I Got youy, Norman Brando

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Ovyron
Posts: 2712
Joined: Tue Jul 03, 2007 2:30 am

Re: I Got youy, Norman Brando

Post by Ovyron » Sat Dec 30, 2017 7:57 am

Michael Sherwin wrote:And yes my code converts a pgn game to the internal format and learns it exactly like Romi would learn one of its own games.
Okay, I finished feeding the 2.2 millions games to Romi, which now has a 1.2GB learn.dat file. However, after typing "pr 0", I get this:

Record 0 sib 47237813 chd 47237811 fs 0 ts 0 t 0 f 0 d 0 s 0 score 0 w 0 l 0 d 4724

Is this normal? Do Record, score, w or l ever increase?

Tobber
Posts: 379
Joined: Fri Sep 28, 2012 3:53 pm
Location: Sweden

Re: I Got youy, Norman Brando

Post by Tobber » Sat Dec 30, 2017 4:27 pm

I must confess I have never used Romi for anything, not even tournaments.
This learn discussion made me curious so I downloaded Romi and tried some fast tournaments under Arena. It worked fine with the internal books in Arena, but not a very strong engine.

Like Ulysses I fed Romi with the 2.2 million pgn and restarted the tournaments. Parameters book_on and learn_on.
Not good at all, Romi now seems to surrender after 7-8 moves, a few games will continue. Tried 3 different opponents but the same behavior. What do I do wrong? And what is the learn file supposed to do? Sorry for the trivial questions but I can't find any readme or other description.

Best
John

carldaman
Posts: 1719
Joined: Sat Jun 02, 2012 12:13 am

Re: I Got youy, Norman Brando

Post by carldaman » Sat Dec 30, 2017 7:43 pm

Tobber wrote:I must confess I have never used Romi for anything, not even tournaments.
This learn discussion made me curious so I downloaded Romi and tried some fast tournaments under Arena. It worked fine with the internal books in Arena, but not a very strong engine.

Like Ulysses I fed Romi with the 2.2 million pgn and restarted the tournaments. Parameters book_on and learn_on.
Not good at all, Romi now seems to surrender after 7-8 moves, a few games will continue. Tried 3 different opponents but the same behavior. What do I do wrong? And what is the learn file supposed to do? Sorry for the trivial questions but I can't find any readme or other description.

Best
John
Once Romi starts learning it will choke up on certain positions and just resign. :( It's a known problem. What I've discovered is that it's not entirely random, and it will resign certain positions all the time, even if the learning file consists of 1 game.

For example: 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 will always cause learning-enabled Romi to resign on my computer. The only way I can test this position with Romi is if Romi has no learning file for that game.

According to Mike Sherwin, this glitch shouldn't happen on Win10 systems, but I don't have such a PC. It is a serious bug, one that seriously hampers testing of the learning feature.

Tobber
Posts: 379
Joined: Fri Sep 28, 2012 3:53 pm
Location: Sweden

Re: I Got youy, Norman Brando

Post by Tobber » Sat Dec 30, 2017 8:12 pm

carldaman wrote:
Tobber wrote:I must confess I have never used Romi for anything, not even tournaments.
This learn discussion made me curious so I downloaded Romi and tried some fast tournaments under Arena. It worked fine with the internal books in Arena, but not a very strong engine.

Like Ulysses I fed Romi with the 2.2 million pgn and restarted the tournaments. Parameters book_on and learn_on.
Not good at all, Romi now seems to surrender after 7-8 moves, a few games will continue. Tried 3 different opponents but the same behavior. What do I do wrong? And what is the learn file supposed to do? Sorry for the trivial questions but I can't find any readme or other description.

Best
John
Once Romi starts learning it will choke up on certain positions and just resign. :( It's a known problem. What I've discovered is that it's not entirely random, and it will resign certain positions all the time, even if the learning file consists of 1 game.

For example: 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 will always cause learning-enabled Romi to resign on my computer. The only way I can test this position with Romi is if Romi has no learning file for that game.

According to Mike Sherwin, this glitch shouldn't happen on Win10 systems, but I don't have such a PC. It is a serious bug, one that seriously hampers testing of the learning feature.
Thank You for the reply. I have a Win10 system so that seems not to be the solution.

Edit: And I tried with command line parameter learn_off in Arena but that didn't help.

/John

Michael Sherwin
Posts: 3041
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: I Got youy, Norman Brando

Post by Michael Sherwin » Sat Dec 30, 2017 8:36 pm

carldaman wrote:
Tobber wrote:I must confess I have never used Romi for anything, not even tournaments.
This learn discussion made me curious so I downloaded Romi and tried some fast tournaments under Arena. It worked fine with the internal books in Arena, but not a very strong engine.

Like Ulysses I fed Romi with the 2.2 million pgn and restarted the tournaments. Parameters book_on and learn_on.
Not good at all, Romi now seems to surrender after 7-8 moves, a few games will continue. Tried 3 different opponents but the same behavior. What do I do wrong? And what is the learn file supposed to do? Sorry for the trivial questions but I can't find any readme or other description.

Best
John
Once Romi starts learning it will choke up on certain positions and just resign. :( It's a known problem. What I've discovered is that it's not entirely random, and it will resign certain positions all the time, even if the learning file consists of 1 game.

For example: 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 will always cause learning-enabled Romi to resign on my computer. The only way I can test this position with Romi is if Romi has no learning file for that game.

According to Mike Sherwin, this glitch shouldn't happen on Win10 systems, but I don't have such a PC. It is a serious bug, one that seriously hampers testing of the learning feature.
Romi never used to do that at all. No change to the code base. And it never does that in windows 10. Not on my computer anyway. I have figured out a fix for the problem. Just going to disable resigning from the learn file result. Guenther Simon suggested that solution and it looks good to me so I'll make a bug fix version and look for a better solution later.
I hate if statements. Pawns demand if statements. Therefore I hate pawns.

carldaman
Posts: 1719
Joined: Sat Jun 02, 2012 12:13 am

Re: I Got youy, Norman Brando

Post by carldaman » Sat Dec 30, 2017 8:56 pm

+1 Thanks a lot, Mike! :D Looking forward to it.

(New Year) Cheers,
CL

Michael Sherwin
Posts: 3041
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: I Got youy, Norman Brando

Post by Michael Sherwin » Sat Dec 30, 2017 8:59 pm

Michael Sherwin wrote:
carldaman wrote:
Tobber wrote:I must confess I have never used Romi for anything, not even tournaments.
This learn discussion made me curious so I downloaded Romi and tried some fast tournaments under Arena. It worked fine with the internal books in Arena, but not a very strong engine.

Like Ulysses I fed Romi with the 2.2 million pgn and restarted the tournaments. Parameters book_on and learn_on.
Not good at all, Romi now seems to surrender after 7-8 moves, a few games will continue. Tried 3 different opponents but the same behavior. What do I do wrong? And what is the learn file supposed to do? Sorry for the trivial questions but I can't find any readme or other description.

Best
John
Romi starts learning it will choke up on certain positions and just resign. :( It's a known problem. What I've discovered is that it's not entirely random, and it will resign certain positions all the time, even if the learning file consists of 1 game.

For example: 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 will always cause learning-enabled Romi to resign on my computer. The only way I can test this position with Romi is if Romi has no learning file for that game.

According to Mike Sherwin, this glitch shouldn't happen on Win10 systems, but I don't have such a PC. It is a serious bug, one that seriously hampers testing of the learning feature.
Romi never used to do that at all. No change to the code base. And it never does that in windows 10. Not on my computer anyway. I have figured out a fix for the problem. Just going to disable resigning from the learn file result. Guenther Simon suggested that solution and it looks good to me so I'll make a bug fix version and look for a better solution later.
Bug fix version is testing now. Will send it out later today, tomorrow at the latest.

About Romi not being a strong engine. My purpose for creating RomiChess was as a sparring partner for humans. I wanted RomiChess to be beatable and have an interesting playing style. The learning in RomiChess is designed to play back a players winning lines against them or to find winning lines if the player does not switch sides. Romi will also repeat its winning lines against the player without using time on the clock. As designed RomiChess does its function very well. This resigning bug from the learn file is something new though that I should have fixed weeks ago so I apologise for that.
I hate if statements. Pawns demand if statements. Therefore I hate pawns.

Michael Sherwin
Posts: 3041
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: I Got youy, Norman Brando

Post by Michael Sherwin » Sat Dec 30, 2017 9:57 pm

Ovyron wrote:
Michael Sherwin wrote:And yes my code converts a pgn game to the internal format and learns it exactly like Romi would learn one of its own games.
Okay, I finished feeding the 2.2 millions games to Romi, which now has a 1.2GB learn.dat file. However, after typing "pr 0", I get this:

Record 0 sib 47237813 chd 47237811 fs 0 ts 0 t 0 f 0 d 0 s 0 score 0 w 0 l 0 d 4724

Is this normal? Do Record, score, w or l ever increase?
Sorry Ulysses, I just saw this post. The first move of the tree is record 1. Record 0 is used for other things. sib 47237813 should be the total number of positions in the tree.
I hate if statements. Pawns demand if statements. Therefore I hate pawns.

User avatar
Ovyron
Posts: 2712
Joined: Tue Jul 03, 2007 2:30 am

Re: I Got youy, Norman Brando

Post by Ovyron » Sun Dec 31, 2017 1:45 am

Thanks Michael, glad to hear a new version of Romi is coming :)

Michael Sherwin
Posts: 3041
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: I Got youy, Norman Brando

Post by Michael Sherwin » Sun Dec 31, 2017 8:23 pm

Michael Sherwin wrote:
Michael Sherwin wrote:
carldaman wrote:
Tobber wrote:I must confess I have never used Romi for anything, not even tournaments.
This learn discussion made me curious so I downloaded Romi and tried some fast tournaments under Arena. It worked fine with the internal books in Arena, but not a very strong engine.

Like Ulysses I fed Romi with the 2.2 million pgn and restarted the tournaments. Parameters book_on and learn_on.
Not good at all, Romi now seems to surrender after 7-8 moves, a few games will continue. Tried 3 different opponents but the same behavior. What do I do wrong? And what is the learn file supposed to do? Sorry for the trivial questions but I can't find any readme or other description.

Best
John
Romi starts learning it will choke up on certain positions and just resign. :( It's a known problem. What I've discovered is that it's not entirely random, and it will resign certain positions all the time, even if the learning file consists of 1 game.

For example: 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 will always cause learning-enabled Romi to resign on my computer. The only way I can test this position with Romi is if Romi has no learning file for that game.

According to Mike Sherwin, this glitch shouldn't happen on Win10 systems, but I don't have such a PC. It is a serious bug, one that seriously hampers testing of the learning feature.
Romi never used to do that at all. No change to the code base. And it never does that in windows 10. Not on my computer anyway. I have figured out a fix for the problem. Just going to disable resigning from the learn file result. Guenther Simon suggested that solution and it looks good to me so I'll make a bug fix version and look for a better solution later.
Bug fix version is testing now. Will send it out later today, tomorrow at the latest.

About Romi not being a strong engine. My purpose for creating RomiChess was as a sparring partner for humans. I wanted RomiChess to be beatable and have an interesting playing style. The learning in RomiChess is designed to play back a players winning lines against them or to find winning lines if the player does not switch sides. Romi will also repeat its winning lines against the player without using time on the clock. As designed RomiChess does its function very well. This resigning bug from the learn file is something new though that I should have fixed weeks ago so I apologise for that.
The bug fix version has tested ok. Now all I have to do is a 32 bit compile and write a little better readme and send it to Graham. I was testing the bug fix version on my old Vista Q4. All I have loaded on it is SF 6 so that is what I was testing against. After 551 games Romi finally learned to beat SF 6.

[pgn][Event "Arena tournament"]
[Site "Q4"]
[Date "2017.12.31"]
[Round "551"]
[White "RomiChessP3m"]
[Black "Stockfish-6-64"]
[Result "1-0"]
[BlackElo "2200"]
[ECO "D05"]
[Opening "Colle"]
[Time "08:32:04"]
[Variation "Zukertort Variation"]
[WhiteElo "2200"]
[TimeControl "10+1"]
[Termination "normal"]
[PlyCount "110"]
[WhiteType "human"]
[BlackType "human"]

1. b3 d5 2. Nf3 Bf5 3. Bb2 Nf6 4. e3 e6 5. d4 Be7 6. Bd3 Bxd3 7. Qxd3 c5 8.
dxc5 Nbd7 9. b4 Rc8 10. Nbd2 b6 11. cxb6 Qxb6 12. a3 O-O 13. Rc1 a5 14. b5
Nc5 15. Qf1 Nfd7 16. Qe2 Na4 17. Bd4 Qb8 18. Qd3 e5 19. Ba1 e4 20. Qd4 Bf6
21. Qxa4 Nc5 22. Qxa5 Bd8 23. Qc3 Nd3+ 24. cxd3 Rxc3 25. Bxc3 exf3 26. Nxf3
Qxb5 27. Ke2 Qb3 28. Bb4 Qa2+ 29. Nd2 Re8 30. d4 Bh4 31. g3 Be7 32. Bxe7
Qb2 33. Rb1 Qc3 34. Rb7 g5 35. Rb3 Qa5 36. Bxg5 Rc8 37. Rd1 Qa6+ 38. Ke1 f6
39. Bf4 Kg7 40. Ra1 h5 41. a4 Rc2 42. Kd1 Rc6 43. f3 Qa5 44. Rb5 Qc3 45.
Ra2 Qd3 46. Rxd5 Kg6 47. a5 Ra6 48. Rb2 Rc6 49. h4 Rc3 50. g4 hxg4 51. fxg4
Ra3 52. Ke1 Ra1+ 53. Kf2 Qa6 54. Rd7 f5 55. Rb6+ Qxb6 {Black resigns} 1-0[/pgn]

How Romi stumbled upon and decided to play the Nimzo-Larsen attack one can only guess without looking at the 550 preceding games. :lol:
I hate if statements. Pawns demand if statements. Therefore I hate pawns.

Post Reply