier than the best (computer) response.

"What's difficult to find for a human is not necessarily what's difficult to find for a computer so every problem has to be manually revised then."

The suggested solutions don't need any attention from the site administrators (besides implementing them): solution 1 does not require revisions, 1' makes users revise the problems, 1'' does not require revisions.

"Also a problem is that there is subjectivity involved in what's the hardest solution for a human to spot."

This is a theoretical problem, not practical. The three solutions suggested do not solve the theoretical problem of finding the hardest line (besides 1'', which makes the problem nonexistent), but they surely improve the "hardness" of the line, solving a practical problem (of computer responses that are easy to met compared to other responses).

"A computer alhorithm for deciding what's hard to spot for a human eye, I assume is very difficult to construct."

This is why constructing such an algorithm is not suggested.

"Since the tactics trainer were revised there is only one solution that is best. Other will lead to equal or lost positions. Winning moves are therefore no longer penalized."

That's great!
The user-feedback based system you suggested could be a good idea. Users submit a move they find more challenging to meet and that is used to revise the puzzle.

This topic has been archived and can no longer be replied to.