- Blind mode tutorial
lichess.org
Donate

Chess engines need to search from solid positions or by phase.

#1

We have opening books and basically all that's needed is to cross the middle game to reach the endgame tables.
So the middle games need to be analysed for faults and corrected. An engine once out of book starts analysing. It can analyse 5 moves and re-evaluate from the new position that looked solid. Working up the ladder. It can do that until it reaches the end game table. Within a few seconds, it will have found a full line that the first few moves are solid enough to continue from there. Once it reaches the end that was done in steps, it will know if the horizon is favorable or not.

There is not need to re-invent an engine, just need to teach it to see the horizon by letting it breath. The engine will take a new positional breath from it's best lines. It most revaluate from that new position, until it reaches an end. The solution does not have to be the quickest, nor guaranteed perfect ending. Winning is all that matters. So the engine took the scenic route to win. Who cares, it wins. Rather than analysing 50 moves deep, it took a best line and used the first 5 moves and continued it's analysis from move 5. It probably will take about 12 positional evaluations to reach a calculate (not fully analysed) conclusion.

Example ... Initial positon evaluated: Stops at move 5 and considers this the new step in it's initial position analyses and does a new search to move 10... and so forth...
https://lichess.org/GxaW5MV7#5
This line was made using ChessX and Stockfish 15. I imported it to see the move number of the debut of the middle game.
The analysis was started from the initial positon and I left it run for days. Which was not needed to build a game in parts.

Even when I left it continue to do a depth search to create a better full line, the end of the lines were not always accurate. That's what lead me to believe that it's not needed to leave the engine analysis so long to build a solid game with zero centipawn loss.

If only an engine would not try to eat the whole game in one mouth full. But rather take it like in spoon fulls. It's lines would have calculated lines to the end. One step at a time to reach a new favorable position, then it jumps to that new psotion to rech another favorable horizon or end game.

If the full main line searched is a winner, it can then play with the move lengths to see if there is a quicker end game that could be less risky. If I can do that manually (resetting every 5 moves), it surely can be programmed to do that.

Chess engines developers, you can now have fun making this add-on idea open source.
Engines need to create positional milestones to calculate an end game, not just analyse none stop. That's a stepping analysis method that can be used to calculate until the end of the game. :)

We have opening books and basically all that's needed is to cross the middle game to reach the endgame tables. So the middle games need to be analysed for faults and corrected. An engine once out of book starts analysing. It can analyse 5 moves and re-evaluate from the new position that looked solid. Working up the ladder. It can do that until it reaches the end game table. Within a few seconds, it will have found a full line that the first few moves are solid enough to continue from there. Once it reaches the end that was done in steps, it will know if the horizon is favorable or not. There is not need to re-invent an engine, just need to teach it to see the horizon by letting it breath. The engine will take a new positional breath from it's best lines. It most revaluate from that new position, until it reaches an end. The solution does not have to be the quickest, nor guaranteed perfect ending. Winning is all that matters. So the engine took the scenic route to win. Who cares, it wins. Rather than analysing 50 moves deep, it took a best line and used the first 5 moves and continued it's analysis from move 5. It probably will take about 12 positional evaluations to reach a calculate (not fully analysed) conclusion. Example ... Initial positon evaluated: Stops at move 5 and considers this the new step in it's initial position analyses and does a new search to move 10... and so forth... https://lichess.org/GxaW5MV7#5 This line was made using ChessX and Stockfish 15. I imported it to see the move number of the debut of the middle game. The analysis was started from the initial positon and I left it run for days. Which was not needed to build a game in parts. Even when I left it continue to do a depth search to create a better full line, the end of the lines were not always accurate. That's what lead me to believe that it's not needed to leave the engine analysis so long to build a solid game with zero centipawn loss. If only an engine would not try to eat the whole game in one mouth full. But rather take it like in spoon fulls. It's lines would have calculated lines to the end. One step at a time to reach a new favorable position, then it jumps to that new psotion to rech another favorable horizon or end game. If the full main line searched is a winner, it can then play with the move lengths to see if there is a quicker end game that could be less risky. If I can do that manually (resetting every 5 moves), it surely can be programmed to do that. Chess engines developers, you can now have fun making this add-on idea open source. Engines need to create positional milestones to calculate an end game, not just analyse none stop. That's a stepping analysis method that can be used to calculate until the end of the game. :)
#3

What in the world what. Something was unclear? Need to create the app my self to see the benefits. Need tosee and touch to understand. Let me describe the idea in simpler terms....

Ever climb a ladder. One step at a time, or climbed stairs, two steps at a time.
Well chess engines are trying to climb the stair case completely in one step while keeping one foot on the first floor and the other stretching slowly to the top floor ... Metaphorically speaking.

So engines only need to evaluate lets say an X number of steps or moves which is very similar. Once it's convinced the first move is doable, it moves to that position on that stair case. It moves both black and white and does a new analysis on move 2 which is the second step.

If it wants to step to stair cases at the same time, then it searches that much deeper until it can play 2 moves of white and black and then starts a new analysis. All this happens until it reaches an end. When it sees its not a mate, it Analyzes backwards from end to beginning in the same fashion. If it used even number moves to calculate forward, it uses the odd number of moves to calculate backwards. Finds a move that is different and changes direction again until the line is a winning line.

What in the world what. Something was unclear? Need to create the app my self to see the benefits. Need tosee and touch to understand. Let me describe the idea in simpler terms.... Ever climb a ladder. One step at a time, or climbed stairs, two steps at a time. Well chess engines are trying to climb the stair case completely in one step while keeping one foot on the first floor and the other stretching slowly to the top floor ... Metaphorically speaking. So engines only need to evaluate lets say an X number of steps or moves which is very similar. Once it's convinced the first move is doable, it moves to that position on that stair case. It moves both black and white and does a new analysis on move 2 which is the second step. If it wants to step to stair cases at the same time, then it searches that much deeper until it can play 2 moves of white and black and then starts a new analysis. All this happens until it reaches an end. When it sees its not a mate, it Analyzes backwards from end to beginning in the same fashion. If it used even number moves to calculate forward, it uses the odd number of moves to calculate backwards. Finds a move that is different and changes direction again until the line is a winning line.
#4

maybe you should study how computer engines work before trying improve upon. also historical and abandoned ideas.

maybe you should study how computer engines work before trying improve upon. also historical and abandoned ideas.
#5

@Toscani
In short what do you exactly mean?
That engines need to analyse for few minutes about good enough moves to continue game?

@Toscani In short what do you exactly mean? That engines need to analyse for few minutes about good enough moves to continue game?
#6

Are you analyzing an opening with an engine? Don't do that. Engines are known to be bad in openings and human theory is better.
There is no need to analyze openings with an engine.

Are you analyzing an opening with an engine? Don't do that. Engines are known to be bad in openings and human theory is better. There is no need to analyze openings with an engine.
#7

Humans have broken down the game in three parts. The majority aim to develop and prep threats for the middle game.
Engines analysis and prune beyond the opening.
What needs to be done, is break the phases into 2, 3 or 4 plies. I tried the Lichess analysis, leaving the engine analyse until it stopped unless I would press the blue plus sign, which i did not do. Intend of making it analyse even deeper, I pressed on the line 1, the 4th ply. It played 2 moves at once and then the analysis started again. Again I waiting for it to stop. Again I pressed the 4th ply of the first line. After a about 20 moves, I was convinced it's not heading in the expected direction (by always using the line 1). The centipawn values were starting to drop to zero. So I went back and picked the 2nd line at an earlier point in the game, but I always picked the 4th (black piece) ply move. The game ends with white winning a game with my intervention. It would have been a draw, if I left Stockfish play all it's 1st line choice at ply 4. The engine needs some indication that the centipawn values are dropping and it needs to use the second line like I did to see if the value continues to rise or drop. If they drop, it needs to go back even further to correct what it though was the best 1st line. Obviously it's not in the programming or else these lines would not have mistakes in them.

The engine gets a horizon if it steps up it's analysis by stepping the analysis to the next move and using that move like it was the initial position. Ever move is an initial position for an engine. Obviously when it creates lines the first move is stronger than the second move of that line and so forth or else it would not make mistakes in it's lines. If it makes mistakes in it's lines, it's analysis is incorrect. The lines it creates needs to be 100% correct if followed through. I proved the lines have mistakes in the first two moves at 22 deep, by pressing ply 4 to play out the game to the end. I then let the Lichess server reanalyse it with Stockfish 15. I was using the local stockfish 14.

Here is the game: It has mistakes and imperfections.

https://lichess.org/study/SNZDIC4M

Correct the line using Stockfish second line choice on move 7. ... (line 2, 4 ply) and resubmit a new chapter in the study if you want.

Humans have broken down the game in three parts. The majority aim to develop and prep threats for the middle game. Engines analysis and prune beyond the opening. What needs to be done, is break the phases into 2, 3 or 4 plies. I tried the Lichess analysis, leaving the engine analyse until it stopped unless I would press the blue plus sign, which i did not do. Intend of making it analyse even deeper, I pressed on the line 1, the 4th ply. It played 2 moves at once and then the analysis started again. Again I waiting for it to stop. Again I pressed the 4th ply of the first line. After a about 20 moves, I was convinced it's not heading in the expected direction (by always using the line 1). The centipawn values were starting to drop to zero. So I went back and picked the 2nd line at an earlier point in the game, but I always picked the 4th (black piece) ply move. The game ends with white winning a game with my intervention. It would have been a draw, if I left Stockfish play all it's 1st line choice at ply 4. The engine needs some indication that the centipawn values are dropping and it needs to use the second line like I did to see if the value continues to rise or drop. If they drop, it needs to go back even further to correct what it though was the best 1st line. Obviously it's not in the programming or else these lines would not have mistakes in them. The engine gets a horizon if it steps up it's analysis by stepping the analysis to the next move and using that move like it was the initial position. Ever move is an initial position for an engine. Obviously when it creates lines the first move is stronger than the second move of that line and so forth or else it would not make mistakes in it's lines. If it makes mistakes in it's lines, it's analysis is incorrect. The lines it creates needs to be 100% correct if followed through. I proved the lines have mistakes in the first two moves at 22 deep, by pressing ply 4 to play out the game to the end. I then let the Lichess server reanalyse it with Stockfish 15. I was using the local stockfish 14. Here is the game: It has mistakes and imperfections. https://lichess.org/study/SNZDIC4M Correct the line using Stockfish second line choice on move 7. ... (line 2, 4 ply) and resubmit a new chapter in the study if you want.
#9

@Toscani

Your idea has several faults. One is that there is a reachable capped limit that's enough to prove one move is not faulty. That's generally not the case. Another fault is the idea that the engine is wasting a lot of time exploring irrelevant variations when it could just move forward a few moves. The branching factor of modern top engines is already insanely low, and wasting as little time as possible exploring irrelevant variations while trying to not miss any relevant one is already a central challenge. Yet another is that you're talking about finding a win. Sure, if the position is very clearly won, the engine might not need to run long to find good enough moves to convert the win, but then again it also won't waste a win if more time is used to find a good way to convert. And in the general case, the position is more likely drawn.

Now, all engines are tuned for bullet-play, so they are not tuned ideally for very long searches, and solving/backpropagating positions can help to be more efficient when doing a very deep analysis... But your method as described would not work

@asdf5656 said in #6:

Are you analyzing an opening with an engine? Don't do that. Engines are known to be bad in openings and human theory is better.
There is no need to analyze openings with an engine.

Modern top engines play the opening perfectly fine. The idea that they play poorly in the opening is severely outdated.

@Toscani Your idea has several faults. One is that there is a reachable capped limit that's enough to prove one move is not faulty. That's generally not the case. Another fault is the idea that the engine is wasting a lot of time exploring irrelevant variations when it could just move forward a few moves. The branching factor of modern top engines is already insanely low, and wasting as little time as possible exploring irrelevant variations while trying to not miss any relevant one is already a central challenge. Yet another is that you're talking about finding a win. Sure, if the position is very clearly won, the engine might not need to run long to find good enough moves to convert the win, but then again it also won't waste a win if more time is used to find a good way to convert. And in the general case, the position is more likely drawn. Now, all engines are tuned for bullet-play, so they are not tuned ideally for very long searches, and solving/backpropagating positions can help to be more efficient when doing a very deep analysis... But your method as described would not work @asdf5656 said in #6: > Are you analyzing an opening with an engine? Don't do that. Engines are known to be bad in openings and human theory is better. > There is no need to analyze openings with an engine. Modern top engines play the opening perfectly fine. The idea that they play poorly in the opening is severely outdated.
#10

Chapter one had mistakes. So I created Chapter 2.
Using main line 2: to avoid the 1st main line move 7. .. Ng4 which was not best.
Main Line 2. (7. c3 Be7 8. g3 b6 )
So I press 8. ... b6 (4th ply in the line).

  • Then I continued to use line 1, 4th plies until I reached move 51. The problem was it proposed only one ply on line 1. So I avoided the situation by using line 2 on move 51.

  • Then I continued with line 1, 4th ply, but again it gave me a line 1 with only one choice the first ply. It felt incomplete, so I picked line 2, 4th ply of move 57. (Is it incomplete if it does not give more than one ply?)

Chess has a trunk, not just branches. I'm trying to determine the trunk length of a tree. The wider the trunk of the tree, the more branches it can hold, but there is one trunk holding all main branches. The trunk has a height, which in chess terms is how long that game line should be expected to last.

Chapter one had mistakes. So I created Chapter 2. Using main line 2: to avoid the 1st main line move 7. .. Ng4 which was not best. Main Line 2. (7. c3 Be7 8. g3 b6 ) So I press 8. ... b6 (4th ply in the line). * Then I continued to use line 1, 4th plies until I reached move 51. The problem was it proposed only one ply on line 1. So I avoided the situation by using line 2 on move 51. * Then I continued with line 1, 4th ply, but again it gave me a line 1 with only one choice the first ply. It felt incomplete, so I picked line 2, 4th ply of move 57. (Is it incomplete if it does not give more than one ply?) Chess has a trunk, not just branches. I'm trying to determine the trunk length of a tree. The wider the trunk of the tree, the more branches it can hold, but there is one trunk holding all main branches. The trunk has a height, which in chess terms is how long that game line should be expected to last.
#11

The C10 French Defense: Rubinstein Variation, Kasparov Attack, Chapter 2 is complete.
It has only one inaccuracy (move 31) that should be fixed when Chapter 3 comes out.

Is an analysis of line 1 incomplete, when it only gives me 1 ply choice to select on that first line ?

The C10 French Defense: Rubinstein Variation, Kasparov Attack, Chapter 2 is complete. It has only one inaccuracy (move 31) that should be fixed when Chapter 3 comes out. Is an analysis of line 1 incomplete, when it only gives me 1 ply choice to select on that first line ?

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