Today I had a discussion whether it would be possible to have a real time analysis of the games played for the coming WC match between Schwartzmaan and Georgiev by one or a few of today's draught engines.
The idea is to obtain the data from the livedraughts site (I can help with that) and analyse the position, give the best move, the value of that move and (if possible) the PV for say 8 or 10 plies.
Also you can think of a nice grafic display of the development of the score during the game.
The basic question, is somebody willing to think about it, and if so, to participate to have a real project.
Feike.
Draught Engines to comment WC draughts in June
-
FeikeBoomstra
- Posts: 306
- Joined: Mon Dec 19, 2005 16:48
- Location: Emmen
-
wellnesswrotter
- Posts: 323
- Joined: Mon May 21, 2007 15:10
- Location: www.snukenkuzco.nl
- Contact:
liveComment
ofcourse this is possible.
and it should be very nice to do so!
Also give comments on moves that are not played, but give nice analysis capabilities!
and it should be very nice to do so!
Also give comments on moves that are not played, but give nice analysis capabilities!
-
FeikeBoomstra
- Posts: 306
- Joined: Mon Dec 19, 2005 16:48
- Location: Emmen
During the coming time you can view with the aid of a "live" board the match between Schwartzmann and Georgiev. And with a live chart from Horizon.
http://www.wkdammen2009.nl
http://www.wkdammen2009.nl
-
Wieger Wesselink
- Posts: 1157
- Joined: Sat Jun 28, 2003 13:22
- Location: Eindhoven, The Netherlands
- Contact:
It's a nice feature!FeikeBoomstra wrote:During the coming time you can view with the aid of a "live" board the match between Schwartzmann and Georgiev. And with a live chart from Horizon.
http://www.wkdammen2009.nl
-
FeikeBoomstra
- Posts: 306
- Joined: Mon Dec 19, 2005 16:48
- Location: Emmen
On request of Bert I will explain shortly how the system is set up.
1) From www.livedraughts.com/wk2009/ronde1 (ronde2/ronde3 etc) you can download the file 1.LDF.gz. This file describes the game until now. The layout is quite simple and straightforward. I poll this file every 20 seconds.
2) As explained earlier my GUI is written in Python, so this is a quite simple extension.
Next step is to compare the current state of the game with my saved game state. If new plies are played, I start the Horizon engine to obtain the best move and best move value in that position. (search time 20-60 seconds, 4 cores).
3) In Python there is a module pyChart, to make nice charts. I had to write the x-axis and y-axis module myself. For the y-axis I use a clipped tanh scale, with the points 0, 0.2, 0.5, 1.0, 9.9, these have about an equal distance in between.
The x-axis is auto scaling <10, <25, <50, <75 moves. Above 75, the max value on the x scale is the number of moves.
So form the game (I save the value and best move in the comment field of the ply) I prepare a list of values and the last best move, and pychart makes a nice chart out of it.
4) Save all things and beam the chart up to the website, restart polling.
That's all it took me about 8 hours to finish, most of the time spent 2 problems, I couldn't get FTP working, but the problem was in the firewall of the website, and I had a problem with the trying to read a .gz file image directly from the buffer in memory. I had to save the file first and the reading it.
I am glad I spent the amount of time, I've got a lot of positive reactions, and consider it really as a compliment Ton Sijbrands asked the audience yesterday: and what move expects the computer?
1) From www.livedraughts.com/wk2009/ronde1 (ronde2/ronde3 etc) you can download the file 1.LDF.gz. This file describes the game until now. The layout is quite simple and straightforward. I poll this file every 20 seconds.
2) As explained earlier my GUI is written in Python, so this is a quite simple extension.
Next step is to compare the current state of the game with my saved game state. If new plies are played, I start the Horizon engine to obtain the best move and best move value in that position. (search time 20-60 seconds, 4 cores).
3) In Python there is a module pyChart, to make nice charts. I had to write the x-axis and y-axis module myself. For the y-axis I use a clipped tanh scale, with the points 0, 0.2, 0.5, 1.0, 9.9, these have about an equal distance in between.
The x-axis is auto scaling <10, <25, <50, <75 moves. Above 75, the max value on the x scale is the number of moves.
So form the game (I save the value and best move in the comment field of the ply) I prepare a list of values and the last best move, and pychart makes a nice chart out of it.
4) Save all things and beam the chart up to the website, restart polling.
That's all it took me about 8 hours to finish, most of the time spent 2 problems, I couldn't get FTP working, but the problem was in the firewall of the website, and I had a problem with the trying to read a .gz file image directly from the buffer in memory. I had to save the file first and the reading it.
I am glad I spent the amount of time, I've got a lot of positive reactions, and consider it really as a compliment Ton Sijbrands asked the audience yesterday: and what move expects the computer?
-
Piet Bouma
- Posts: 3574
- Joined: Sun Nov 02, 2003 13:05
- Location: Harlingen
Feike,FeikeBoomstra wrote: I am glad I spent the amount of time, I've got a lot of positive reactions, and consider it really as a compliment Ton Sijbrands asked the audience yesterday: and what move expects the computer?
It is a really nice feature. Compliments for programming it.
Do you also save the charts at each move?
Interesting how many times the computer matches with the choice of the players....
In the future I think, there must/can be - with the new .pdn standard? - Wieger is this possible? - a way to reproduce at each move a chart of a computerprogramm (an url of the .png?).
Also interesting is further analysis (more moves predicted) of the computer and maybe more choices, with scores?
For this match to premature, but in the future......
-
Wieger Wesselink
- Posts: 1157
- Joined: Sat Jun 28, 2003 13:22
- Location: Eindhoven, The Netherlands
- Contact:
I think the most logical place to store such information is inside special comments. For example it might look like this:Piet Bouma wrote: Feike,
It is a really nice feature. Compliments for programming it.
Do you also save the charts at each move?
Interesting how many times the computer matches with the choice of the players....
In the future I think, there must/can be - with the new .pdn standard? - Wieger is this possible? - a way to reproduce at each move a chart of a computerprogramm (an url of the .png?).
Also interesting is further analysis (more moves predicted) of the computer and maybe more choices, with scores?
For this match to premature, but in the future......
32-28 {[%program=horizon value=0.53 depth=15]}
-
FeikeBoomstra
- Posts: 306
- Joined: Mon Dec 19, 2005 16:48
- Location: Emmen
This is the pdn as I saved it, but of course there is no problem at all to have it in a more standardized format.
some remarks:
scale: 1000 = 1 man
value 0 means not calculated
value +1/-1 means draw from calculation
value +2/-2, +3/-3 etc. means draw from (6 pieces) endgame database.
The first three moves I don't try to predict, as that makes no sense at all.
The Result tag is not updated, I will look at that.
[White "A. Georgiev"]
[Black "A. Schwarzman"]
[Result "0-0"]
[Event "WK match 2009"]
[Round "ronde2"]
01.33-28 {0: None } 17-21 {0: None } 02.38-33 {0: None } 21-26 {0: None }
03.42-38 {0: None } 18-23 {0: None } 04.34-29 {0: 23x34} 23x34 {-40: 40x29}
05.39x30 {20: 11-17} 20-25 {-40: 47-42}
06.44-39 {0: 25x34} 25x34 {-20: 40x29}
07.39x30 {60: 11-17} 12-18 {-40: 50-44}
08.43-39 {80: 07-12} 07-12 {-80: 40-34}
09.49-43 {60: 01-07} 11-17 {-60: 40-34} 10.30-24 {0: 19x30} 19x30 {0: 35x24}
11.35x24 {80: 06-11} 01-07 {-80: 40-34}
12.47-42 {80: 14-20} 14-20 {-80: 50-44} 13.50-44 {0: 20x29} 20x29 {0: 33x24}
14.33x24 {120: 10-14} 10-14 {-80: 40-34}
15.40-34 {120: 07-11} 07-11 {-80: 34-29}
16.45-40 {100: 05-10} 17-22 {0: 28x17} 17.28x17 {1: 11x22} 11x22 {20: 32-28}
18.34-29 {100: 05-10} 05-10 {-140: 40-34}
19.39-33 {80: 12-17} 12-17 {-80: 32-28} 20.32-28 {80: 17-21} 17-21 {0: 28x17}
21.28x17 {0: 21x12} 21x12 {-60: 40-34}
22.40-34 {80: 14-19} 14-20 {-60: 44-40}
23.44-40 {120: 20-25} 10-14 {-20: 31-27}
24.31-27 {20: 14-19} 20-25 {40: 24-20} 25.24-20 {0: 15x24} 15x24 {0: 29x20}
26.29x20 {-40: 14-19} 02-07 {60: 37-31}
27.43-39 {-40: 14-19} 14-19 {20: 20-15}
28.20-15 {-20: 18-23} 09-14 {60: 37-31} 29.37-31 {0: 26x37} 26x37 {60: 41x32}
30.41x32 {-60: 18-23} 19-23 {80: 33-28}
31.36-31 {-20: 13-19} 13-19 {40: 33-28}
32.31-26 {20: 07-11} 07-11 {-40: 42-37}
33.40-35 {40: 12-17} 19-24 {-20: 27-22} 34.27-21 {0: 16x27} 16x27 {0: 32x21}
35.32x21 {260: 14-19} 14-19 {-240: 21-16}
36.21-16 {420: 08-13} 08-13 {0: 16x07}
37.16x07 {0: 12x01} 12x01 {-420: 38-32}
38.38-32 {280: 24-30} 06-11 {-280: 42-37}
39.46-41 {420: 24-30} 11-17 {-240: 42-38}
40.32-27 {500: 24-30} 24-30 {0: 35x24}
41.35x24 {0: 19x30} 19x30 {-440: 42-38}
42.42-38 {480: 30-35} 13-19 {-380: 38-32}
43.41-37 {260: 30-35} 19-24 {-160: 48-43}
44.37-32 {380: 03-08} 17-22 {-60: 48-43} 45.48-43 {0: 22x31} 22x31 {0: 26x37}
46.26x37 {720: 18-22} 18-22 {-740: 37-31}
47.37-31 {720: 22-28} 22-28 {0: 33x22}
48.33x22 {720: 24-29} 24-29 {-660: 32-28}
49.32-28 {740: 29x40} 29x40 {0: 28x19}
50.28x19 {660: 40-45} 40-45 {-720: 38-32}
51.38-32 {900: 30-34} 30-34 {0: 39x30}
52.39x30 {0: 25x34} 25x34 {-720: 19-14}
53.19-14 {740: 45-50} 45-50 {-740: 22-18}
54.22-18 {580: 50-39} 04-09 {-600: 15-10}
55.15-10 {0: 09x20} 09x20 {-600: 31-27}
56.10-04 {520: 50-11} 50-11 {-520: 18-13}
57.18-13 {500: 11-07} 20-25 {-40: 13-09}
58.13-09 {0: 03x14} 03x14 {-20: 04-18} 59.04-18 {2: 14-19} 11-17 {-2: 18x45}
0-0
The charts are constructed on the fly, the data is in the pdn, but I don't keep the charts somewhere. And of course not all charts are generated, if there is more than one ply between to LDF samples, I just skip a chart.
some remarks:
scale: 1000 = 1 man
value 0 means not calculated
value +1/-1 means draw from calculation
value +2/-2, +3/-3 etc. means draw from (6 pieces) endgame database.
The first three moves I don't try to predict, as that makes no sense at all.
The Result tag is not updated, I will look at that.
[White "A. Georgiev"]
[Black "A. Schwarzman"]
[Result "0-0"]
[Event "WK match 2009"]
[Round "ronde2"]
01.33-28 {0: None } 17-21 {0: None } 02.38-33 {0: None } 21-26 {0: None }
03.42-38 {0: None } 18-23 {0: None } 04.34-29 {0: 23x34} 23x34 {-40: 40x29}
05.39x30 {20: 11-17} 20-25 {-40: 47-42}
06.44-39 {0: 25x34} 25x34 {-20: 40x29}
07.39x30 {60: 11-17} 12-18 {-40: 50-44}
08.43-39 {80: 07-12} 07-12 {-80: 40-34}
09.49-43 {60: 01-07} 11-17 {-60: 40-34} 10.30-24 {0: 19x30} 19x30 {0: 35x24}
11.35x24 {80: 06-11} 01-07 {-80: 40-34}
12.47-42 {80: 14-20} 14-20 {-80: 50-44} 13.50-44 {0: 20x29} 20x29 {0: 33x24}
14.33x24 {120: 10-14} 10-14 {-80: 40-34}
15.40-34 {120: 07-11} 07-11 {-80: 34-29}
16.45-40 {100: 05-10} 17-22 {0: 28x17} 17.28x17 {1: 11x22} 11x22 {20: 32-28}
18.34-29 {100: 05-10} 05-10 {-140: 40-34}
19.39-33 {80: 12-17} 12-17 {-80: 32-28} 20.32-28 {80: 17-21} 17-21 {0: 28x17}
21.28x17 {0: 21x12} 21x12 {-60: 40-34}
22.40-34 {80: 14-19} 14-20 {-60: 44-40}
23.44-40 {120: 20-25} 10-14 {-20: 31-27}
24.31-27 {20: 14-19} 20-25 {40: 24-20} 25.24-20 {0: 15x24} 15x24 {0: 29x20}
26.29x20 {-40: 14-19} 02-07 {60: 37-31}
27.43-39 {-40: 14-19} 14-19 {20: 20-15}
28.20-15 {-20: 18-23} 09-14 {60: 37-31} 29.37-31 {0: 26x37} 26x37 {60: 41x32}
30.41x32 {-60: 18-23} 19-23 {80: 33-28}
31.36-31 {-20: 13-19} 13-19 {40: 33-28}
32.31-26 {20: 07-11} 07-11 {-40: 42-37}
33.40-35 {40: 12-17} 19-24 {-20: 27-22} 34.27-21 {0: 16x27} 16x27 {0: 32x21}
35.32x21 {260: 14-19} 14-19 {-240: 21-16}
36.21-16 {420: 08-13} 08-13 {0: 16x07}
37.16x07 {0: 12x01} 12x01 {-420: 38-32}
38.38-32 {280: 24-30} 06-11 {-280: 42-37}
39.46-41 {420: 24-30} 11-17 {-240: 42-38}
40.32-27 {500: 24-30} 24-30 {0: 35x24}
41.35x24 {0: 19x30} 19x30 {-440: 42-38}
42.42-38 {480: 30-35} 13-19 {-380: 38-32}
43.41-37 {260: 30-35} 19-24 {-160: 48-43}
44.37-32 {380: 03-08} 17-22 {-60: 48-43} 45.48-43 {0: 22x31} 22x31 {0: 26x37}
46.26x37 {720: 18-22} 18-22 {-740: 37-31}
47.37-31 {720: 22-28} 22-28 {0: 33x22}
48.33x22 {720: 24-29} 24-29 {-660: 32-28}
49.32-28 {740: 29x40} 29x40 {0: 28x19}
50.28x19 {660: 40-45} 40-45 {-720: 38-32}
51.38-32 {900: 30-34} 30-34 {0: 39x30}
52.39x30 {0: 25x34} 25x34 {-720: 19-14}
53.19-14 {740: 45-50} 45-50 {-740: 22-18}
54.22-18 {580: 50-39} 04-09 {-600: 15-10}
55.15-10 {0: 09x20} 09x20 {-600: 31-27}
56.10-04 {520: 50-11} 50-11 {-520: 18-13}
57.18-13 {500: 11-07} 20-25 {-40: 13-09}
58.13-09 {0: 03x14} 03x14 {-20: 04-18} 59.04-18 {2: 14-19} 11-17 {-2: 18x45}
0-0
The charts are constructed on the fly, the data is in the pdn, but I don't keep the charts somewhere. And of course not all charts are generated, if there is more than one ply between to LDF samples, I just skip a chart.
-
FeikeBoomstra
- Posts: 306
- Joined: Mon Dec 19, 2005 16:48
- Location: Emmen
Possible extensions:
1) I would like to present also some alternative plies, but that's very time consuming.
2) What about showing the PV? Will people understand the limitations of a PV.
Or would it be better to show a few possible future moves generated by self play.
To generate the HTML for such an applet is not the problem. Time may be a problem.
3) A very nice thing would be to show the "threat" in a position, but I have no idea how to find "a non-trivial path to loose a piece"
1) I would like to present also some alternative plies, but that's very time consuming.
2) What about showing the PV? Will people understand the limitations of a PV.
Or would it be better to show a few possible future moves generated by self play.
To generate the HTML for such an applet is not the problem. Time may be a problem.
3) A very nice thing would be to show the "threat" in a position, but I have no idea how to find "a non-trivial path to loose a piece"
-
Ed Gilbert
- Posts: 860
- Joined: Sat Apr 28, 2007 14:53
- Real name: Ed Gilbert
- Location: Morristown, NJ USA
- Contact:
Feike, fully agree with Gerard !!
I followed all the games during the weekend, and your engine comment /graphs , really was an additional asset.
Now lets play those people, and see if they still smile in the end [img]images/smilies/icon_smile.gif[/img]
Bert
I followed all the games during the weekend, and your engine comment /graphs , really was an additional asset.
Now lets play those people, and see if they still smile in the end [img]images/smilies/icon_smile.gif[/img]
Bert
-
FeikeBoomstra
- Posts: 306
- Joined: Mon Dec 19, 2005 16:48
- Location: Emmen
-
Piet Bouma
- Posts: 3574
- Joined: Sun Nov 02, 2003 13:05
- Location: Harlingen
For people who are interested:
Sjende Blyn is playing "out of competition" in "Open Gorredijk".
http://toernooibase.kndb.nl/opvraag/sta ... &afko=0&r=
The computer took the lead with 10 pnts out of 5 games.
Sjende Blyn is playing "out of competition" in "Open Gorredijk".
http://toernooibase.kndb.nl/opvraag/sta ... &afko=0&r=
The computer took the lead with 10 pnts out of 5 games.
-
Piet Bouma
- Posts: 3574
- Joined: Sun Nov 02, 2003 13:05
- Location: Harlingen
Meanwhile, Sjende Blyn is a sensation in "Open Gorredijk".Piet Bouma wrote:For people who are interested:
Sjende Blyn is playing "out of competition" in "Open Gorredijk".
http://toernooibase.kndb.nl/opvraag/sta ... &afko=0&r=
The computer took the lead with 10 pnts out of 5 games.
He has beaten the Frisian topplayers J.T. Dekker and A.F. Schotanus.
It is predictable that the computerprogramm will win the tournament.
