ENGINE SPEED AND STRONG

[Krzysztof Grzelak]

Give an example of how it looks in practice.

Starts in the first laptop program Kingsrow - program playing white.

Starts in the second laptop program Argus - program playing black.

I set the time on both laptops, for example - 30 minutes for 75 moves.

The program Kingsrow made a move 34-30. Computer mouse on a second laptop in the Argus I perform movement 34-30.
At this point, I wait until the program Argus will make a move. I do move that pointed program Argus on the laptop program Kingsrow. And so alternately until the party ends. Later, I write the party and played another involving other programs.

[Joost Buijs]

Give an example of how it looks in practice.

Starts in the first laptop program Kingsrow - program playing white.

Starts in the second laptop program Argus - program playing black.

I set the time on both laptops, for example - 30 minutes for 75 moves.

The program Kingsrow made a move 34-30. Computer mouse on a second laptop in the Argus I perform movement 34-30.
At this point, I wait until the program Argus will make a move. I do move that pointed program Argus on the laptop program Kingsrow. And so alternately until the party ends. Later, I write the party and played another involving other programs.

Well, for me the most easy will be when you type the move (for instance 34-30) in the console, this is like my old program worked 25 years ago, I will see what I can do.

I can add the time and the number of moves in the configuration file, this is probably the most easy because you only have to set it once. Do you also have Ed's EGDB's? My program can use them as an option but it is not really necessary, it won't change the playing strength by much.

Unfortunately I have to leave now, I will get your reply tomorrow. Talk to you later.

Joost

[Krzysztof Grzelak]

Hi Joost

So I have the base ends of Ed's.

[Sidiki]

I can add the time and the number of moves in the configuration file, this is probably the most easy because you only have to set it once. Do you also have Ed's EGDB's? My program can use them as an option but it is not really necessary, it won't change the playing strength by much.

Unfortunately I have to leave now, I will get your reply tomorrow. Talk to you later.

Joost

Hi Joost,

I launched a dxp match between Scan and Kingsrow, Scan runned under Damage 2016 hub of Bert and used 5 EGDB and Kingsrow used 6 EGDB.
1 min/75 moves i saw that Kingsrow crashed Scan, after what i enabled the 6 EGDB of Scan and the match became better.

I thought that with a good evaluation and deep search and speed of nodes/s an engine which own these characteristics will outplay any other engine. Because for me EGDB was to gain time. It was the case in his time with Kingsrow which done several matches against Flits and Truus without EGDB and won these game easily.

Isn't only possible when the elo points between these two engine it's enough?
Because everybody Know that Scan, Kingsrow, Damage, Argus levels are closer.
So, in this case if two of these engines play a match, the engine wich will get the biggest EGDB will have the best chance to win the match?

[Joost Buijs]

I can add the time and the number of moves in the configuration file, this is probably the most easy because you only have to set it once. Do you also have Ed's EGDB's? My program can use them as an option but it is not really necessary, it won't change the playing strength by much.

Unfortunately I have to leave now, I will get your reply tomorrow. Talk to you later.

Joost

Hi Joost,

I launched a dxp match between Scan and Kingsrow, Scan runned under Damage 2016 hub of Bert and used 5 EGDB and Kingsrow used 6 EGDB.
1 min/75 moves i saw that Kingsrow crashed Scan, after what i enabled the 6 EGDB of Scan and the match became better.

I thought that with a good evaluation and deep search and speed of nodes/s an engine which own these characteristics will outplay any other engine. Because for me EGDB was to gain time. It was the case in his time with Kingsrow which done several matches against Flits and Truus without EGDB and won these game easily.

Isn't only possible when the elo points between these two engine it's enough?
Because everybody Know that Scan, Kingsrow, Damage, Argus levels are closer.
So, in this case if two of these engines play a match, the engine wich will get the biggest EGDB will have the best chance to win the match?

Hi Sidiki,

Indeed, although I have hardly any experience with draughts and EGDB my feeling about this is the same. When you have two programs of comparable strength the one with the largest (most pieces) EGDB will have a greater chance to win the match. Lately evaluation-functions are getting better and because of this the importance of EGDB's are getting less, at least I don't see much difference with my program, the slowdown of probing the EGDB more or less annihilates the advantage it gives, it is nice that it can see a win or draw much earlier but it doesn't change the outcome of the game, I guess a game has to be won in an earlier stage.

When I look at the wins or losses of my program the blunder is usually made very early in the game and I don't think that EGDB will help in such cases. The 4x4 pattern evaluation works very well in practice but it contains holes which in most cases are plugged by the search, on some occasions though one of these holes slips through, a good opening book will help, unfortunately I don't have one yet.

You need to play thousands of games to determine true strength, since I have no means to play very fast games yet I usually play matches of 100 games and sometimes I see a variance in strength of ~25 Elo. There is a lot of randomness added by the SMP search which is the main cause of this phenomenon.

I don't know which program is stronger, Scan or Kingsrow, when I let them play on my hardware (without EGDB) they are of comparable strength, my program is still a tiny bit weaker (probably because it is all new and not very well tuned yet), this is when I test it with the slow version, the fast version is comparable in strength with the other two programs.

Joost

[Sidiki]

Hi Sidiki,

Indeed, although I have hardly any experience with draughts and EGDB my feeling about this is the same. When you have two programs of comparable strength the one with the largest (most pieces) EGDB will have a greater chance to win the match. Lately evaluation-functions are getting better and because of this the importance of EGDB's are getting less, at least I don't see much difference with my program, the slowdown of probing the EGDB more or less annihilates the advantage it gives, it is nice that it can see a win or draw much earlier but it doesn't change the outcome of the game, I guess a game has to be won in an earlier stage.

When I look at the wins or losses of my program the blunder is usually made very early in the game and I don't think that EGDB will help in such cases. The 4x4 pattern evaluation works very well in practice but it contains holes which in most cases are plugged by the search, on some occasions though one of these holes slips through, a good opening book will help, unfortunately I don't have one yet.

You need to play thousands of games to determine true strength, since I have no means to play very fast games yet I usually play matches of 100 games and sometimes I see a variance in strength of ~25 Elo. There is a lot of randomness added by the SMP search which is the main cause of this phenomenon.

I don't know which program is stronger, Scan or Kingsrow, when I let them play on my hardware (without EGDB) they are of comparable strength, my program is still a tiny bit weaker (probably because it is all new and not very well tuned yet), this is when I test it with the slow version, the fast version is comparable in strength with the other two programs.

Joost

Hi Joost,

You are right, effectivly, the eval function is more important than EGDB.
Like you said, EGDB permit to see early a win/draw/loss, but haven't a significant impact on the playing strenght.

The idea of opening book is also good.
This question is for you and all the others programmers:
"Is't possible to built a deep opening bookmuch than 30 "perfect" moves to permit to an engine to be more stronger?

This idea come every day into my mind.
Thank again to all programmers of this forum for the share spirit.

Sidiki

[Joost Buijs]

Hi Sidiki,

Indeed, although I have hardly any experience with draughts and EGDB my feeling about this is the same. When you have two programs of comparable strength the one with the largest (most pieces) EGDB will have a greater chance to win the match. Lately evaluation-functions are getting better and because of this the importance of EGDB's are getting less, at least I don't see much difference with my program, the slowdown of probing the EGDB more or less annihilates the advantage it gives, it is nice that it can see a win or draw much earlier but it doesn't change the outcome of the game, I guess a game has to be won in an earlier stage.

When I look at the wins or losses of my program the blunder is usually made very early in the game and I don't think that EGDB will help in such cases. The 4x4 pattern evaluation works very well in practice but it contains holes which in most cases are plugged by the search, on some occasions though one of these holes slips through, a good opening book will help, unfortunately I don't have one yet.

You need to play thousands of games to determine true strength, since I have no means to play very fast games yet I usually play matches of 100 games and sometimes I see a variance in strength of ~25 Elo. There is a lot of randomness added by the SMP search which is the main cause of this phenomenon.

I don't know which program is stronger, Scan or Kingsrow, when I let them play on my hardware (without EGDB) they are of comparable strength, my program is still a tiny bit weaker (probably because it is all new and not very well tuned yet), this is when I test it with the slow version, the fast version is comparable in strength with the other two programs.

Joost

Hi Joost,

You are right, effectivly, the eval function is more important than EGDB.
Like you said, EGDB permit to see early a win/draw/loss, but haven't a significant impact on the playing strenght.

The idea of opening book is also good.
This question is for you and all the others programmers:
"Is't possible to built a deep opening bookmuch than 30 "perfect" moves to permit to an engine to be more stronger?

This idea come every day into my mind.
Thank again to all programmers of this forum for the share spirit.

Sidiki

Hi Sidiki,

You can only build an opening-book with "perfect" moves if you can solve the game of 10x10 draughts, I'm afraid that with current hardware the game is too complex to solve, but it is possible to build a book with "near perfect" moves, moves that are very deeply analyzed or have very reliable statistics.

The problem with draughts is that it is very drawish, there is a strong indication that the game will always be a draw when you don't make very big mistakes, so you might argue that all moves leading to a draw are in a sense "perfect".

Actually I find it a bit boring when you play a match of hundreds of games that >95% of them are drawn, this is with short thinking times of less than a second, with longer thinking times this percentage gets even higher. Maybe there is a possibility to improve the situation by using larger patterns in the evaluation function, but I doubt whether it will have much effect.

I read on the forum that there is something called 'Killer Draughts' which seems to be less drawish, I don't know the rules and I don't know if there are engines supporting this, when there are more people working on this I will add this variation to my engine as well.

Joost

[TAILLE]

Hi,

The current discussion looks a little strange for me. Comparing the strength due to eval function vs EGDB is a non-sense because these are not built for the same kind of positions. The eval function is build in order to play the beginning and the middle game (typically the 45 first moves) and the EGDB is build for playing the endgame (typically after the 45th move).
My view is the following: draugths is clearly a drawing game. Of course if a program A has a far better evaluation function than program B, it will win the game in the beginning and middle game and the EGDB is of no use but this case is not very interesting is it?
The most interesting situation is the following: program A eval function is a LITTLE better than program B eval function and, as a consequence, program A reaches quite often the endgame with a clear advantage but not a clear win! Now the EGDB becomes the key point: if program A has the best EGDB it will win many of these games but if program B has the best EGDB it will manage to draw a lot of these.

Let's take the following problem presented by Jaap Bus


White to play

We all see a clear white advantage but who is able to see a win? If the black man on 40 can receive the support of another black man then a black promotion will be possible with a good chance for a draw.

Damy, with the 8 pieces EGDB, is able to find the winning move but with only the 6 pieces EGDB I have to admit that Damy is not able to find this winning move.

A strong chess players liked to say that he could loose a game only if its opponent plays a better beginnig game AND a better middle game AND a better endgame!

Isn't it the same in draughts?

[Joost Buijs]

Hi,

The current discussion looks a little strange for me. Comparing the strength due to eval function vs EGDB is a non-sense because these are not built for the same kind of positions. The eval function is build in order to play the beginning and the middle game (typically the 45 first moves) and the EGDB is build for playing the endgame (typically after the 45th move).
My view is the following: draugths is clearly a drawing game. Of course if a program A has a far better evaluation function than program B, it will win the game in the beginning and middle game and the EGDB is of no use but this case is not very interesting is it?
The most interesting situation is the following: program A eval function is a LITTLE better than program B eval function and, as a consequence, program A reaches quite often the endgame with a clear advantage but not a clear win! Now the EGDB becomes the key point: if program A has the best EGDB it will win many of these games but if program B has the best EGDB it will manage to draw a lot of these.

Let's take the following problem presented by Jaap Bus


White to play

We all see a clear white advantage but who is able to see a win? If the black man on 40 can receive the support of another black man then a black promotion will be possible with a good chance for a draw.

Damy, with the 8 pieces EGDB, is able to find the winning move but with only the 6 pieces EGDB I have to admit that Damy is not able to find this winning move.

A strong chess players liked to say that he could loose a game only if its opponent plays a better beginnig game AND a better middle game AND a better endgame!

Isn't it the same in draughts?

Hi,

How long does it take before Damy finds the winning move in this position? If it takes several minutes it is not very useful when you play a tournament don't you think so? And what is the advantage of seeing a winning move when the program plays the same winning move without using the EGDB?

I just look at win/loss percentages and the gain I get by using EGDB is minimal, the same holds for chess btw.

It takes my program with 8p EGDB 10 minutes to find 37-31 with a win in 45 ply, but when I disable the EGDB it also plays 37-31 because it sees after 12 seconds that all other moves are worse, so in this case the EGDB doesn't help at all.

Joost

[TAILLE]

Hi,

The current discussion looks a little strange for me. Comparing the strength due to eval function vs EGDB is a non-sense because these are not built for the same kind of positions. The eval function is build in order to play the beginning and the middle game (typically the 45 first moves) and the EGDB is build for playing the endgame (typically after the 45th move).
My view is the following: draugths is clearly a drawing game. Of course if a program A has a far better evaluation function than program B, it will win the game in the beginning and middle game and the EGDB is of no use but this case is not very interesting is it?
The most interesting situation is the following: program A eval function is a LITTLE better than program B eval function and, as a consequence, program A reaches quite often the endgame with a clear advantage but not a clear win! Now the EGDB becomes the key point: if program A has the best EGDB it will win many of these games but if program B has the best EGDB it will manage to draw a lot of these.

Let's take the following problem presented by Jaap Bus


White to play

We all see a clear white advantage but who is able to see a win? If the black man on 40 can receive the support of another black man then a black promotion will be possible with a good chance for a draw.

Damy, with the 8 pieces EGDB, is able to find the winning move but with only the 6 pieces EGDB I have to admit that Damy is not able to find this winning move.

A strong chess players liked to say that he could loose a game only if its opponent plays a better beginnig game AND a better middle game AND a better endgame!

Isn't it the same in draughts?

Hi,

How long does it take before Damy finds the winning move in this position? If it takes several minutes it is not very useful when you play a tournament don't you think so? And what is the advantage of seeing a winning move when the program plays the same winning move without using the EGDB?

I just look at win/loss percentages and the gain I get by using EGDB is minimal, the same holds for chess btw.

It takes my program with 8p EGDB 10 minutes to find 37-31 with a win in 45 ply, but when I disable the EGDB it also plays 37-31 because it sees after 12 seconds that all other moves are worse, so in this case the EGDB doesn't help at all.

Joost

Hi Joost,

With the EGDB cache empty Damy chooses 37-31 after 10" and the win is proved after 25".
If now the cache is already loaded with the relevant db for this position then 37-31 is chosen after 5" and the win is proved after 9".
In a real game the cache is never empty and we can guess that the win is proved between 9" and 25".

With only the 6 pieces EGDB then Damy changeg regularly is move (37-31, 37-32, 28-23, 29-23, 29-24) and after several minutes Damy prefered the apparently strong positionnel move 28-23?

In a match between programs don't you think that it may exist some undiscovered wins?
When I see a result like 4 wins, 7 losses, 147 draws, 0 unknowns I conclude that in 11 games a program takes an advantage and manage to win but couldn't we also conclude that in the 147 drawing game it may exist may be 10 other games in which a win was not found?

[Fabien Letouzey]

Hi Joost,

I read on the forum that there is something called 'Killer Draughts' which seems to be less drawish, I don't know the rules and I don't know if there are engines supporting this, when there are more people working on this I will add this variation to my engine as well.

Indeed the rules seem hard to find! I expected them to be immediately visible on http://www.killerdraughts.org but they are nowhere in sight? It's as follows: when the last piece you capture with a king is also a king, you have to stop immediately behind it. In Scan, that's two additional lines of code.

I talk a little bit about the draw rate in this thread: http://fmjd.org/bb3/viewtopic.php?f=53&t=7018

I have interest in killer draughts, which I consider the only promising rule set so far. Scan 2.0, however, does not support it out of the box. I recall that Walter made a version available.

Fabien.

[Joost Buijs]

Hi Joost,

I read on the forum that there is something called 'Killer Draughts' which seems to be less drawish, I don't know the rules and I don't know if there are engines supporting this, when there are more people working on this I will add this variation to my engine as well.

Indeed the rules seem hard to find! I expected them to be immediately visible on http://www.killerdraughts.org but they are nowhere in sight? It's as follows: when the last piece you capture with a king is also a king, you have to stop immediately behind it. In Scan, that's two additional lines of code.

I talk a little bit about the draw rate in this thread: http://fmjd.org/bb3/viewtopic.php?f=53&t=7018

I have interest in killer draughts, which I consider the only promising rule set so far. Scan 2.0, however, does not support it out of the box. I recall that Walter made a version available.

Fabien.

Hi Fabien,

Thanks!

Your description of KD looks very simple to implement, I will take a look at my king-capture generator to see if this rule is easy to add, maybe the easiest and the fastest is to add a second kc-generator for KD only. I guess it will have some consequences for the evaluation-function as well but that is of later concern.

Joost

[Joost Buijs]

Hi,

The current discussion looks a little strange for me. Comparing the strength due to eval function vs EGDB is a non-sense because these are not built for the same kind of positions. The eval function is build in order to play the beginning and the middle game (typically the 45 first moves) and the EGDB is build for playing the endgame (typically after the 45th move).
My view is the following: draugths is clearly a drawing game. Of course if a program A has a far better evaluation function than program B, it will win the game in the beginning and middle game and the EGDB is of no use but this case is not very interesting is it?
The most interesting situation is the following: program A eval function is a LITTLE better than program B eval function and, as a consequence, program A reaches quite often the endgame with a clear advantage but not a clear win! Now the EGDB becomes the key point: if program A has the best EGDB it will win many of these games but if program B has the best EGDB it will manage to draw a lot of these.

Let's take the following problem presented by Jaap Bus


White to play

We all see a clear white advantage but who is able to see a win? If the black man on 40 can receive the support of another black man then a black promotion will be possible with a good chance for a draw.

Damy, with the 8 pieces EGDB, is able to find the winning move but with only the 6 pieces EGDB I have to admit that Damy is not able to find this winning move.

A strong chess players liked to say that he could loose a game only if its opponent plays a better beginnig game AND a better middle game AND a better endgame!

Isn't it the same in draughts?

Hi,

How long does it take before Damy finds the winning move in this position? If it takes several minutes it is not very useful when you play a tournament don't you think so? And what is the advantage of seeing a winning move when the program plays the same winning move without using the EGDB?

I just look at win/loss percentages and the gain I get by using EGDB is minimal, the same holds for chess btw.

It takes my program with 8p EGDB 10 minutes to find 37-31 with a win in 45 ply, but when I disable the EGDB it also plays 37-31 because it sees after 12 seconds that all other moves are worse, so in this case the EGDB doesn't help at all.

Joost

Hi Joost,

With the EGDB cache empty Damy chooses 37-31 after 10" and the win is proved after 25".
If now the cache is already loaded with the relevant db for this position then 37-31 is chosen after 5" and the win is proved after 9".
In a real game the cache is never empty and we can guess that the win is proved between 9" and 25".

With only the 6 pieces EGDB then Damy changeg regularly is move (37-31, 37-32, 28-23, 29-23, 29-24) and after several minutes Damy prefered the apparently strong positionnel move 28-23?

In a match between programs don't you think that it may exist some undiscovered wins?
When I see a result like 4 wins, 7 losses, 147 draws, 0 unknowns I conclude that in 11 games a program takes an advantage and manage to win but couldn't we also conclude that in the 147 drawing game it may exist may be 10 other games in which a win was not found?

Hi Gérard,

That you are able to find the win in this position within 25 seconds is amazing, this is very fast, in a second run my program took ~7 minutes to find the win (the time varies due to SMP), the variation to the won 8p EGDB position is 45 ply long and starts with: 37-31 12-17 22x11 16x07 27-21 10-14 21-16 14-19.

It is possible that there are undiscovered wins when you don't use EGDB, but when I have my program play with or without EGDB the match scores don't change in a significant way, this is what I look at. Maybe my implementation is wrong, but I don't think so because with test-positions it all works fine but with game-play it does not seem to help much.

The point is that without EGDB my program finds after 12 seconds that 37-31 is the best move and with EGDB enabled it takes 26 seconds before it sees that 37-31 is the best move, this is due to the slowdown caused by probing and this can have an adversary effect during game-play.

Usually the errors are made earlier on in the game, for instance the position you show above is already very bad for black and that should not happen when your evaluation and search during the opening and midgame are working like they should.

Joost

[TAILLE]

Hi Gérard,

That you are able to find the win in this position within 25 seconds is amazing, this is very fast, in a second run my program took ~7 minutes to find the win (the time varies due to SMP), the variation to the won 8p EGDB position is 45 ply long and starts with: 37-31 12-17 22x11 16x07 27-21 10-14 21-16 14-19.

It is possible that there are undiscovered wins when you don't use EGDB, but when I have my program play with or without EGDB the match scores don't change in a significant way, this is what I look at. Maybe my implementation is wrong, but I don't think so because with test-positions it all works fine but with game-play it does not seem to help much.

The point is that without EGDB my program finds after 12 seconds that 37-31 is the best move and with EGDB enabled it takes 26 seconds before it sees that 37-31 is the best move, this is due to the slowdown caused by probing and this can have an adversary effect during game-play.

Usually the errors are made earlier on in the game, for instance the position you show above is already very bad for black and that should not happen when your evaluation and search during the opening and midgame are working like they should.

Joost

Damy do not give me the longest variant but it seems to me that 45 plies is quite high. Could we try to find the best PV?
After 37-31 12-17 22x11 16x07 27-21 10-14 21-16 14-19 Damy continue with 29-23 19-24 23-19 24x13 28-22 15-20 22-17 20-24 17-11 7-12 11-6 24-29 31-26
do you agree up to this point?

[Rein Halbersma]

Hi Joost,

I read on the forum that there is something called 'Killer Draughts' which seems to be less drawish, I don't know the rules and I don't know if there are engines supporting this, when there are more people working on this I will add this variation to my engine as well.

Indeed the rules seem hard to find! I expected them to be immediately visible on http://www.killerdraughts.org but they are nowhere in sight? It's as follows: when the last piece you capture with a king is also a king, you have to stop immediately behind it. In Scan, that's two additional lines of code.

I talk a little bit about the draw rate in this thread: http://fmjd.org/bb3/viewtopic.php?f=53&t=7018

I have interest in killer draughts, which I consider the only promising rule set so far. Scan 2.0, however, does not support it out of the box. I recall that Walter made a version available.

Fabien.

Hi Fabien,

Thanks!

Your description of KD looks very simple to implement, I will take a look at my king-capture generator to see if this rule is easy to add, maybe the easiest and the fastest is to add a second kc-generator for KD only. I guess it will have some consequences for the evaluation-function as well but that is of later concern.

Joost

This is the essential Killer endgame: white to move plays 25-14 and black has to capture 46x10 (stopping immediately behind the white king) and white wins 15x4. This means that almost every two versus one endgame is won for the majority side. Still, I expect the drawing rate to be 50% for top programs, perhaps similar to chess. Michel Grimminck's Dragon program has computed 6 piece endgames for it, stats are here: https://mdgsoft.home.xs4all.nl/draughts ... index.html