can romi use both learning and book

[ozziejoe]

I find that if i give romi a book to use, she does not update her learning file. This happens in both chessbase and arena. Can someone confirm?



best
J

[Michael Sherwin]

The interface sends a result command at the end of the game to trigger learning. It may be that the interface does not send the result command when an interface book is being used.

Romi can not use her learning if the game is started from a fen position as all the moves of the game are stored from the original starting position. Also the interface most likely does not send a result command if the game is not started at the beginning.

[David Dahlem]

Hi Michael

After Romi creates a large learn file, how about an option to convert this learn file to a pgn file?



Regards
Dave

[Michael Sherwin]

Hi Michael

After Romi creates a large learn file, how about an option to convert this learn file to a pgn file?



Regards
Dave

Hi David,

What would it be used for?

Study?

Adding lines to another engines book?

Making a specialized book for ...?

Any other reasons?

[David Dahlem]

Hi Michael

After Romi creates a large learn file, how about an option to convert this learn file to a pgn file?



Regards
Dave

Hi David,

What would it be used for?

Study?

Adding lines to another engines book?

Making a specialized book for ...?

Any other reasons?

It could be used for all of these, i had in mind using the pgn to make Arena .abk books, or CB .ctg books.

Regards
Dave

[Eelco de Groot]

Hi Michael

After Romi creates a large learn file, how about an option to convert this learn file to a pgn file?



Regards
Dave

Hi David,

What would it be used for?

Study?

Adding lines to another engines book?

Making a specialized book for ...?

Any other reasons?

It could be used for all of these, i had in mind using the pgn to make Arena .abk books, or CB .ctg books.

Regards
Dave

I don't know if this is already present in RomiChess, but if you would want to make a tool for helping to create opening books, I think a useful feature could be if you could "freeze" certain positions. These would be the positions that Romi trusts well in its opening book and will keep on trying to reach, and any learning will focus especially on the movesequences after these root positions. For instance the position in the Spanish after 1. e4 e5 2. Nf3 Nc6 3. Bb5 f5 could be a starting point, and the program would play on a server to collect responses from White after 3... f5 but would not change the probability of playing 3... f5. Having collected games playing with Black, you would probably have to add a layer of learning, to get good moves with Black to the most common responses from White. You could automate this but it is a time expensive operation to get a quality book this way. But it is possible to get good moves if you are willing to let the program do a deep analysis.
Then it is back to the server to get the booked responses to these moves, or just trust your own analysis if you do not want to give your best lines away in three minute games.

I noticed that often the move that was played by the top GMs if it was not found by a computer as best move, would still after a while come bubbling up in a n-best moves search. Sometimes this can take a day or more I think for really difficult moves, and the progress after this can halt, because the program is missing some positional knowledge for instance. But if the move is both tactically and positionally good but difficult to judge right, it can still get a good ranking in a n-best analysis by the computer.

I wanted to give this example I was analyzing, the game Anand played against Radjabov in Morelia yesterday in the Jaenisch Gambit I think it should be called, that Radjabov often exploits even at this very high super GM level. See also the Chessbase site http://www.chessbase.com/newsdetail.asp?newsid=4464 for an analysis by GM Dorian Rogozenko.


[Event "XXV SuperGM"]
[Site "Morelia/Linares MEX/ESP"]
[Date "2008.02.19"]
[Round "4"]
[White "Anand,V"]
[Black "Radjabov,T"]
[Result "1/2-1/2"]
[WhiteElo "2799"]
[BlackElo "2735"]
[EventDate "2008.02.15"]
[ECO "C63"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 f5 4. d3 fxe4 5. dxe4 Nf6 6. O-O Bc5

[d]r1bqk2r/pppp2pp/2n2n2/1Bb1p3/4P3/5N2/PPP2PPP/RNBQ1RK1 w kq -

My opening book of choice in the Shredder interface made by Harry Schnapp has the move 7. Qe2 here which is probably from the mentioned Carlsen,M (2693)-Radjabov,T (2747)/Porto Vecchio 2007 game, as that was a Blitz game and Anand diverges here, it seems a good point to start analyze here and see what the computer comes up with.

Anand's move 7. Bxc6 wins material but initially it is not liked by the program, I used an older Toga with my own settings. But finally, after almost three hours, the move appears in a two best search as the second best, and after that, on ply 21, 7. Qe2, which was the first choice drops further towards a drawish value, so 7.Bxc6 could come on top!

It is likely that Anand has done a more interactive analysis of his own line, making his own choices but he has probably checked his line also thoroughly with the computer, if only to check what will be the likely responses because the opponents will also have used their computers Like I said the interactive approach works best but the computer can find some good moves on its own these days if it gets enough time!

This was the output, I collected the PVs that had possible answers from White that you might want to check when playing this system from the Black side.

After 6... Bc5 from Radjabov


r1bqk2r/pppp2pp/2n2n2/1Bb1p3/4P3/5N2/PPP2PPP/RNBQ1RK1 w kq -

Engine: Toga II MP Blueberry Winfinder 01 (256 MB)
by Thomas Gaksch and Fabien Letouzey

14 1:22 +0.54 7.Qd3 d6 8.b4 Bb6 9.a4 a5 10.bxa5 Bxa5 11.Bc4 Nb4 12.Qe2 d5 13.exd5 O-O 14.d6+ Nfd5 15.dxc7 Bxc7
16.Nxe5 (78.525.449) 898
14 2:20 +0.36 7.Bxc6 bxc6 8.Nxe5 O-O 9.Nc3 Ba6 10.Nd3 Qe7 11.Bg5 Rab8 12.Qd2 Bd4 13.Rfe1 h6 14.Bh4 Bc4 15.Bxf6 Qxf6 (125.923.783) 898
.
.
.
______________________________________________________________

16 5:50 +0.59 7.Qd3 d6 8.Qc4 Bd7 9.Nc3 Na5 10.Qe6+
Kf8 11.Bxd7 Qxd7 12.Qxd7 Nxd7 13.Nd5 Bb6 14.b3 c6 15.Nxb6 axb6
16.Rd1 Ke7 17.Bg5+ Ke6 18.a4 (293.116.386) 832
16 7:24 +0.46 7.Bg5 d6 8.Nc3 O-O 9.Nd5 Be6 10.Nxf6+ gxf6 11.Bh6 Rf7 12.Qd2 d5 13.Qc3 Qd6 14.Rad1 Kh8 15.Bc4 Rg8 16.Bxd5 (370.518.139) 832
.
.
.
______________________________________________________________

17 9:38 +0.51 7.Qd3 d6 8.Qc4 Bd7 9.Nc3 Na5 10.Qe6+
Kf8 11.Bxd7 Qxd7 12.Qxd7 Nxd7 13.Rd1 Nc6 14.Nd5 Bb6 15.Bg5 Nc5
16.Re1 Ba5 17.c3 Kf7 (478.618.802) 832
17 31:21 -0.07 7.Bd2 O-O 8.Nc3 d6 9.Qe2 Kh8 10.Bxc6
bxc6 11.Na4 Bb6 12.h3 Qe7 13.Qc4 Qe8 14.Ng5 d5 15.Qb4 a5 16.Qa3
(1.567.100.714) 832
.
.
.
______________________________________________________________

18 34:55 +0.52 7.Qd3 d6 8.Qc4 Bd7 9.Nc3 Na5 10.Qe6+
Kf8 11.Bxd7 Qxd7 12.Qxd7 Nxd7 13.Rd1 Bb6 14.Nd5 h6 15.b4 Nc6
16.Be3 Bxe3 17.fxe3 Rc8 18.b5 Ne7 (1.740.880.528) 819
18 50:59 +0.27 7.Bc4 d6 8.Nc3 Na5 9.Bb5+ Bd7 10.Bxd7+
Qxd7 11.Bg5 Nc6 12.Bxf6 gxf6 13.Nd5 O-O-O 14.Nh4 Nb4 15.Nxf6
Qg7 16.a3 Nxc2 17.Nf5 Qxf6 18.Qxc2 (2.506.534.797) 819
______________________________________________________________

18 34:55 +0.52 7.Qd3 d6 8.Qc4 Bd7 9.Nc3 Na5 10.Qe6+ Kf8
11.Bxd7 Qxd7 12.Qxd7 Nxd7 13.Rd1 Bb6
14.Nd5 h6 15.b4 Nc6 16.Be3 Bxe3
17.fxe3 Rc8 18.b5 Ne7 (1.740.880.528)

818

18 52:14 +0.52 7.Qe2 d6 8.Qc4 Bd7 9.Nc3 Na5 10.Qe6+ Kf8
11.Bxd7 Qxd7 12.Qxd7 Nxd7 13.Rd1 Bb6
14.Nd5 h6 15.b4 Nc6 16.Be3 Bxe3
17.fxe3 Rc8 18.b5 Ne7 (2.564.293.883)

818

______________________________________________________________


19 59:47 +0.43 7.Qd3 d6 8.Qc4 Bd7 9.Nc3 Na5 10.Qe6+ Kf8 11.Bxd7 Qxd7 12.Qxd7 Nxd7 13.Rd1 Bb6 14.Nd5 h6 15.Nh4 Kf7 16.Nf5 Nc6 17.Be3 Bxe3 18.Nfxe3 Rac8 19.Nf5 (2.927.336.748) 809
19 94:34 +0.01 7.a3 O-O 8.Nc3 d6 9.Na4 Bb6 10.Qd3 Kh8 11.h3 Qe7 12.Nxb6 axb6 13.Be3 Bd7 14.a4 Be6 15.Ng5 Nb4 16.Qd2 Nc6 17.Nxe6 Qxe6 (4.593.514.190) 809
_______________________________________________________________

19 101:44 +0.47 7.Qe2 d6 8.Qc4 Qe7 9.Nc3 Bd7 10.Bg5 a6 11.Bxf6 Qxf6 12.Nd5 Qd8 13.Bxc6 bxc6 14.Ne3 Qf6 15.b4 Bb6 16.Rfd1 Qe6 17.Qxe6+ Bxe6 18.Ng5 Bd7 19.Nf5 Bxf5 20.exf5 (4.964.252.565) 813
19 59:47 +0.43 7.Qd3 d6 8.Qc4 Bd7 9.Nc3 Na5 10.Qe6+ Kf8 11.Bxd7 Qxd7 12.Qxd7 Nxd7 13.Rd1 Bb6 14.Nd5 h6 15.Nh4 Kf7 16.Nf5 Nc6 17.Be3 Bxe3 18.Nfxe3 Rac8 19.Nf5 (2.927.336.748) 813
_______________________________________________________________

20 192:06 +0.48 7.Qe2 d6 8.Qc4 Qe7 9.Be3 Bd7 10.Bxc5 dxc5
11.Bxc6 Bxc6 12.Nc3 O-O-O 13.Rad1 Kb8
14.Rfe1 Rhf8 15.a4 a6 16.a5 h6 17.h3 g6
18.b3 g5 (9.430.479.741) 818

20 163:02 +0.42 7.Bxc6 bxc6 8.Nxe5 O-O 9.Nc3 d6
10.Nd3 Bd4 11.Ne2 Bb6 12.Bg5 h6
13.Bxf6 Qxf6 14.Nef4 g5 15.Nh5 Qg6
16.Ng3 Be6 17.Qd2 Kh7 18.Qc3 Qe8
19.Rfd1 (8.029.750.379) 818

[Eelco de Groot]

At 21 ply, after 270 minutes, 7.Bxc6 is on top:

r1bqk2r/pppp2pp/2n2n2/1Bb1p3/4P3/5N2/PPP2PPP/RNBQ1RK1 w kq -

Engine: Toga II MP Blueberry Winfinder 01 (Athlon 2009 MHz, 256 MB)
by Thomas Gaksch and Fabien Letouzey

21 270:00 +0.50 7.Bxc6 bxc6 8.Nxe5 O-O 9.Nc3 d6
10.Nd3 Bd4 11.Ne2 Bb6 12.Bg5 h6
13.Bxf6 Qxf6 14.Nef4 Bd7 15.c3 Qg5
16.Qf3 Rae8 17.Rfe1 Be6 18.Qg3 Qxg3
19.hxg3 Bg4 20.a4 (13.330.621.782) 820

21 321:03 +0.41 7.Qe2 d6 8.Qc4 Qe7 9.Be3 Bd7 10.Bxc5 dxc5
11.Bxc6 Bxc6 12.Nc3 O-O-O 13.Rad1 Kb8
14.Rfe1 Rhf8 15.a4 a5 16.h3 Rfe8
17.b3 Rf8 18.Nd5 Qd6 19.Nxf6 Qxd1
20.Rxd1 (15.799.369.042) 820