Beowulf curve calibration results for depth = 5 plies

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 07 June 2004 19:36:04:

The high value for the queen and the bishop are surprising to me. I thought that the bishop value would decrease, since it did decrease with earlier plies.
use_delta
(x=0.000000, y=5821.000000) t=3180.000000,
(x=1.000000, y=5818.000000) t=3261.000000, 0.000000 is better for use_delta at depth 5
use_eval_sc
(x=0.000000, y=5821.000000) t=3387.000000,
(x=1.000000, y=5821.000000) t=2882.000000, 1.000000 is better for use_eval_sc at depth 5
use_hash
(x=0.000000, y=5816.000000) t=5875.000000,
(x=1.000000, y=5821.000000) t=2810.000000, 1.000000 is better for use_hash at depth 5
use_history
(x=0.000000, y=5807.000000) t=3267.000000,
(x=1.000000, y=5821.000000) t=2834.000000, 1.000000 is better for use_history at depth 5
use_iid
(x=0.000000, y=5820.000000) t=2789.000000,
(x=1.000000, y=5821.000000) t=2853.000000,
use_killers
(x=0.000000, y=5827.000000) t=2859.000000,
(x=1.000000, y=5821.000000) t=2846.000000,
use_null
(x=0.000000, y=5891.000000) t=4140.000000,
(x=1.000000, y=5821.000000) t=2860.000000,
use_razoring
(x=0.000000, y=5821.000000) t=2859.000000,
(x=1.000000, y=5822.000000) t=2858.000000, 1.000000 is better for use_razoring at depth 5
use_see
(x=0.000000, y=5807.000000) t=4205.000000,
(x=1.000000, y=5822.000000) t=2858.000000, 1.000000 is better for use_see at depth 5
use_verification
(x=0.000000, y=5821.000000) t=2858.000000,
(x=1.000000, y=5822.000000) t=2858.000000, 1.000000 is better for use_verification at depth 5
use_window
(x=0.000000, y=5811.000000) t=2910.000000,
(x=1.000000, y=5822.000000) t=2857.000000, 1.000000 is better for use_window at depth 5
bishop_score=354 at 5; stddev=11.806867 : -0.0649647*x^2 + 45.9411*x + -2296.14
(x=291.000000, y=5569.000000), t=2809.000000
(x=307.000000, y=5685.000000), t=2837.000000
(x=323.000000, y=5775.000000), t=2846.000000
(x=339.000000, y=5801.000000), t=2860.000000
(x=355.000000, y=5836.000000), t=2858.000000
(x=371.000000, y=5794.000000), t=2855.000000
(x=387.000000, y=5759.000000), t=2858.000000
(xmax=355.000000, ymax=5836.000000), xmax seen verses curve xmax [355 354]
knight_score=319 at 5; stddev=13.190923 : -0.0730562*x^2 + 46.643*x + -1620.68
(x=286.000000, y=5750.000000), t=2859.000000
(x=302.000000, y=5795.000000), t=2856.000000
(x=318.000000, y=5824.000000), t=2861.000000
(x=334.000000, y=5795.000000), t=2858.000000
(x=350.000000, y=5773.000000), t=2847.000000
(x=366.000000, y=5667.000000), t=2830.000000
(x=382.000000, y=5530.000000), t=2810.000000
(xmax=318.000000, ymax=5824.000000), xmax seen verses curve xmax [318 319]
queen_score=1100 at 5; stddev=4.098942 : -0.000190476*x^2 + 0.458095*x + 5555
(x=850.000000, y=5804.000000), t=2867.000000
(x=900.000000, y=5819.000000), t=2862.000000
(x=950.000000, y=5816.000000), t=2859.000000
(x=1000.000000, y=5821.000000), t=2859.000000
(x=1050.000000, y=5826.000000), t=2857.000000
(x=1100.000000, y=5829.000000), t=2856.000000
(x=1150.000000, y=5830.000000), t=2856.000000
(xmax=1150.000000, ymax=5830.000000), xmax seen verses curve xmax [1150 1100]
TIME queen_score=1100 at 5; stddev=1.419368 : 0.000147619*x^2 + -0.32881*x + 3039.14
TIME (xmin=1100.000000, tmin=2856.000000), TIME xmin seen verses curve xmin [1100 1100]
rook_score=492 at 5; stddev=7.479562 : -0.00404924*x^2 + 3.98785*x + 4847.15
(x=471.000000, y=5826.000000), t=2861.000000
(x=492.000000, y=5826.000000), t=2860.000000
(x=513.000000, y=5836.000000), t=2858.000000
(x=534.000000, y=5820.000000), t=2855.000000
(x=555.000000, y=5814.000000), t=2855.000000
(x=576.000000, y=5792.000000), t=2851.000000
(x=597.000000, y=5790.000000), t=2851.000000
(xmax=513.000000, ymax=5836.000000), xmax seen verses curve xmax [513 492]
TIME rook_score=580 at 5; stddev=1.494030 : 2.69949e-005*x^2 + -0.115565*x + 2909.82
TIME (xmin=576.000000, tmin=2851.000000), TIME xmin seen verses curve xmin [576 580]
avoidance_depth_razor=13 at 5; stddev=2.279797 : -0.30303*x^2 + 7.6*x + 5746.31
(x=7.000000, y=5785.000000), t=2601.000000
(x=8.000000, y=5785.000000), t=2600.000000
(x=9.000000, y=5793.000000), t=2851.000000
(x=10.000000, y=5793.000000), t=2852.000000
(x=11.000000, y=5793.000000), t=2850.000000
(x=12.000000, y=5793.000000), t=2851.000000
(x=13.000000, y=5793.000000), t=2853.000000
(x=14.000000, y=5793.000000), t=2857.000000
(x=15.000000, y=5793.000000), t=2861.000000
(xmax=9.000000, ymax=5793.000000), xmax seen verses curve xmax [9 13]

Bad data this pass -- no fit. (avoid_null_mat):
(x=3.000000, y=5793.000000), t=2855.000000
(x=6.000000, y=5793.000000), t=2854.000000
(x=9.000000, y=5793.000000), t=2853.000000
(x=12.000000, y=5793.000000), t=2854.000000
(x=15.000000, y=5793.000000), t=2853.000000
(x=18.000000, y=5793.000000), t=2852.000000
(x=21.000000, y=5793.000000), t=2855.000000 xmax seen verses curve xmax [3 0]
avoid_null_mat=21 at 5; {badfit}
TIME avoid_null_mat=13 at 5; stddev=1.475900 : 0.021164*x^2 + -0.555556*x + 2856.57
TIME (xmax=3.000000, ymax=5793.000000), TIME xmax seen verses curve xmax [3 13]

Bad data this pass -- no fit. (avoid_null_pieces):
(x=1.000000, y=5793.000000), t=2848.000000
(x=2.000000, y=5793.000000), t=2847.000000
(x=3.000000, y=5793.000000), t=2848.000000
(x=4.000000, y=5793.000000), t=2848.000000
(x=5.000000, y=5793.000000), t=2846.000000
(x=6.000000, y=5793.000000), t=2850.000000
(x=7.000000, y=5792.000000), t=2847.000000
(x=8.000000, y=5793.000000), t=2849.000000
(x=9.000000, y=5793.000000), t=2849.000000
(x=10.000000, y=5793.000000), t=2852.000000
(x=11.000000, y=5795.000000), t=2859.000000 xmax seen verses curve xmax [11 5]
avoid_null_pieces=11 at 5; {badfit}
TIME avoid_null_pieces=4 at 5; stddev=2.432244 : 0.213287*x^2 + -1.82308*x + 2850.49
TIME (xmax=11.000000, ymax=5795.000000), TIME xmax seen verses curve xmax [11 4]

Bad data this pass -- no fit. (avoid_raz_num):
(x=3.000000, y=5793.000000), t=2847.000000
(x=4.000000, y=5793.000000), t=2847.000000
(x=5.000000, y=5793.000000), t=2847.000000
(x=6.000000, y=5793.000000), t=2849.000000
(x=7.000000, y=5794.000000), t=2850.000000
(x=8.000000, y=5794.000000), t=2852.000000
(x=9.000000, y=5794.000000), t=2853.000000 xmax seen verses curve xmax [7 1]
avoid_raz_num=9 at 5; {badfit}
TIME avoid_raz_num=2 at 5; stddev=1.023268 : 0.154762*x^2 + -0.75*x + 2847.6
TIME (xmax=7.000000, ymax=5794.000000), TIME xmax seen verses curve xmax [7 2]



my ftp site {remove http:// unless you like error messages}

[Stefan Knappe]

Geschrieben von:/Posted by: Stefan Knappe at 08. June 2004 09:24:
Als Antwort auf:/In reply to: Beowulf curve calibration results for depth = 5 plies geschrieben von:/posted by: Dann Corbit at 07 June 2004 19:36:04:

Hi Dann,
how much time did you need for a calibration on depth 5? I have read, you used more than 1000 positions to fit the parameters?!
I would expect, there are endless combinations of optimized parameters, which all produce the same results - am I right? Probably you can compensate a decrease of the value of the queen due to an increase of the mobility?! Do you think, this compensation is possible? How do you prevent strange combinations?
Best regards,
Stefan

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 09 June 2004 02:38:21:
Als Antwort auf:/In reply to: Re: Beowulf curve calibration results for depth = 5 plies geschrieben von:/posted by: Stefan Knappe at 08. June 2004 09:24:

Hi Dann,
how much time did you need for a calibration on depth 5? I have read, you used more than 1000 positions to fit the parameters?!
I would expect, there are endless combinations of optimized parameters, which all produce the same results - am I right?
Probably you can compensate a decrease of the value of the queen due to an increase of the mobility?! Do you think, this compensation is possible?
How do you prevent strange combinations?

There are exactly 11,748 distinct positions.
Each position will be solved with many different values for the parameters. For a few paramters that are boolean, only "on" and "off" are tried.
For most paramters, there are at least 6 different values attempted for the whole set.
There are 118 paramters. If all were two valued, then there would be 2^118 distinct possibilities = 3.323e+35
Of course, there are many, many times more than that.
Anything is possible
I don't I am hoping for an effect similar to simulated annealing. After iteration to a solution set, I run it again and again until it settles down to some happy medium. Then I increase the depth and do it again.
You can see for yourself how long a run takes, exactly.
The t= column is the time in seconds.
For instance,
bishop_score=354 at 5; stddev=11.806867 : -0.0649647*x^2 + 45.9411*x + -2296.14
(x=291.000000, y=5569.000000), t=2809.000000
(x=307.000000, y=5685.000000), t=2837.000000
(x=323.000000, y=5775.000000), t=2846.000000
(x=339.000000, y=5801.000000), t=2860.000000
(x=355.000000, y=5836.000000), t=2858.000000
(x=371.000000, y=5794.000000), t=2855.000000
(x=387.000000, y=5759.000000), t=2858.000000
(xmax=355.000000, ymax=5836.000000), xmax seen verses curve xmax [355 354]
When a bishop was considered to be worth 291 centipawns 5,569 out of 11,748 positions were solved. The time for that run was 2809 seconds = 41.82 minutes.
Total time to resolve the bishop score curve was 2809+2837+2846+2860+2858+2855+2858) = 19923 seconds = 332.05 minutes = 5.5 hours
To solve at lower depths is exponentially faster. For some of the important factors, I might try 6 plies. 7 plies is right out.



my ftp site {remove http:// unless you like error messages}

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 09 June 2004 02:54:53:
Als Antwort auf:/In reply to: Re: Beowulf curve calibration results for depth = 5 plies geschrieben von:/posted by: Dann Corbit at 09 June 2004 02:38:21:

use_delta
(x=0.000000, y=5821.000000) t=3180.000000,
(x=1.000000, y=5818.000000) t=3261.000000, 0.000000 is better for use_delta at depth 5
use_eval_sc
(x=0.000000, y=5821.000000) t=3387.000000,
(x=1.000000, y=5821.000000) t=2882.000000, 1.000000 is better for use_eval_sc at depth 5
use_hash
(x=0.000000, y=5816.000000) t=5875.000000,
(x=1.000000, y=5821.000000) t=2810.000000, 1.000000 is better for use_hash at depth 5
use_history
(x=0.000000, y=5807.000000) t=3267.000000,
(x=1.000000, y=5821.000000) t=2834.000000, 1.000000 is better for use_history at depth 5
use_iid
(x=0.000000, y=5820.000000) t=2789.000000,
(x=1.000000, y=5821.000000) t=2853.000000,
use_killers
(x=0.000000, y=5827.000000) t=2859.000000,
(x=1.000000, y=5821.000000) t=2846.000000,
use_null
(x=0.000000, y=5891.000000) t=4140.000000,
(x=1.000000, y=5821.000000) t=2860.000000,
use_razoring
(x=0.000000, y=5821.000000) t=2859.000000,
(x=1.000000, y=5822.000000) t=2858.000000, 1.000000 is better for use_razoring at depth 5
use_see
(x=0.000000, y=5807.000000) t=4205.000000,
(x=1.000000, y=5822.000000) t=2858.000000, 1.000000 is better for use_see at depth 5
use_verification
(x=0.000000, y=5821.000000) t=2858.000000,
(x=1.000000, y=5822.000000) t=2858.000000, 1.000000 is better for use_verification at depth 5
use_window
(x=0.000000, y=5811.000000) t=2910.000000,
(x=1.000000, y=5822.000000) t=2857.000000, 1.000000 is better for use_window at depth 5
bishop_score=354 at 5; stddev=11.806867 : -0.0649647*x^2 + 45.9411*x + -2296.14
(x=291.000000, y=5569.000000), t=2809.000000
(x=307.000000, y=5685.000000), t=2837.000000
(x=323.000000, y=5775.000000), t=2846.000000
(x=339.000000, y=5801.000000), t=2860.000000
(x=355.000000, y=5836.000000), t=2858.000000
(x=371.000000, y=5794.000000), t=2855.000000
(x=387.000000, y=5759.000000), t=2858.000000
(xmax=355.000000, ymax=5836.000000), xmax seen verses curve xmax [355 354]
knight_score=319 at 5; stddev=13.190923 : -0.0730562*x^2 + 46.643*x + -1620.68
(x=286.000000, y=5750.000000), t=2859.000000
(x=302.000000, y=5795.000000), t=2856.000000
(x=318.000000, y=5824.000000), t=2861.000000
(x=334.000000, y=5795.000000), t=2858.000000
(x=350.000000, y=5773.000000), t=2847.000000
(x=366.000000, y=5667.000000), t=2830.000000
(x=382.000000, y=5530.000000), t=2810.000000
(xmax=318.000000, ymax=5824.000000), xmax seen verses curve xmax [318 319]
queen_score=1100 at 5; stddev=4.098942 : -0.000190476*x^2 + 0.458095*x + 5555
(x=850.000000, y=5804.000000), t=2867.000000
(x=900.000000, y=5819.000000), t=2862.000000
(x=950.000000, y=5816.000000), t=2859.000000
(x=1000.000000, y=5821.000000), t=2859.000000
(x=1050.000000, y=5826.000000), t=2857.000000
(x=1100.000000, y=5829.000000), t=2856.000000
(x=1150.000000, y=5830.000000), t=2856.000000
(xmax=1150.000000, ymax=5830.000000), xmax seen verses curve xmax [1150 1100]
TIME queen_score=1100 at 5; stddev=1.419368 : 0.000147619*x^2 + -0.32881*x + 3039.14
TIME (xmin=1100.000000, tmin=2856.000000), TIME xmin seen verses curve xmin [1100 1100]
rook_score=492 at 5; stddev=7.479562 : -0.00404924*x^2 + 3.98785*x + 4847.15
(x=471.000000, y=5826.000000), t=2861.000000
(x=492.000000, y=5826.000000), t=2860.000000
(x=513.000000, y=5836.000000), t=2858.000000
(x=534.000000, y=5820.000000), t=2855.000000
(x=555.000000, y=5814.000000), t=2855.000000
(x=576.000000, y=5792.000000), t=2851.000000
(x=597.000000, y=5790.000000), t=2851.000000
(xmax=513.000000, ymax=5836.000000), xmax seen verses curve xmax [513 492]
TIME rook_score=580 at 5; stddev=1.494030 : 2.69949e-005*x^2 + -0.115565*x + 2909.82
TIME (xmin=576.000000, tmin=2851.000000), TIME xmin seen verses curve xmin [576 580]
avoidance_depth_razor=13 at 5; stddev=2.279797 : -0.30303*x^2 + 7.6*x + 5746.31
(x=7.000000, y=5785.000000), t=2601.000000
(x=8.000000, y=5785.000000), t=2600.000000
(x=9.000000, y=5793.000000), t=2851.000000
(x=10.000000, y=5793.000000), t=2852.000000
(x=11.000000, y=5793.000000), t=2850.000000
(x=12.000000, y=5793.000000), t=2851.000000
(x=13.000000, y=5793.000000), t=2853.000000
(x=14.000000, y=5793.000000), t=2857.000000
(x=15.000000, y=5793.000000), t=2861.000000
(xmax=9.000000, ymax=5793.000000), xmax seen verses curve xmax [9 13]

Bad data this pass -- no fit. (avoid_null_mat):
(x=3.000000, y=5793.000000), t=2855.000000
(x=6.000000, y=5793.000000), t=2854.000000
(x=9.000000, y=5793.000000), t=2853.000000
(x=12.000000, y=5793.000000), t=2854.000000
(x=15.000000, y=5793.000000), t=2853.000000
(x=18.000000, y=5793.000000), t=2852.000000
(x=21.000000, y=5793.000000), t=2855.000000 xmax seen verses curve xmax [3 0]
avoid_null_mat=21 at 5; {badfit}
TIME avoid_null_mat=13 at 5; stddev=1.475900 : 0.021164*x^2 + -0.555556*x + 2856.57
TIME (xmax=3.000000, ymax=5793.000000), TIME xmax seen verses curve xmax [3 13]

Bad data this pass -- no fit. (avoid_null_pieces):
(x=1.000000, y=5793.000000), t=2848.000000
(x=2.000000, y=5793.000000), t=2847.000000
(x=3.000000, y=5793.000000), t=2848.000000
(x=4.000000, y=5793.000000), t=2848.000000
(x=5.000000, y=5793.000000), t=2846.000000
(x=6.000000, y=5793.000000), t=2850.000000
(x=7.000000, y=5792.000000), t=2847.000000
(x=8.000000, y=5793.000000), t=2849.000000
(x=9.000000, y=5793.000000), t=2849.000000
(x=10.000000, y=5793.000000), t=2852.000000
(x=11.000000, y=5795.000000), t=2859.000000 xmax seen verses curve xmax [11 5]
avoid_null_pieces=11 at 5; {badfit}
TIME avoid_null_pieces=4 at 5; stddev=2.432244 : 0.213287*x^2 + -1.82308*x + 2850.49
TIME (xmax=11.000000, ymax=5795.000000), TIME xmax seen verses curve xmax [11 4]

Bad data this pass -- no fit. (avoid_raz_num):
(x=3.000000, y=5793.000000), t=2847.000000
(x=4.000000, y=5793.000000), t=2847.000000
(x=5.000000, y=5793.000000), t=2847.000000
(x=6.000000, y=5793.000000), t=2849.000000
(x=7.000000, y=5794.000000), t=2850.000000
(x=8.000000, y=5794.000000), t=2852.000000
(x=9.000000, y=5794.000000), t=2853.000000 xmax seen verses curve xmax [7 1]
avoid_raz_num=9 at 5; {badfit}
TIME avoid_raz_num=2 at 5; stddev=1.023268 : 0.154762*x^2 + -0.75*x + 2847.6
TIME (xmax=7.000000, ymax=5794.000000), TIME xmax seen verses curve xmax [7 2]

Bad data this pass -- no fit. (back_rank_unsafe):
(x=3.000000, y=5792.000000), t=2847.000000
(x=4.000000, y=5792.000000), t=2847.000000
(x=5.000000, y=5792.000000), t=2906.000000
(x=6.000000, y=5792.000000), t=2856.000000
(x=7.000000, y=5792.000000), t=2859.000000
(x=8.000000, y=5792.000000), t=3032.000000
(x=9.000000, y=5792.000000), t=2996.000000
(x=10.000000, y=5792.000000), t=2910.000000 xmax seen verses curve xmax [3 0]
back_rank_unsafe=10 at 5; {badfit}

Bad data this pass -- no fit. (backward_pawn_1):
(x=3.000000, y=5792.000000), t=3106.000000
(x=4.000000, y=5792.000000), t=3011.000000
(x=5.000000, y=5792.000000), t=2936.000000
(x=6.000000, y=5792.000000), t=2934.000000
(x=7.000000, y=5792.000000), t=2958.000000
(x=8.000000, y=5792.000000), t=2939.000000
(x=9.000000, y=5792.000000), t=2936.000000
(x=10.000000, y=5792.000000), t=2938.000000
(x=11.000000, y=5792.000000), t=2936.000000
(x=12.000000, y=5792.000000), t=2940.000000 xmax seen verses curve xmax [3 0]
backward_pawn_1=12 at 5; {badfit}
TIME backward_pawn_1=9 at 5; stddev=29.805511 : 4.15909*x^2 + -74.6409*x + 3254.95
TIME (xmax=3.000000, ymax=5792.000000), TIME xmax seen verses curve xmax [3 9]

Bad data this pass -- no fit. (backward_pawn_2):
(x=7.000000, y=5792.000000), t=2938.000000
(x=10.000000, y=5792.000000), t=2938.000000
(x=13.000000, y=5792.000000), t=2937.000000
(x=16.000000, y=5792.000000), t=2938.000000
(x=19.000000, y=5792.000000), t=2938.000000
(x=22.000000, y=5792.000000), t=2937.000000
(x=25.000000, y=5792.000000), t=2938.000000
(x=28.000000, y=5792.000000), t=2940.000000 xmax seen verses curve xmax [7 0]
backward_pawn_2=28 at 5; {badfit}
TIME backward_pawn_2=15 at 5; stddev=1.200340 : 0.0132275*x^2 + -0.407407*x + 2940.45
TIME (xmax=7.000000, ymax=5792.000000), TIME xmax seen verses curve xmax [7 15]

Bad data this pass -- no fit. (backward_pawn_3):
(x=15.000000, y=5792.000000), t=2938.000000
(x=21.000000, y=5792.000000), t=2938.000000
(x=27.000000, y=5792.000000), t=2940.000000
(x=33.000000, y=5792.000000), t=2940.000000
(x=39.000000, y=5792.000000), t=2939.000000
(x=45.000000, y=5792.000000), t=3025.000000
(x=51.000000, y=5792.000000), t=3117.000000 xmax seen verses curve xmax [15 0]
backward_pawn_3=51 at 5; {badfit}
TIME backward_pawn_3=26 at 5; stddev=22.560523 : 0.290344*x^2 + -14.9365*x + 3111.62
TIME (xmax=15.000000, ymax=5792.000000), TIME xmax seen verses curve xmax [15 26]



my ftp site {remove http:// unless you like error messages}