Rebel wrote:
We have been here before. Remove null-move and LMR and the similarity will remain. Why don't you do experiments yourself first? Some of us did.
If my target is to change move choice by search change then my first thinking is not to remove null move or LMR but changing the order of moves
The point is that I believe that there are many positions when 2 moves have the same score and the move choice is dependent on the move that the program search first.
Of course you need to search hash move good captures and killers first but after it or when you have no hash move and no killers or good captures(that is usually the case at depth 1) you should use the opposite order of moves if you want to change the move choice.
That is easy to test. I sort my root move list but I could simply shuffle that list and see how it does on the sim test.
Don
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
Rebel wrote:
We have been here before. Remove null-move and LMR and the similarity will remain. Why don't you do experiments yourself first? Some of us did.
If my target is to change move choice by search change then my first thinking is not to remove null move or LMR but changing the order of moves
The point is that I believe that there are many positions when 2 moves have the same score and the move choice is dependent on the move that the program search first.
Of course you need to search hash move good captures and killers first but after it or when you have no hash move and no killers or good captures(that is usually the case at depth 1) you should use the opposite order of moves if you want to change the move choice.
I don't that will make much of a difference because removing null-move and especially LMR will have a greater impact on move selection but I will give it a shot. Also realize the nature of the 8xxx positions, there is a lot of move choices (non forced positions) as the following statistic with 99 different engines shows:
Rebel wrote:
We have been here before. Remove null-move and LMR and the similarity will remain. Why don't you do experiments yourself first? Some of us did.
If my target is to change move choice by search change then my first thinking is not to remove null move or LMR but changing the order of moves
The point is that I believe that there are many positions when 2 moves have the same score and the move choice is dependent on the move that the program search first.
Of course you need to search hash move good captures and killers first but after it or when you have no hash move and no killers or good captures(that is usually the case at depth 1) you should use the opposite order of moves if you want to change the move choice.
I don't that will make much of a difference because removing null-move and especially LMR will have a greater impact on move selection but I will give it a shot. Also realize the nature of the 8xxx positions, there is a lot of move choices (non forced positions) as the following statistic with 99 different engines shows:
1. means there were 24 positions of the 8xxx all 99 engines played the same move.
2. means 682 positions with 2 move choices by 99 engines, etc.
In non forced positions when there is no one best move
order of moves may be very important.
Let take an extreme example.
In KNN vs K positions
assuming that you evaluate them as 0.00 when the lone king is not in the corner LMR or null move is not going to change nothing but changing the order of moves is always going to change the move choice(if you have at least 2 drawing moves).
Most positions are not obviously drawn positions but I believe that there are significant number of positions when at least 2 best moves have the same evaluation and it is part of the reason that you get many moves by 99 engines.
Rebel wrote:
We have been here before. Remove null-move and LMR and the similarity will remain. Why don't you do experiments yourself first? Some of us did.
If my target is to change move choice by search change then my first thinking is not to remove null move or LMR but changing the order of moves
The point is that I believe that there are many positions when 2 moves have the same score and the move choice is dependent on the move that the program search first.
Of course you need to search hash move good captures and killers first but after it or when you have no hash move and no killers or good captures(that is usually the case at depth 1) you should use the opposite order of moves if you want to change the move choice.
I don't that will make much of a difference because removing null-move and especially LMR will have a greater impact on move selection but I will give it a shot. Also realize the nature of the 8xxx positions, there is a lot of move choices (non forced positions) as the following statistic with 99 different engines shows:
1. means there were 24 positions of the 8xxx all 99 engines played the same move.
2. means 682 positions with 2 move choices by 99 engines, etc.
In non forced positions when there is no one best move
order of moves may be very important.
Let take an extreme example.
In KNN vs K positions
assuming that you evaluate them as 0.00 when the lone king is not in the corner LMR or null move is not going to change nothing but changing the order of moves is always going to change the move choice(if you have at least 2 drawing moves).
Most positions are not obviously drawn positions but I believe that there are significant number of positions when at least 2 best moves have the same evaluation and it is part of the reason that you get many moves by 99 engines.
Okay, criticism should be taken serious, there are some more unexplored areas as well.
Don suggested: That is easy to test. I sort my root move list but I could simply shuffle that list and see how it does on the sim test.
Rebel wrote:
We have been here before. Remove null-move and LMR and the similarity will remain. Why don't you do experiments yourself first? Some of us did.
If my target is to change move choice by search change then my first thinking is not to remove null move or LMR but changing the order of moves
The point is that I believe that there are many positions when 2 moves have the same score and the move choice is dependent on the move that the program search first.
Of course you need to search hash move good captures and killers first but after it or when you have no hash move and no killers or good captures(that is usually the case at depth 1) you should use the opposite order of moves if you want to change the move choice.
I don't that will make much of a difference because removing null-move and especially LMR will have a greater impact on move selection but I will give it a shot. Also realize the nature of the 8xxx positions, there is a lot of move choices (non forced positions) as the following statistic with 99 different engines shows:
1. means there were 24 positions of the 8xxx all 99 engines played the same move.
2. means 682 positions with 2 move choices by 99 engines, etc.
In non forced positions when there is no one best move
order of moves may be very important.
Let take an extreme example.
In KNN vs K positions
assuming that you evaluate them as 0.00 when the lone king is not in the corner LMR or null move is not going to change nothing but changing the order of moves is always going to change the move choice(if you have at least 2 drawing moves).
Most positions are not obviously drawn positions but I believe that there are significant number of positions when at least 2 best moves have the same evaluation and it is part of the reason that you get many moves by 99 engines.
Okay, criticism should be taken serious, there are some more unexplored areas as well.
Don suggested: That is easy to test. I sort my root move list but I could simply shuffle that list and see how it does on the sim test.
What do you suggest?
I think that it is a good idea to test it.
You can also test suffling the move order(after the obvious hash captures and killers) not only in the root but I guess that this is less important for move choice relative to doing it in the root.
Uri Blass wrote: I think that it is a good idea to test it.
You can also test suffling the move order(after the obvious hash captures and killers) not only in the root but I guess that this is less important for move choice relative to doing it in the root.
Here are the results of shuffling the move order throughout the whole tree.
A more important unexplored area (and challenge) would be to measure the minimum elo loss to hide the origin of an engine. The elo loss is a fact, its minimum number is unknown.
