Positive Reinforcement During Computer Analysis • page 1/2 • Lichess Feedback • lichess.org

Apologies if this has already been addressed; I did not see it in the FAQ nor in a brief forum search.

I am still relatively new to serious chess and lichess, and I tend to use the computer analysis tool quite heavily after games to understand where I can improve. The details/graphs/etc provided are enormously helpful in identifying missed opportunities, blunders, etc. This is great for learning what I'm doing wrong, but is not so helpful in finding out what I'm doing right.

I understand that eventually the "right" moves will be obvious/reinforced via wins. But in lower rated games where the value of a particular good move may be lost/rendered unimportant by both my own and my opponent's mistakes and blunders, I think it could be valuable to show when a good/great move was made.

I've heard that -other site- has something like this behind a paywall. Something like "accuracy against Stockfish-X" or "number of perfect/good moves as thought by the engine" would definitely benefit my game review.

I've only briefly looked at the Lichess source, I do not know how difficult something like this would be to implement. I know the computation already exists somewhere for the lessons/practice ("Good Move!" "You can do better" etc.), so it is probably more on the UI side + summary statistics.

Apologies if this has already been addressed; I did not see it in the FAQ nor in a brief forum search. I am still relatively new to serious chess and lichess, and I tend to use the computer analysis tool quite heavily after games to understand where I can improve. The details/graphs/etc provided are enormously helpful in identifying missed opportunities, blunders, etc. This is great for learning what I'm doing wrong, but is not so helpful in finding out what I'm doing right. I understand that eventually the "right" moves will be obvious/reinforced via wins. But in lower rated games where the value of a particular good move may be lost/rendered unimportant by both my own and my opponent's mistakes and blunders, I think it could be valuable to show when a good/great move was made. I've heard that -other site- has something like this behind a paywall. Something like "accuracy against Stockfish-X" or "number of perfect/good moves as thought by the engine" would definitely benefit my game review. I've only briefly looked at the Lichess source, I do not know how difficult something like this would be to implement. I know the computation already exists somewhere for the lessons/practice ("Good Move!" "You can do better" etc.), so it is probably more on the UI side + summary statistics.

tpr

#2

The analysis gives you the average centipawn loss, which is is direct measure of your accuracy against Stockfish

rmeh

#3

I understand that, my main point is all measures of success are "how bad your move was" and none are "how good your move was". It is all negative reinforcement. As I mentioned the feedback is great, but as is typically known positive reinforcement is significantly more productive for learning and improvement.

This could be as simple as just showing how many moves you played with centipawn loss lower than X. Anything that is POSITIVE feedback instead of negative.

I understand that, my main point is all measures of success are "how bad your move was" and none are "how good your move was". It is all negative reinforcement. As I mentioned the feedback is great, but as is typically known positive reinforcement is significantly more productive for learning and improvement. This could be as simple as just showing how many moves you played with centipawn loss lower than X. Anything that is POSITIVE feedback instead of negative.

Sybotes

#4

There has been a thread with the suggestion: Put a ! behind the moves in the notation - not only ? and ?? ...
But an algorithm for that would be very complicated and probably resource-eating. What would a good move be? Surely not only what the engine considers best. But even if you move the best - you do not want to be patronized wit a "goog move" mark if you recapture in a trade - though it is probably the only move that doesn't lose instantly.
So is a good move one among a couple of candidates all not far from the optimum? Still hard to decide.
When you encounter a ! in a chess book, it is usually put when a move is good against first sight. So for honouring a move as good, the engine might have to try - 4 moves deep: looks bad; 5 moves deep - looks bad; 6 moves deep: good evaluation. And the later a move shows a good evaluation, the more it deserves a "!".
But if you analyze your games here - just be pleased if your move was amonng the 2 or 3 best from the Stockfish lines. And try to find out what made you not to move the best!

There has been a thread with the suggestion: Put a ! behind the moves in the notation - not only ? and ?? ... But an algorithm for that would be very complicated and probably resource-eating. What would a good move be? Surely not only what the engine considers best. But even if you move the best - you do not want to be patronized wit a "goog move" mark if you recapture in a trade - though it is probably the only move that doesn't lose instantly. So is a good move one among a couple of candidates all not far from the optimum? Still hard to decide. When you encounter a ! in a chess book, it is usually put when a move is good against first sight. So for honouring a move as good, the engine might have to try - 4 moves deep: looks bad; 5 moves deep - looks bad; 6 moves deep: good evaluation. And the later a move shows a good evaluation, the more it deserves a "!". But if you analyze your games here - just be pleased if your move was amonng the 2 or 3 best from the Stockfish lines. And try to find out what made you not to move the best!

Toadofsky

#5

#1 That's fine but please understand I have addressed this topic at length... repeatedly suggesting people develop and test a solution which doesn't require additional CPU, memory, etc.

jperkins

#6

Maybe use a red font or background for when the <strike>normal</strike>default user makes a blunder and allow an option for the OP to have ROYGBIV wheel of choices for inaccuracies, mistakes, blunder, and not_bad moves.

reminded me of this:
https://time.com/31960/school-bans-teachers-from-using-red-ink-because-its-too-mean/
can read it but cant relate. I am old so forgive me.

For me during analysis, the stockfish analysis number at top right of board gives me all the feedback needed to determine if my last move got me closer to a win or a loss.

Maybe use a red font or background for when the <strike>normal</strike>default user makes a blunder and allow an option for the OP to have ROYGBIV wheel of choices for inaccuracies, mistakes, blunder, and not_bad moves. reminded me of this: https://time.com/31960/school-bans-teachers-from-using-red-ink-because-its-too-mean/ can read it but cant relate. I am old so forgive me. For me during analysis, the stockfish analysis number at top right of board gives me all the feedback needed to determine if my last move got me closer to a win or a loss.

jperkins edited

#7

added to css . not exactly what I was thinking but along those lines. probably have to massage the code to get it to show up in the actual analysis line. And I am sure lichess is a lot more complicated than this, so I could be way off base here

comment.mistake {
background-color: pink;
}

in my case I think it was this css file but appears to be related to my preferences
https://lichess1.org/assets/_L9umnl/css/analyse.round.light.min.css

not sure if this pic will stay valid as I didnt join imgur to post it
https://i.imgur.com/IpHTNEn.png

added to css . not exactly what I was thinking but along those lines. probably have to massage the code to get it to show up in the actual analysis line. And I am sure lichess is a lot more complicated than this, so I could be way off base here comment.mistake { background-color: pink; } in my case I think it was this css file but appears to be related to my preferences https://lichess1.org/assets/_L9umnl/css/analyse.round.light.min.css not sure if this pic will stay valid as I didnt join imgur to post it https://i.imgur.com/IpHTNEn.png

jperkins edited

#8

test of alternate png location for pic display - and it failed. now assuming it is limited to imgur

Toadofsky

#9

My guess is that nobody can find an efficient way to provide automated analysis which provides meaningful positive reinforcement, at least not before complexity and other metrics in positions are established.

jperkins edited

#10

I agree and like that lichess 'keeps it simple' It could be considered that the job of positive reinforcement would fall under the definition of 'Chess Coach' and not 'Chess analysis'. Another solution could be that one could post their games in one of the other forum topics and request an analysis of 'what I did correctly in this game'

Also noticed that there is already a color indication already in lichess. the small horizontal line at the right side of the 'mistake' line. sorta makes my earlier comment redundant in a way. It did get me wondering if the <eval>+0.2</eval> element could be be used to determine the background color of that element , by javascript, on the client side. a positive value would be more green and a negative would be more red. Might have to take into consideration what color the user is though. and it just gets more complicated after that for a feature that isnt critical to me.

I agree and like that lichess 'keeps it simple' It could be considered that the job of positive reinforcement would fall under the definition of 'Chess Coach' and not 'Chess analysis'. Another solution could be that one could post their games in one of the other forum topics and request an analysis of 'what I did correctly in this game' Also noticed that there is already a color indication already in lichess. the small horizontal line at the right side of the 'mistake' line. sorta makes my earlier comment redundant in a way. It did get me wondering if the <eval>+0.2</eval> element could be be used to determine the background color of that element , by javascript, on the client side. a positive value would be more green and a negative would be more red. Might have to take into consideration what color the user is though. and it just gets more complicated after that for a feature that isnt critical to me.