ENGINE SPEED AND STRONG

Discussion about development of draughts in the time of computer and Internet.
Post Reply
Joost Buijs
Posts: 317
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: ENGINE SPEED AND STRONG

Post by Joost Buijs » Thu Feb 23, 2017 07:09

My guess is that wilo is invariant with time only when you have two very similar versions of the same engine playing each other, when you have two totally different engines playing each other this won't be the case.

ML solely based on win/los ratio derived by self-play does not seem optimal either, I think you have to throw several other engines with totally different characteristics into the mix to get good results, the more diversity the better.

Joost

Catherine
Posts: 129
Joined: Tue Aug 14, 2012 22:24
Real name: Catherine Bourneuf

Re: ENGINE SPEED AND STRONG

Post by Catherine » Fri Feb 24, 2017 04:36

Joost Buijs wrote:My guess is that wilo is invariant with time only when you have two very similar versions of the same engine playing each other, when you have two totally different engines playing each other this won't be the case.

ML solely based on win/los ratio derived by self-play does not seem optimal either, I think you have to throw several other engines with totally different characteristics into the mix to get good results, the more diversity the better.

Joost
Hi Joost and the others,

You said, almost programmers said it also, that draughts is drawish!!! I think it so, but i have some questions and i don't if it can help to increase win average.

1--- There is some positions or some kind of games that force the win?
For example half classical is it stronger than classical or is it the perfect counter game of classical, same thing Roozenburg against Keller system?
So there is some positions to avoid in the opening?

2--- Is it possible to generate a big opening book by self playing or self learning to try to have "perfect moves of most positions that can bring statisticly win?

PS: For Joost, can we, if possible know how the project of the latest version of Argus is going, and when we can have the pleasure to try it?

Thank
Catherine.

Joost Buijs
Posts: 317
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: ENGINE SPEED AND STRONG

Post by Joost Buijs » Fri Feb 24, 2017 08:01

Catherine wrote:
Joost Buijs wrote:My guess is that wilo is invariant with time only when you have two very similar versions of the same engine playing each other, when you have two totally different engines playing each other this won't be the case.

ML solely based on win/los ratio derived by self-play does not seem optimal either, I think you have to throw several other engines with totally different characteristics into the mix to get good results, the more diversity the better.

Joost
Hi Joost and the others,

You said, almost programmers said it also, that draughts is drawish!!! I think it so, but i have some questions and i don't if it can help to increase win average.

1--- There is some positions or some kind of games that force the win?
For example half classical is it stronger than classical or is it the perfect counter game of classical, same thing Roozenburg against Keller system?
So there is some positions to avoid in the opening?

2--- Is it possible to generate a big opening book by self playing or self learning to try to have "perfect moves of most positions that can bring statisticly win?

PS: For Joost, can we, if possible know how the project of the latest version of Argus is going, and when we can have the pleasure to try it?

Thank
Catherine.
Hi Catherine,

Unfortunately I don't know anything about draughts theory, I never studied the game and played maybe 3 draught games in my whole life.

It is indeed possible to generate a big opening book by self-play and to use the statistical information to determine what might be the best move.

Argus is getting along fine, I removed several bugs and it plays very decently at the moment, comparable to the other top-engines but not tuned as good yet, it is a work in progress and I still have to add many things, everything that I add has to be tuned, this is a very tedious process which take weeks. The next thing I want to add is singular extensions, this is not very difficult as such but tuning it is another problem, so this has to wait for a next version of the engine.

At the moment I use very fast games of 1 second per game to verify things, when I start doing other things with my computer (like I'm doing right now) and because I test with SMP enabled it has influence on the win/loss score, so I guess I have to install another computer for the usual things.

I still have to add a simple console interface and pondering, this can be done in a few hours but I find it more interesting to work on the search, that's why I still didn't add these things.

This weekend I hope upload the latest version.

Joost

Catherine
Posts: 129
Joined: Tue Aug 14, 2012 22:24
Real name: Catherine Bourneuf

Re: ENGINE SPEED AND STRONG

Post by Catherine » Fri Feb 24, 2017 09:26

Joost Buijs wrote:
Catherine wrote:
Joost Buijs wrote:My guess is that wilo is invariant with time only when you have two very similar versions of the same engine playing each other, when you have two totally different engines playing each other this won't be the case.

ML solely based on win/los ratio derived by self-play does not seem optimal either, I think you have to throw several other engines with totally different characteristics into the mix to get good results, the more diversity the better.

Joost
Hi Joost and the others,

You said, almost programmers said it also, that draughts is drawish!!! I think it so, but i have some questions and i don't if it can help to increase win average.

1--- There is some positions or some kind of games that force the win?
For example half classical is it stronger than classical or is it the perfect counter game of classical, same thing Roozenburg against Keller system?
So there is some positions to avoid in the opening?

2--- Is it possible to generate a big opening book by self playing or self learning to try to have "perfect moves of most positions that can bring statisticly win?

PS: For Joost, can we, if possible know how the project of the latest version of Argus is going, and when we can have the pleasure to try it?

Thank
Catherine.
Hi Catherine,

Unfortunately I don't know anything about draughts theory, I never studied the game and played maybe 3 draught games in my whole life.

It is indeed possible to generate a big opening book by self-play and to use the statistical information to determine what might be the best move.

Argus is getting along fine, I removed several bugs and it plays very decently at the moment, comparable to the other top-engines but not tuned as good yet, it is a work in progress and I still have to add many things, everything that I add has to be tuned, this is a very tedious process which take weeks. The next thing I want to add is singular extensions, this is not very difficult as such but tuning it is another problem, so this has to wait for a next version of the engine.

At the moment I use very fast games of 1 second per game to verify things, when I start doing other things with my computer (like I'm doing right now) and because I test with SMP enabled it has influence on the win/loss score, so I guess I have to install another computer for the usual things.

I still have to add a simple console interface and pondering, this can be done in a few hours but I find it more interesting to work on the search, that's why I still didn't add these things.

This weekend I hope upload the latest version.

Joost
Hi Joost,

Thank again for taking time to answer so clearly.
So a big opening book can be generated to find best moves by self play ?!! Great
I think that his building must be a gain of time.

Also to you and all the others to permit to draughts community to use engines more stronger to discover others faces of this game.

This do in same time the happiness of us others engine fans.
We will be glad to see the latest version of Argus.
Thank

Catherine

Joost Buijs
Posts: 317
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: ENGINE SPEED AND STRONG

Post by Joost Buijs » Sat Feb 25, 2017 17:09

Hi Catherine,

The last version is on dropbox:

https://www.dropbox.com/s/85fmvcw7nws5v ... a.rar?dl=0

It is now possible to play a game via the console but I didn't implement pondering yet, this has to wait for the next version. I was never my intention to play games via the console so it is more or less hacked in and not very well tested. My plan is to make a simple GUI and to remove the console stuff and configuration file as quick as possible.

There are two problems, sometimes the search can stay away very long on a single move with the consequence that it loses a lot of time on its internal clock. The second problem is that there are some issues with my implementation of Ed's EGDB which takes time to solve properly so I disabled EGDB in this version, it does not influence playing strength very much though.

I will let it rest for a couple of weeks now, because I have enough other things to do.
When you encounter problems just let me know and I will try to fix them.

Joost

Catherine
Posts: 129
Joined: Tue Aug 14, 2012 22:24
Real name: Catherine Bourneuf

Re: ENGINE SPEED AND STRONG

Post by Catherine » Sat Feb 25, 2017 17:59

Joost Buijs wrote:Hi Catherine,

The last version is on dropbox:

https://www.dropbox.com/s/85fmvcw7nws5v ... a.rar?dl=0

It is now possible to play a game via the console but I didn't implement pondering yet, this has to wait for the next version. I was never my intention to play games via the console so it is more or less hacked in and not very well tested. My plan is to make a simple GUI and to remove the console stuff and configuration file as quick as possible.

There are two problems, sometimes the search can stay away very long on a single move with the consequence that it loses a lot of time on its internal clock. The second problem is that there are some issues with my implementation of Ed's EGDB which takes time to solve properly so I disabled EGDB in this version, it does not influence playing strength very much though.

I will let it rest for a couple of weeks now, because I have enough other things to do.
When you encounter problems just let me know and I will try to fix them.

Joost
Hi Joost,

Thank very much for your disponibility of your kindness for taking time to improve Argus for the draughts community. We will test it play against it and you will have a feedback. Thank and God bless you.

PS: please how to run a dxp game and use egdb?

Catherine

Joost Buijs
Posts: 317
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: ENGINE SPEED AND STRONG

Post by Joost Buijs » Sat Feb 25, 2017 19:20

Catherine wrote:
Joost Buijs wrote:Hi Catherine,

The last version is on dropbox:

https://www.dropbox.com/s/85fmvcw7nws5v ... a.rar?dl=0

It is now possible to play a game via the console but I didn't implement pondering yet, this has to wait for the next version. I was never my intention to play games via the console so it is more or less hacked in and not very well tested. My plan is to make a simple GUI and to remove the console stuff and configuration file as quick as possible.

There are two problems, sometimes the search can stay away very long on a single move with the consequence that it loses a lot of time on its internal clock. The second problem is that there are some issues with my implementation of Ed's EGDB which takes time to solve properly so I disabled EGDB in this version, it does not influence playing strength very much though.

I will let it rest for a couple of weeks now, because I have enough other things to do.
When you encounter problems just let me know and I will try to fix them.

Joost
Hi Joost,

Thank very much for your disponibility of your kindness for taking time to improve Argus for the draughts community. We will test it play against it and you will have a feedback. Thank and God bless you.

PS: please how to run a dxp game and use egdb?

Catherine
You can type dxp in the console or edit the argusv107a.cfg file and set 'dxp = yes'. Unfortunately this version has no EGDB.

Joost

Catherine
Posts: 129
Joined: Tue Aug 14, 2012 22:24
Real name: Catherine Bourneuf

Re: ENGINE SPEED AND STRONG

Post by Catherine » Sat Feb 25, 2017 22:26

Joost Buijs wrote:
Catherine wrote:
Joost Buijs wrote:Hi Catherine,

The last version is on dropbox:

https://www.dropbox.com/s/85fmvcw7nws5v ... a.rar?dl=0

It is now possible to play a game via the console but I didn't implement pondering yet, this has to wait for the next version. I was never my intention to play games via the console so it is more or less hacked in and not very well tested. My plan is to make a simple GUI and to remove the console stuff and configuration file as quick as possible.

There are two problems, sometimes the search can stay away very long on a single move with the consequence that it loses a lot of time on its internal clock. The second problem is that there are some issues with my implementation of Ed's EGDB which takes time to solve properly so I disabled EGDB in this version, it does not influence playing strength very much though.

I will let it rest for a couple of weeks now, because I have enough other things to do.
When you encounter problems just let me know and I will try to fix them.

Joost
Hi Joost,

Thank very much for your disponibility of your kindness for taking time to improve Argus for the draughts community. We will test it play against it and you will have a feedback. Thank and God bless you.

PS: please how to run a dxp game and use egdb?

Catherine
You can type dxp in the console or edit the argusv107a.cfg file and set 'dxp = yes'. Unfortunately this version has no EGDB.

Joost
Thank very much Joost,
You will get the feedback.

Catherine

MichelG
Posts: 244
Joined: Sun Dec 28, 2003 20:24
Contact:

Re: ENGINE SPEED AND STRONG

Post by MichelG » Sun Feb 26, 2017 01:39

Joost Buijs wrote:Hi Catherine,

The last version is on dropbox:

[...]

It is now possible to play a game via the console but I didn't implement pondering yet, this has to wait for the next version. I was never my intention to play games via the console so it is more or less hacked in and not very well tested. My plan is to make a simple GUI and to remove the console stuff and configuration file as quick as possible.

There are two problems, sometimes the search can stay away very long on a single move with the consequence that it loses a lot of time on its internal clock. The second problem is that there are some issues with my implementation of Ed's EGDB which takes time to solve properly so I disabled EGDB in this version, it does not influence playing strength very much though.

I will let it rest for a couple of weeks now, because I have enough other things to do.
When you encounter problems just let me know and I will try to fix them.

Joost
Hi Joost,

Fabien pointed out earlier tonight that the rar-file contains the evaluation function data from Scan 2.0, reordered and encrypted. I can confirm that this is indeed the case.

Sharing the program in this fashion violates the copyright rules under which Scan was published. Please respect Scan's license, which is the GNU GENERAL PUBLIC LICENSE, see https://www.gnu.org/licenses/gpl-3.0.en.html.

Michel

Joost Buijs
Posts: 317
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: ENGINE SPEED AND STRONG

Post by Joost Buijs » Sun Feb 26, 2017 08:59

MichelG wrote:
Joost Buijs wrote:Hi Catherine,

The last version is on dropbox:

[...]

It is now possible to play a game via the console but I didn't implement pondering yet, this has to wait for the next version. I was never my intention to play games via the console so it is more or less hacked in and not very well tested. My plan is to make a simple GUI and to remove the console stuff and configuration file as quick as possible.

There are two problems, sometimes the search can stay away very long on a single move with the consequence that it loses a lot of time on its internal clock. The second problem is that there are some issues with my implementation of Ed's EGDB which takes time to solve properly so I disabled EGDB in this version, it does not influence playing strength very much though.

I will let it rest for a couple of weeks now, because I have enough other things to do.
When you encounter problems just let me know and I will try to fix them.

Joost
Hi Joost,

Fabien pointed out earlier tonight that the rar-file contains the evaluation function data from Scan 2.0, reordered and encrypted. I can confirm that this is indeed the case.

Sharing the program in this fashion violates the copyright rules under which Scan was published. Please respect Scan's license, which is the GNU GENERAL PUBLIC LICENSE, see https://www.gnu.org/licenses/gpl-3.0.en.html.

Michel
Hi Michel,

First of all it is no secret that the current evaluation function of the program is based on scans 4x4 patterns (albeit coded in a totally different way) and that I temporary derived the evaluation values from scans eval data, as everybody knows I don't have my own ML routines up and running yet. If you can tell me what I have to do to adhere to the GPL that will be fine, I've been reading it throughout and for me it is not clear how to interpret it.

Secondly it was never my intention to share the program in any way at least not till I have my own ML up and running, that I did this because people are so eager to test the program is possibly a mistake so I will remove the program from 'Dropbox'.

I wonder why Fabien needs you to tell me this and why he doesn't tell me this himself?

Maybe a good time to stop with draughts all together because I have the feeling it is a 'dead end street' anyway.

Joost

Fabien Letouzey
Posts: 283
Joined: Tue Jul 07, 2015 07:48
Real name: Fabien Letouzey

Re: ENGINE SPEED AND STRONG

Post by Fabien Letouzey » Sun Feb 26, 2017 09:39

Joost Buijs wrote:First of all it is no secret that the current evaluation function of the program is based on scans 4x4 patterns (albeit coded in a totally different way) and that I temporary derived the evaluation values from scans eval data, as everybody knows I don't have my own ML routines up and running yet. If you can tell me what I have to do to adhere to the GPL that will be fine, I've been reading it throughout and for me it is not clear how to interpret it.
We can easily check that it's no secret. Let's ask Krzysztof, Catherine, and Sidiki (not to mention others) whether it was 100% clear that Argus was actually using Scan's evaluation function (and not just the idea of 4x4 patterns which would have been OK). Or maybe I'm clearly listed as co-author? I don't have Windows.
I wonder why Fabien needs you to tell me this and why he doesn't tell me this himself?
Because the proof was above the level of the average reader here. Michel took the initiative to publish his own confirmation, leaving no room for doubt.

Joost Buijs
Posts: 317
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: ENGINE SPEED AND STRONG

Post by Joost Buijs » Sun Feb 26, 2017 10:16

Fabien Letouzey wrote:
Joost Buijs wrote:First of all it is no secret that the current evaluation function of the program is based on scans 4x4 patterns (albeit coded in a totally different way) and that I temporary derived the evaluation values from scans eval data, as everybody knows I don't have my own ML routines up and running yet. If you can tell me what I have to do to adhere to the GPL that will be fine, I've been reading it throughout and for me it is not clear how to interpret it.
We can easily check that it's no secret. Let's ask Krzysztof, Catherine, and Sidiki (not to mention others) whether it was 100% clear that Argus was actually using Scan's evaluation function (and not just the idea of 4x4 patterns which would have been OK). Or maybe I'm clearly listed as co-author? I don't have Windows.
I wonder why Fabien needs you to tell me this and why he doesn't tell me this himself?
Because the proof was above the level of the average reader here. Michel took the initiative to publish his own confirmation, leaving no room for doubt.
The program temporarily uses your 4x4 evaluation data and not your evaluation function as a whole, it is my intention to have this data replaced pretty soon because I'm going to switch to larger patterns anyway. Maybe this was not clear to the average reader, well now it is.

Now I have to rethink whether I will continue with the project or not, for me the fun is programming and I'm not looking after recognition of any kind, which seems very important for you. I already had my doubts about putting so much effort into a game that appears to be a 'black hole'. Maybe all that effort is a waste and I rather could have spend it on my chess program that I didn't touch for four years.

Joost
Last edited by Joost Buijs on Sun Feb 26, 2017 10:34, edited 1 time in total.

Fabien Letouzey
Posts: 283
Joined: Tue Jul 07, 2015 07:48
Real name: Fabien Letouzey

Re: ENGINE SPEED AND STRONG

Post by Fabien Letouzey » Sun Feb 26, 2017 10:31

Joost Buijs wrote:The program temporarily uses your 4x4 evaluation data and not your evaluation function as a whole, it is my intention to have this data replaced pretty soon because I'm going to switch to larger patterns anyway. Maybe this was not clear to the average reader, well now it is.
For the record, what is the official explanation about the AAAA XORing?

Joost Buijs
Posts: 317
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: ENGINE SPEED AND STRONG

Post by Joost Buijs » Sun Feb 26, 2017 11:02

Fabien Letouzey wrote:
Joost Buijs wrote:The program temporarily uses your 4x4 evaluation data and not your evaluation function as a whole, it is my intention to have this data replaced pretty soon because I'm going to switch to larger patterns anyway. Maybe this was not clear to the average reader, well now it is.
For the record, what is the official explanation about the AAAA XORing?
I remember I tried this during a test to verify if the checksum worked, after this I was probably to lazy to generate an other table, at least this is not meant to 'encrypt' things like you seem to suggest, if I wanted to encrypt things I would have thrown a random function over it.

Rein Halbersma
Posts: 1625
Joined: Wed Apr 14, 2004 16:04
Contact:

Re: ENGINE SPEED AND STRONG

Post by Rein Halbersma » Sun Feb 26, 2017 11:23

Joost Buijs wrote: The program temporarily uses your 4x4 evaluation data and not your evaluation function as a whole, it is my intention to have this data replaced pretty soon because I'm going to switch to larger patterns anyway. Maybe this was not clear to the average reader, well now it is.

Now I have to rethink whether I will continue with the project or not, for me the fun is programming and I'm not looking after recognition of any kind, which seems very important for you. I already had my doubts about putting so much effort into a game that appears to be a 'black hole'. Maybe all that effort is a waste and I rather could have spend it on my chess program that I didn't touch for four years.
I think the draughts community is so small and everyone is in it for the fun.

But just consider Fabien's position: he already had one piece of labor stolen from him (Fruit). And now you carelessly use and distribute his essential draughts contribution (Scan eval) without proper attribution. Can you imagine how that might spoil his fun?

Suppose you enter a draughts tournament, what would you propose as a procedure to convince others of the fact that you are entering with an original program?

Post Reply