Huh, learning, what is it good for? Absolutly ...

[Tony Thomas]

Howdy,

I will give you some unscientific feedback. I downloaded RomiP3jx64 to test against LearningLemming using the Manhattan book some time ago. I don't have the numbers, but Romi very clearly improved its play dramatically from its learning, much more than I would have expected given what I know of your technique. It went from losing almost all its games to being quite competitive pretty quickly...maybe 20 games or so. This was true even when I had my learning on also. The improvement did not seem to transfer over that well if I switched time controls, however. I am not sure if that was only true when switching to a slower time control, or was bidirectional. I also never tried Romi against other opponents, so I am not sure how well its learning generalizes.

One thing I did note is learning seemed more effective in openings that were prone to positional maneuvering with lots of transpositions (like c4/g3 systems) than anything else. If there is a more recent version / opening book let me know and I will try to do a little testing and post results if you want.

Your nodes per second are amazing by the way. Much much better than LearningLemming.

By the way x 2, Romi's presence is missed in the ACCA events. Any reason why you haven't been entering?

-Sam

I am not Michael Sherwin, but I can answer few things. Yes, there is a new book, made by our own Wael. Yes, there are recent versions, I have the version made two days ago, if you could pm me your email address, I will send it to you. I dont think Romi's nodes/second is as amazing as it used to be, it used to search about a 1.2 million or more nodes/second with the old versions, currently it searches only half of that at the same time control and same computer. I have tried Romi against 12 different opponents in 30 game matches, when learning was on Romi used to kill the weaker opponents, sometimes scoring as much as 95% by using the same winning line over and over.

Romi doesnt compete in ACCA because he cant find an operator. If I had a decent computer I would have volunteered, but I am not interested in seeing Romi getting her tail whopped on my celeron 2800 that is extremely slow.

[BubbaTough]

I have pm'ed you my email. If you are interested I will try it out in a fairly long match with learning enabled and send you the results. I would hope LearningLemming does no start losing 95% of the games, otherwise it is misnamed. If its just a computer between you an ACCA events I might be able to operate for you on a semi-decent computer, we can talk about it when the next one roles around.

-Sam

[Tony Thomas]

I have pm'ed you my email. If you are interested I will try it out in a fairly long match with learning enabled and send you the results. I would hope LearningLemming does no start losing 95% of the games, otherwise it is misnamed. If its just a computer between you an ACCA events I might be able to operate for you on a semi-decent computer, we can talk about it when the next one roles around.

-Sam

I will get Michael to join if you can be the operator. I will send you the engine and learn file in few minutes. I will also include commands, currently I think learning is default, but I have options to turn it off, on, use it etc..I think that Lemming has gained a lot recently from watching its games as opposed to Romi, she is going through a rather secret rewrite. She will get all your asses back for the embarassment. Note that the current experimental version is called Goatka 1.0beta due to recent goat joke exchanges between me and Michael.

[Michael Sherwin]

Howdy,

I will give you some unscientific feedback. I downloaded RomiP3jx64 to test against LearningLemming using the Manhattan book some time ago. I don't have the numbers, but Romi very clearly improved its play dramatically from its learning, much more than I would have expected given what I know of your technique. It went from losing almost all its games to being quite competitive pretty quickly...maybe 20 games or so. This was true even when I had my learning on also. The improvement did not seem to transfer over that well if I switched time controls, however. I am not sure if that was only true when switching to a slower time control, or was bidirectional. I also never tried Romi against other opponents, so I am not sure how well its learning generalizes.

One thing I did note is learning seemed more effective in openings that were prone to positional maneuvering with lots of transpositions (like c4/g3 systems) than anything else. If there is a more recent version / opening book let me know and I will try to do a little testing and post results if you want.

Your nodes per second are amazing by the way. Much much better than LearningLemming.

By the way x 2, Romi's presence is missed in the ACCA events. Any reason why you haven't been entering?

-Sam

Hi Sam,

Thanks for the feedback!

The biggest reason that I have not entered the ACCA tournaments is that both my parents have dimentia and that is a constant battle that I am fighting. Having a consistant and repeating operator would make it appealing for me to have Romi compete. Romi's previous operators did a fantastic job and I thank them, but they have not volunteered for more events and I did not want to impose upon them to do so.

The learning has not changed since the last time that you tested it, however, Romi may have gotten a bit stronger. So running more learning test between two engines that have gotten stronger may be interesting. Note that Goatka is a highly experimental beta. That is another reason for the alternate name.

I should have an official release of RomiChess within a couple of weeks, but that is assuming that something goes right for a change!

Best,
Mike

[Michael Sherwin]

Here is an interesting little game by the newest Romi beta aginst Hamsters2 at 4+4 time limit. I do not know how good or bad Romi played, but if you look at the search depths then an idea of where Romi is headed may be seen!

[Event "Arena tournament"]
[Site "A34"]
[Date "2007.11.15"]
[Round "3"]
[White "RomiChessNG8"]
[Black "Hamsters2"]
[Result "1-0"]
[ECO "A90"]
[Opening "Dutch Defence"]
[Time "06:37:09"]
[Variation "120+2"]
[TimeControl "240+4"]
[Termination "normal"]
[PlyCount "102"]
[WhiteType "program"]
[BlackType "program"]


{C:\Chess\Pgn\Sherwin50.pgn 2/50 (Owner-Owner, ?, 2007.04.04, *)}
1. d4 f5
2. c4 Nf6
3. g3 e6
4. Bg2 Ne4

{(Nf6e4 Nb1c3 Bf8b4 Qd1d3 d7d5 a2a3 Bb4e7 Ng1f3 00 Bc1f4) -0.21/10 12}

5. Bxe4

{(g2e4f5e4 b1c3 d7d5 g1h3 f8e7 e1g1 e8g8 c4d5 e6d5 h3f4 c7c6 d1b3 ) 0.30/15
16}

5. ... fxe4

{(f5xe4 d4d5 Qd8f6 Nb1c3 Bf8b4 Qd1b3 Nb8a6 a2a3 Bb4a5 Bc1f4 e6xd5 c4xd5 00)
0.08/10 12}

6. Nc3

{(b1c3d7d5 e2e3 f8b4 g1e2 b8c6 a2a3 b4c3 e2c3 d5c4 e1g1 e6e5 d4d5 c6e7 d1a4
c8d7 a4c4 d7h3 f1d1 h3g4 ) 0.47/15 15}

6. ... d5

{(d7d5 Qd1b3 c7c6 Bc1f4 Bf8e7 c4c5 00 Ke1f1 Qd8a5 Ra1d1) 0.09/10 11}

7. e3

{(e2e3 g7g6 g1e2 c7c6 e2f4 f8d6 c1d2 e8g8 e1g1 c8d7 g1g2 g6g5 ) 0.21/14 14}

7. ... c6

{(c7c6 Ng1h3 Bf8e7 00 00 Nh3f4 Qd8b6 c4c5 Qb6b4 Qd1c2) 0.20/10 11}

8. f3

{(f2f3 f8b4 f3e4 d5c4 g1f3 b8d7 e1g1 e8g8 h2h3 d7f6 d1c2 b7b6 g1h2 c8b7
f3e5 ) 0.13/13 13}

8. ... exf3

{(e4xf3 Ng1xf3 Bf8e7 00 00 Bc1d2 d5xc4 Nf3e5 Rf8xf1+ Qd1xf1 b7b5 b2b3
Be7d6) 0.57/11 11}

9. Nxf3

{(g1f3 f8e7 e3e4 e8g8 c1f4 d5c4 e1g1 g7g5 f4e3 b8d7 d1c2 g8g7 c2f2 )
0.12/14 12}

9. ... Be7

{(Bf8e7 00 00 Qd1d3 Qd8e8 b2b3 Qe8h5 Nf3e5 Nb8d7 Rf1xf8+ Be7xf8 Ne5xd7
Bc8xd7) 0.48/11 11}

10. e4

{(e3e4d5c4 e1g1 e8g8 g1g2 b8d7 c1e3 d8b6 d1c2 b6a5 a1c1 b7b6 h2h3 c8b7 g2h2
) 0.06/14 13}

10. ... dxc4

{(d5xc4 Nf3e5 00 Ne5xc4 b7b5 Nc4e5 Qd8b6 Bc1e3 Nb8d7 Qd1h5 Nd7xe5 Qh5xe5)
0.55/11 10}

11. O-O

{(e1g1e8g8 g1g2 b8d7 ) 0.08/14 13}

11. ... O-O

{(00 Bc1e3 Nb8d7 Nf3d2 b7b5 Qd1g4 Nd7f6 Qg4h4 h7h5) 0.73/9 10}

12. Be3

{(c1e3 b8d7 f3d2 b7b5 f1f8 d8f8 e4e5 c8b7 d1e2 ) -0.29/14 10}

12. ... Nd7

{(Nb8d7 Qd1d2 b7b5 a2a3 h7h6 d4d5 e6xd5 Nc3xd5 Be7d6 Kg1h1) 0.80/10 10}

13. Ne2

{(c3e2d7f6 d1c2 b7b5 a1c1 c8b7 e2f4 c6c5 f4e6 ) -0.57/13 13}

13. ... Qe8

{(Qd8e8 Ne2f4 Be7d8 Qd1e2 b7b5 a2a4 Bc8b7 a4xb5 c6xb5 d4d5) 0.89/9 10}

14. Qc2

{(d1c2b7b5 e2f4 d7f6 a1e1 f6g4 e3d2 c6c5 h2h3 g4f6 d2a5 c5d4 f3d4 e7d6 )
-0.27/14 12}

14. ... b5

{(b7b5 Ne2f4 Be7d8 Ra1d1 Bc8b7 b2b3 Bb7a6 Nf3e5 Nd7xe5 d4xe5 c4xb3 a2xb3)
0.95/9 9}

15. Nf4

{(e2f4d7f6 h2h3 c6c5 f3g5 e6e5 d4e5 f6h5 f4h5 e8h5 ) -0.20/14 12}

15. ... Bd8

{(Be7d8 e4e5 Bc8b7 Nf4h3 Rf8f5 Nf3h4 Bd8xh4 Rf1xf5 e6xf5 g3xh4) 0.92/10 9}

16. Ne2

{(f4e2c8b7 a1c1 ) -0.58/13 10}

16. ... Bb7

{(Bc8b7 Ne2f4 Ra8c8 a2a4 a7a6 Ra1d1 Nd7b6 a4xb5 c6xb5 Nf3e5) 1.00/9 9}

17. b3

{(b2b3c4b3 c2b3 c6c5 f3g5 d8g5 f1f8 d7f8 e3g5 b7e4 d4c5 a8b8 b3e3 e8c6 a1f1
) -0.60/14 13}

17. ... Ba6

{(Bb7a6 Ne2f4 Rf8f7 e4e5 Rf7f5 Be3d2 c4xb3 a2xb3 Ba6b7) 0.98/8 9}

18. Bf4

{(e3f4 c4b3 a2b3 a6b7 f4d6 d8e7 d6e7 e8e7 a1c1 a8d8 h2h3 e6e5 c1b1 e7d6 )
-0.40/13 7}

18. ... Qh5

{(Qe8h5 a2a4 Nd7b6 b3xc4 b5xc4 Nf3e5 Ra8c8 g3g4 Qh5h3 Ne2c3) 1.00/9 9}

19. bxc4

{(b3c4 b5c4 e2c3 d8a5 a1c1 ) -0.56/13 8}

19. ... bxc4

{(b5xc4 Ne2c3 Bd8a5 Ra1b1 c6c5 Qc2a4 Ba5xc3 Qa4xa6 Bc3xd4+ Nf3xd4 c5xd4
Qa6xe6+ Qh5f7) 1.06/10 9}

20. Nc3

{(e2c3h5g4 c2g2 c6c5 d4d5 e6d5 e4d5 d8f6 a1c1 g7g5 f4d6 f6d4 f3d4 g4d4 g1h1
f8f1 g2f1 a8d8 f1f5 g5g4 ) -0.55/15 10}

20. ... Bb6

{(Bd8b6 Qc2a4 Ba6b5 Nc3xb5 c6xb5 Qa4b4 Nd7f6 Qb4d6 e6e5) 1.09/9 8}

21. Qa4

{(c2a4a6b5 a4d1 b5a6 a1c1 a8d8 g1h1 d7c5 f3e5 h5d1 c1d1 c5d3 ) -0.49/15 9}

21. ... Bb5

{(Ba6b5 Qa4d1 Bb6a5 Qd1c2 Ba5b4 Ra1b1 a7a5 Kg1h1 Rf8f7 Nc3xb5 c6xb5) 1.06/9
8}

22. Qd1

{(a4d1d7f6 f4d6 f8d8 d6e7 d8d7 e7f6 g7f6 g1g2 b6a5 d1c2 a8d8 a1b1 b5a6 h2h3
c6c5 ) -0.72/15 9}

22. ... Nf6

{(Nd7f6 Nf3e5 Ra8d8 Nc3xb5 c6xb5 Ne5c6 Nf6g4 h2h3 Qh5xh3 Nc6xd8 Rf8xd8)
1.09/9 26}

23. Bd6

{(f4d6f8d8 e4e5 f6e8 d6b4 b5a6 b4e7 ) -0.81/15 9}

23. ... Rfd8

{(Rf8d8 e4e5 Qh5h6 Qd1c1 Qh6xc1 Rf1xc1 Rd8d7 Ra1b1 a7a6 Rb1b2) 0.92/9 7}

24. e5

{(e4e5 h5h6 d1c1 h6c1 f1c1 f6d5 a2a4 d5c3 c1c3 b6a5 c3c2 b5a6 ) -0.55/14 6}

24. ... Qh6

{(Qh5h6 Qd1e2 Bb5a6 a2a4 Rd8d7 a4a5 Bb6d8 Nf3d2 Nf6d5) 0.72/9 7}

25. Qe2

{(d1e2 f6d5 c3d5 c6d5 a2a4 b5a6 a4a5 c4c3 e2a6 b6d4 f3d4 h6e3 g1h1 )
-0.05/15 7}

25. ... Ba6

{(Bb5a6 a2a4 Rd8d7 a4a5 Bb6d8 Nf3d2 Qh6h5 Qe2xh5 Nf6xh5) 0.66/9 7}

26. a4

{(a2a4f6e8 a4a5 b6c7 d6a3 g7g6 a3e7 d8d7 e7g5 h6g7 e2e4 c7d8 e4c6 e8c7 a1b1
d8g5 ) 0.77/15 8}

26. ... Ba5

{(Bb6a5 e5xf6 Rd8xd6 Qe2e5 Ba5xc3 Qe5xd6 Qh6e3+ Kg1h1 Bc3xa1 Qd6xc6 Ra8b8
Qc6xa6) 0.20/9 7}

27. exf6

{(e5f6 d8d6 ) 1.53/15 7}

27. ... Rxd6

{(Rd8xd6 Qe2e5 Ba5xc3 Qe5xd6 Bc3xa1 Qd6xe6+ Kg8h8 Qe6xc6 Ba1xd4+ Nf3xd4
Qh6e3+ Rf1f2 Qe3e1+ Kg1g2) -0.12/10 7}

28. Ne4

{(c3e4d6d5 f3g5 a6c8 f6f7 g8f8 e2c4 c8d7 h2h4 a8b8 a1b1 b8b6 c4d3 a5b4 d3e3
h6g6 ) 1.42/15 8}

28. ... Rd5

{(Rd6d5 f6xg7 Ba5b6 a4a5 Rd5xa5 Ne4c5 Ba6c8 Qe2xc4 Qh6e3+ Kg1g2 Kg8xg7)
0.40/8 7}

29. Nfg5

{(f3g5) 1.74/14 7}

29. ... Bc8

{(Ba6c8 Qe2xc4 Ra8b8 Qc4xc6 Bc8d7 Qc6c1 Rd5f5 f6f7+ Kg8h8 Kg1h1) -0.69/9 7}

30. f7+

{(f6f7 g8f8 e2c4 c8d7 h2h4 h6g6 a1b1 a5d8 c4b4 d8e7 b4b7 a8d8 b7a7 e7f6
e4f6 g7f6 a7b6 ) 2.46/14 6}

30. ... Kf8

{(Kg8f8 Qe2xc4 Ra8b8 Ra1b1 Rb8xb1 Qc4xc6 Rb1xf1+ Kg1xf1 Bc8a6+ Kf1g2 Ba6f1+
Kg2h1 Bf1b5 Ng5xe6+ Kf8xf7 Qc6xd5 Qh6xe6 Qd5xe6+ Kf7xe6 a4xb5) -0.98/10 7}

31. Qxc4

{(e2c4 c8d7 h2h4 h6g6 a1b1 a5d8 b1b7 d8g5 e4g5 a8c8 b7a7 ) 2.74/15 5}

31. ... Bd7

{(Bc8d7 h2h4 Kf8e7 Ra1b1 Ra8f8 Rb1b7 e6e5 d4xe5 Ba5b6+ Kg1h1 Rd5xe5)
-1.42/9 7}

32. h4

{(h2h4a5d8 a1b1 h6g6 b1b7 d8g5 e4g5 a7a5 b7d7 d5d7 c4c6 ) 2.74/15 8}

32. ... Bd8

{(Ba5d8 Qc4b4+ Bd8e7 Qb4b7 Ra8d8 Qb7xa7 Rd5f5 Ne4c5 Bd7c8 Rf1xf5 e6xf5)
-1.75/9 6}

33. Rab1

{(a1b1d8e7 b1b7 ) 2.58/15 7}

33. ... Be7

{(Bd8e7 Rb1b7 Ra8d8 Rb7xa7 Qh6g6 g3g4 Qg6h6 Ne4c3 Qh6xh4 Nc3xd5 Qh4xg4+
Kg1h1 Be7xg5) -1.60/9 11}

34. Rb7

{(b1b7 a8d8 b7d7 d8d7 c4c6 e7g5 e4g5 d5d6 c6e4 d7d8 g5h7 f8e7 f7f8q d8f8
h7f8 d6d5 f8g6 e7d8 g6e5 h6d2 ) 5.86/14 5}

34. ... Rd8

{(Ra8d8 Rb7xa7 e6e5 d4xe5 Rd5xe5 Ra7c7 Re5d5 Qc4c3 Rd8a8 Qc3a1) -2.07/9 6}

35. Rxd7

{(b7d7d8d7 c4c6 e7g5 e4g5 d5d6 c6e4 d7d8 g5h7 f8e7 f7f8q d8f8 h7f8 d6d5 )
5.86/17 7}

35. ... R8xd7

{(Rd8xd7 Qc4xc6 Be7xg5 Ne4xg5 Rd5d6 Qc6c2 Rd6xd4 Ng5xe6+ Qh6xe6 Qc2c8+
Rd7d8 Qc8xe6 Rd4xa4) -5.53/11 6}

36. Qxc6

{(c4c6 e7g5 e4g5 d5d6 c6e4 d7d8 g5h7 f8e7 f7f8q d8f8 h7f8 d6d5 g3g4 g7g5
h4g5 h6f8 f1f8 e7f8 g5g6 d5d7 e4e6 ) 6.97/16 5}

36. ... Bxg5

{(Be7xg5 Ne4xg5 Rd5d6 Qc6c2 Rd7d8 Ng5xh7+ Kf8e7 Qc2c7+ Rd6d7 Qc7c5+ Rd7d6
Qc5xa7+ Rd6d7 Qa7c5+ Rd7d6 f7f8Q+ Rd8xf8 Nh7xf8) -5.88/11 6}

37. Nxg5

{(e4g5) 7.29/16 7}

37. ... R5d6

{(Rd5d6 Qc6e4 Rd7d8 d4d5 Rd6xd5 Ng5xe6+ Qh6xe6 Qe4xe6 Rd5d1 Kg1g2 Rd8d2+
Rf1f2 Rd2xf2+ Kg2xf2 g7g6) -6.17/11 6}

38. Qe4

{(c6e4) 7.22/16 6}

38. ... Rd8

{(Rd7d8 d4d5 Rd6xd5 Ng5xe6+ Qh6xe6 Qe4xe6 Rd5d1 Rf1xd1 Rd8xd1+ Kg1f2 Rd1d8
Kf2e3 Rd8d1) -6.22/11 6}

39. Nxh7+

{(g5h7f8e7 f7f8q d8f8 h7f8 d6d5 f8g6 e7d6 g6e5 d6e7 f1f7 ) 7.60/16 6}

39. ... Ke7

{(Kf8e7 Qe4f3 g7g5 f7f8Q+ Rd8xf8 Nh7xf8 g5xh4 Qf3b7+ Ke7e8 Qb7c8+ Ke8e7
Qc8c7+ Ke7e8 Qc7xd6 Qh6e3+ Kg1g2) -8.81/11 6}

40. f8=Q+

{(f7f8qd8f8 h7f8 ) 7.62/16 6}

40. ... Rxf8

{(Rd8xf8 Nh7xf8 Rd6d5 g3g4 Ke7d6 g4g5 Rd5xg5+ h4xg5 Qh6xg5+ Kg1h2 Qg5d5
Qe4g4 g7g5) -7.57/11 5}

41. Nxf8

{(h7f8) 8.24/14 6}

41. ... Rd5

{(Rd6d5 g3g4 a7a5 g4g5 Rd5xg5+ h4xg5 Qh6xg5+ Kg1h2 Qg5d2+ Kh2g3 Qd2c3+
Kg3g2 Qc3c4 Qe4b7+ Ke7d6) -7.73/10 5}

42. g4

{(g3g4) 9.60/15 6}

42. ... a6

{(a7a6 g4g5 Rd5xg5+ h4xg5 Qh6xg5+ Kg1h2 Qg5h5+ Kh2g3 Qh5g5+ Kg3h3 Qg5h5+
Kh3g2 Qh5d5 Nf8g6+ Ke7d7) -7.68/9 5}

43. g5

{(g4g5) 10.22/14 6}

43. ... Rxg5+

{(Rd5xg5+ h4xg5 Qh6xg5+ Kg1h2 Qg5h6+ Kh2g3 Qh6g5+ Kg3h3 Qg5h5+ Qe4h4+
Qh5xh4+ Kh3xh4 a6a5 Rf1f2 Ke7d6 Rf2f7 e6e5 d4xe5+ Kd6xe5 Rf7xg7) -8.26/11 5
}

44. hxg5

{(h4g5h6g5 g1h2 g5h5 ) 10.37/15 5}

44. ... Qxg5+

{(Qh6xg5+ Kg1h2 Qg5h5+ Kh2g3 Qh5g5+ Qe4g4 Qg5xg4+ Kg3xg4 Ke7d6 Rf1e1 Kd6d5
Nf8xe6 g7g6 Re1e5+ Kd5d6) -8.49/11 5}

45. Kh2

{(g1h2) 10.69/14 5}

45. ... Qh5+

{(Qg5h5+ Kh2g3 Qh5g5+ Qe4g4 Qg5xg4+ Kg3xg4 a6a5 Kg4g5 Ke7d6 Rf1e1 Kd6d5
Nf8xe6 Kd5c4 Re1b1) -8.66/11 5}

46. Kg3

{(h2g3h5g5 e4g4 g5g4 g3g4 ) 10.75/15 5}

46. ... Qg5+

{(Qh5g5+ Kg3h3 Qg5h5+ Qe4h4+ Qh5xh4+ Kh3xh4 Ke7d6 Rf1f7 Kd6d5 Rf7d7+ Kd5c4
Nf8xe6 Kc4b4 d4d5) -8.64/10 5}

47. Qg4

{(e4g4) 10.57/16 6}

47. ... Qxg4+

{(Qg5xg4+ Kg3xg4 a6a5 Kg4g5 Ke7d6 Rf1f7 Kd6d5 Rf7d7+ Kd5c4 Nf8xe6 Kc4b4
Rd7b7+ Kb4c4 Rb7xg7) -9.38/12 18}

48. Kxg4

{(g3g4) 11.42/18 5}

48. ... a5

{(a6a5 Kg4g5 Ke7d6 Rf1e1 Kd6e7 Nf8xe6 Ke7d7 Ne6xg7 Kd7c7 Ng7f5 Kc7c6 Kg5f4)
-9.64/11 5}

49. Ng6+

{(f8g6) 12.34/18 5}

49. ... Kd6

{(Ke7d6 Rf1e1 Kd6d5 Re1e5+ Kd5d6 Re5xa5 Kd6c6 Ra5c5+ Kc6b6 Kg4f4 Kb6b7
Kf4e5 Kb7b6 Ke5xe6) -9.77/11 5}

50. Rf7

{(f1f7d6d5 f7g7 d5d4 g7a7 d4c3 a7a5 c3b3 a5a7 b3a3 a4a5 a3b4 g6e5 b4a4 g4g5
) 12.94/18 5}

50. ... Kd5

{(Kd6d5 Rf7d7+ Kd5c6 Ng6e5+ Kc6b6 Rd7xg7 Kb6a6 Ne5c4 e6e5 Rg7g6+ Ka6a7
d4xe5 Ka7b7 Nc4xa5+ Kb7c7) -10.57/10 5}

51. Rxg7

{(f7g7d5d4 g7a7 d4c3 a7a5 ) 12.96/18 5}

51. ... Kxd4

{Black resigns}


1-0

[BubbaTough]

I have set up a match of LearningLemming0.19 vs. RomiChessNG8 of 14 min games with 1 sec increments with book/learning on for some obscene number of games. It looks like it will be a good match (starting score: 5.5 LL, 4.5 Romi). I will post periodic updates. If you want any artifacts from the match (such as learn file or games) let me know and I will email you them. I am not sure how many games it generally takes for Romi learning to start mastering most opponents...if someone has a rule of thumb let me know and I can use it to help determine when to start posting results.

-Sam

[Tony Thomas]

I have set up a match of LearningLemming0.19 vs. RomiChessNG8 of 14 min games with 1 sec increments with book/learning on for some obscene number of games. It looks like it will be a good match (starting score: 5.5 LL, 4.5 Romi). I will post periodic updates. If you want any artifacts from the match (such as learn file or games) let me know and I will email you them. I am not sure how many games it generally takes for Romi learning to start mastering most opponents...if someone has a rule of thumb let me know and I can use it to help determine when to start posting results.

-Sam

Around 50 games, Romi should start to even out with an opponent about 100 points stronger.

[BubbaTough]

So far I think Romi has played Grunfeld in all games as black, and Ruy Lopez in all games as black. If it keeps losing and LearningLemming does not change things up, will it eventually switch openings, or always go to the same starting position and just keep trying to improve its play?

-Sam

[Michael Sherwin]

So far I think Romi has played Grunfeld in all games as black, and Ruy Lopez in all games as black. If it keeps losing and LearningLemming does not change things up, will it eventually switch openings, or always go to the same starting position and just keep trying to improve its play?

-Sam

If you are using the DWD opening book, it is a type two opening book and will always be used. Romi will continue to try to improve her play from the point that she leaves the book.

To have Romi eventually switch its opening moves, no book should be used or a type one book should be used. just rename the DWD learn.dat file to DWDlearn.dat and Romi will create a new learn.dat file the next time she starts. Then if you have a good pgn collection you can just type merge filename.pgn. The pgn must be in the same directory.

If you have a collection of LL games you can merge them and let Romi start with LL experiance. That might be interesting.

Always exit Romi with 'quit' so the files are flushed and closed.

[BubbaTough]

I am using Arena, not sure if it is quitting correctly or not. I am using default settings, which I believe is book on learning on (since it does seem to be using a book and learning). Arema is set to restart each program once every 10 games, I assume this does not cause problems. If you want me to change this in anyway just send me instructions and I will happily obey.

time control: 14 min 1 sec interval. Cache size 32, hardware core 2 duo 2.66

Results so far:

LearningLemming 38.5
RomiChessNG8 24.5

distribution from LearningLemming perspective:
==1010=1==1011111=100=101=101=10111=001110=01011=1=01=1=1=1=010

I have preserved all the games from this match so far if you want to look at them. Besides this I don't hae a good collection of LearningLemming games to use. General conclusions from having looked through a number of games:

Grunfeld play:
These games have been wide open, with LearningLemming generally coming out of the opening with good attacking chances. LearningLemming has been trying out a lot of different things, and the variation has prevented Romichess from being able to develop a cohesive set of moves to consolidate its position (which seems to take a number of games to happen).

Ruy Lopez play:
Both programs have settled on a very deep line of play that locks up the queenside pawns, and gives white (Romi) a space advantage. For a big chunk of the first games Romi played this position pretty planlessly. After a while though, it seemed to figure out it needs to bring pieces over to the king side and try to develop something, and it is doing much better.

General:
Overall, my impression that Romi learning is more effective is closed manevuering is being reinforced, at least when both sides are learning. I will keep running things for a while, and send another update. If you want me to try a different setting / version / learn file let me know.

-Sam