Queen 3.08

[Leen Ammeraal]

Queen 3.08 is now available from my website
http://home.planet.nl/~ammeraal/.
After playing hundred of games, this version seems to be slightly
stronger than Queen 3.07.
Leen Ammeraal

[Norm Pollock]

Queen 3.08 is now available from my website
http://home.planet.nl/~ammeraal/.
After playing hundred of games, this version seems to be slightly
stronger than Queen 3.07.
Leen Ammeraal

Thanks Leen. Queen has a terrific gui which makes it a very practical and small package (3M w/o tbs) that I can take with me on flash memory. It is great for human v computer chess games and for analysis of human-human games but give it a minute at least for each analysis. It's not Fritz, but it is much smaller and much easier to use.

I have one suggestion. I think Queen would play a whole lot better against other engines if it would use flexible time management.

[Leen Ammeraal]

...
I have one suggestion. I think Queen would play a whole lot better against other engines if it would use flexible time management.

Thank you Norm. I would very much like to do this, but
could you possibly give me a hint? I suppose that an engine should
use more time in difficult situations than in easy ones, but when
should a position be considered 'difficult'?
Leen

[Norm Pollock]

...
I have one suggestion. I think Queen would play a whole lot better against other engines if it would use flexible time management.

Thank you Norm. I would very much like to do this, but
could you possibly give me a hint? I suppose that an engine should
use more time in difficult situations than in easy ones, but when
should a position be considered 'difficult'?
Leen

Based on my observations, I believe that many engines use more time when they see that the move they are about to make has a much lower score than their previous move. You also want to use less time when there is only one move to get out of check, or for any other "no-brainer".

[Leen Ammeraal]

Thanks Norm. I had already seen that Queen does not immediately move
when only one move is possible, but I still had not done anything about it.
This other suggestion of yours, taking more time if the score is substantially
dropping, is also very clear to me, so it is time to implement 'flexible time
management' in the next version, due probably in the beginning of 2006.
Leen

[mike schoonover]

good work.

test games,win xp,96 mb hash,3/4 egtb,2.53 ghz cel,533 mhz fsb.

[Event "Computer chess game"]
[Site "SKOONY2"]
[Date "2005.11.29"]
[Round "-"]
[White "queen"]
[Black "SlowChess Blitz WV"]
[Result "1/2-1/2"]
[TimeControl "300"]

1. e4 c6 2. d4 d5 3. Nd2 dxe4 4. Nxe4 Bf5 5. Ng3 Bg6 6. h4 h6 7. Nf3 Nd7 8.
h5 Bh7 9. Bd3 Bxd3 10. Qxd3 Qc7 11. Bd2 e6 12. O-O-O Ngf6 13. Ne4 O-O-O 14.
Rh4 Nxe4 15. Qxe4 Be7 16. Rh3 Nf6 17. Qe2 Kb8 18. Ne5 Rxd4 19. Nxf7 Rf8 20.
Qxe6 Re4 21. Qb3 Bc5 22. Nxh6 gxh6 23. Bxh6 Rd8 24. Qg3 Rxd1+ 25. Kxd1 Bd6
26. Qd3 Re5 27. c4 Qe7 28. Be3 Nxh5 29. g4 Nf6 30. Rh8+ Kc7 31. Bxa7 Re6
32. c5 Re1+ 33. Kd2 Bf4+ 34. Kc3 Be5+ 35. Kb3 Ne8 36. Bb6+ Kb8 37. Rf8 Bc7
38. Bxc7+ Kxc7 39. Qf5 Qe2 40. Rf7+ Kb8 41. Qf4+ Ka7 42. Qa4+ Kb8 43. Qf4+
Kc8 44. Qf5+ Kb8 45. Qf4+
{Draw by repetition} 1/2-1/2

[Event "Computer chess game"]
[Site "SKOONY2"]
[Date "2005.11.29"]
[Round "-"]
[White "SlowChess Blitz WV"]
[Black "queen"]
[Result "1-0"]
[TimeControl "300"]

1. Nf3 d5 2. e3 Nf6 3. d4 c6 4. c4 e6 5. Nc3 Nbd7 6. Be2 dxc4 7. Bxc4 b5 8.
Be2 Bb7 9. O-O a6 10. b3 c5 11. Bb2 Bd6 12. dxc5 Bxc5 13. Qc2 Qc7 14. Rac1
O-O 15. Nxb5 axb5 16. Bxf6 gxf6 17. b4 Bd5 18. Bxb5 Rxa2 19. Qb1 Qb6 20.
bxc5 Nxc5 21. Qb4 Ne4 22. Nd4 Rb8 23. h4 Kh8 24. Qe7 Ra7 25. Bd7 e5 26. Rb1
Qxb1 27. Rxb1 Rxb1+ 28. Kh2 Kg7 29. Nf5+ Kg6 30. f3 Rb8 31. fxe4 Be6 32.
Qd6 Rd8 33. Qd1 h5 34. Qf3 Bxf5 35. Bxf5+ Kh6 36. Bg4 Kg7 37. Bxh5 Ra2 38.
Qg3+ Kf8 39. Bg4 Ke7 40. h5 Ra4 41. Bf5 Rh8 42. Qh3 Rh6 43. g4 Ra2+ 44. Kg3
Kf8 45. Qh1 Kg7 46. Qd1 Rh8 47. Qd7 Ra3 48. Be6 Rf8 49. Qe7 Rxe3+ 50. Kh4
Kg8 51. h6 Rf3 52. g5 Rf4+ 53. Kh5 Rh4+ 54. Kxh4 Kh7 55. Bf5+ Kg8 56. h7+
Kg7 57. Qxf6#
{White Mates} 1-0

regards
mike

[Thomas Mayer]

Hi Leen,

to give you just a little hint how it could work I give you an overview how Quark is doing it:

First I remember the score of the last move. If it's the initial position or a book move I remember 0.00.

Now my time routine gives back 2 times: First the approx time Quark should use for the move and the max time Quark should use. (Just to handle fail low issues when short before the cut or very low in time)

So what's happening now in the search ? Well, in the tree it is the usual: timeused>timeavail stop search. But I change timeavail at the root search from time to time:

a) when I search the first move of an iteration (the best move so far) I double timeavail. (With a max as described above) This is just to make sure that the move will be fully searched. (But sometimes this fails anyway, you know that the search can explode from iteration to iteration - some do try here anyway to complete the search)

b) when the first move is a fail low (and here I mean a fail low of -0.30 under the expected score from previous move) I research with timeavail=4*approxtime. And I let the timeavail that high as long no other move get's back in the expected window. So I keep timeavail=4*approxtime also for the other moves in the root search

c) when the first move fails very high (well, the margin is matter of taste) and I have used more then 1/2 approxtime I stop search.

d) when the first move looks okay and is in the window of the expected score I set timeavail=approxtime for the rest of the moves

e) when I find a new candidate for being bestmove I double timeavail also. (timeavail=2*approxtime). Exception: fail-low situation, then timeavail stays at 4*approxtime

f) when in fail low situation a move is back in the expected window I set timeavail back to approxtime

g) when after an iteration more then 1/2 timeavail is used I stop the search, it's unlikely that the next iteration will be completed.

That's more or less what Quark is doing. It shouldn't be to complicate to implement that and I think it would help Queen also to defend in dangerous situations. Also you can save some time with rule g... Of course all the values are matter of taste, I use here just what my stomach says, I have no idea wether other values would do a better job.

Ah, about the easy move. Well, I have there just a rule for single move possible: timeavail=min(10seconds,approxtime/10)
This means it will not take more then 10 seconds or a tenth of the usual approxtime when just a single move is possible. I give it some seconds to find a good pondering move. This is still somehow buggy in Quark, but at least it should work that way...

I hope this helps a bit or give you an idea how to implement it yourself.

Greets, Thomas

[Leen Ammeraal]

Thank you very much Thomas. I have just printed out your reaction,
which I do only for very important material! So it looks like I have a lot
of homework to do. I will send you private e-mail if I want some
further help on this point.
Leen