Inquiry ( to Stockfish Team)

[neelbasant]

Hello All

There was patch called Intelligent contempt (IC) by jcalovski which passed both STC and LTC.
But I am still unable to find the binary in abrok.stockfish.( added to master)
I just want to know what happened to the patch and why the binary is not
there ?

Neel

[Eelco de Groot]

The name Intelligent Contempt is new to me, I must have missed that particular experiment, sorry. But I am fairly certain, this was never commited to the master. Are you sure it was jcalovski? His username is Mysseno, maybe you could find more there.

There was however once 'Dynamic Contempt' which did pass all STC and LTC tests. But ultimately this also was not committed, and if you are interested in the source of that, it can still be found.

A Fishcooking search turns up reverting dynamic contempt = regression (?)
The auhor of dynamic contempt was as far as I can see Leonid Pechenik, at least he authored the tests mentioned in the Fishcooking thread and his dynamic contempt patch can still be found. Many branches get deleted over time, so if you are interested better download or otherwise copy, clone etc.

[neelbasant]

May be this will help.

https://github.com/jcalovski/Stockfish/ ... b738116bff

Finished on 04/05/2016

Neel

[Eelco de Groot]

May be this will help.

https://github.com/jcalovski/Stockfish/ ... b738116bff

Finished on 04/05/2016

Neel

Hello Neel,

Yes, I saw the new try meanwhile, overlooked that but I don't see any evidence of the patch passing either STC or LTC. And the test done on that date is the only STC test I see with result

My ic diff

LLR: -1.47 (-2.94,2.94) [0.00,5.00]
Total: 26127 W: 4737 L: 4688 D: 16702

sprt @ 10+0.1 th 1 Use intelligent contempt: increase/decrease our contempt for every fail high/low then reset like delta window.

Can you find any evidence that it passed even at STC? We have had one case of a passed test that disappeared from the page with completed tests so we can't exclude the possibility it happened again.

(There are tests done by Mysseno against Stockfish DD but those are clearly different from the tests against Stockfish master. Contempt has to be tested againt weaker opponents. It is not likely to pass against Master, at least not theoretically)

Eelco

[Ajedrecista]

Hello:

My ic diff

LLR: -1.47 (-2.94,2.94) [0.00,5.00]
Total: 26127 W: 4737 L: 4688 D: 16702

sprt @ 10+0.1 th 1 Use intelligent contempt: increase/decrease our contempt for every fail high/low then reset like delta window.

I ran 100000 SPRT(0, 5) simulations (alpha = 5%, beta = 5%) with the starting point 4737 wins, 4688 loses and 16702 draws. I used the parameters taken from this sample of 26127 games: drawelo ~ 262.9849; Elo gain ~ 1.1019 Bayeselo. I know that the Elo gain has uncertainties (± some Bayeselo) but anyway, I took the central point. Then:

Code: Select all

Passes:  5108/100000 ~  5.11%.
Fails:  94892/100000 ~ 94.89%.

----------------------------------------------------------------------------------------

There are  83645 simulations with score > 50% that failed SPRT.
There are   1390 simulations with score = 50% that failed SPRT.

----------------------------------------------------------------------------------------

Shortest simulation:
FAIL after  26699 games (+ 4806 - 4818 = 17075).      {   572 games to go}.

Median simulations (there are three):
FAIL after  38744 games (+ 6968 - 6929 = 24847).      { 12617 games to go}.
FAIL after  38744 games (+ 7073 - 7033 = 24638).      { 12617 games to go}.
FAIL after  38744 games (+ 7049 - 7011 = 24684).      { 12617 games to go}.

Average number of games ~ 47579 games.                { 21452 games to go}.
With 95% confidence intervals ~ [47431, 47728] games. {From 21304 to 21601 games to go}.

Longest simulation:
FAIL after 317368 games (+57786 -56562 =203020).      {291241 games to go}.

----------------------------------------------------------------------------------------

Summary of the distribution of the length of simulations:

From   26000 to   26999 games:     41 simulations (  0.04 %); accumulated:   0.04 %.
From   27000 to   27999 games:   2062 simulations (  2.06 %); accumulated:   2.10 %.
From   28000 to   28999 games:   5028 simulations (  5.03 %); accumulated:   7.13 %.
From   29000 to   29999 games:   6027 simulations (  6.03 %); accumulated:  13.16 %.
From   30000 to   30999 games:   5993 simulations (  5.99 %); accumulated:  19.15 %.
From   31000 to   31999 games:   5532 simulations (  5.53 %); accumulated:  24.68 %.
From   32000 to   32999 games:   4985 simulations (  4.99 %); accumulated:  29.67 %.
From   33000 to   33999 games:   4399 simulations (  4.40 %); accumulated:  34.07 %.
From   34000 to   34999 games:   3990 simulations (  3.99 %); accumulated:  38.06 %.
From   35000 to   35999 games:   3590 simulations (  3.59 %); accumulated:  41.65 %.
From   36000 to   36999 games:   3331 simulations (  3.33 %); accumulated:  44.98 %.
From   37000 to   37999 games:   2906 simulations (  2.91 %); accumulated:  47.88 %.
From   38000 to   38999 games:   2777 simulations (  2.78 %); accumulated:  50.66 %.
From   39000 to   39999 games:   2491 simulations (  2.49 %); accumulated:  53.15 %.
From   40000 to   40999 games:   2396 simulations (  2.40 %); accumulated:  55.55 %.
From   41000 to   41999 games:   2085 simulations (  2.09 %); accumulated:  57.63 %.
From   42000 to   42999 games:   2024 simulations (  2.02 %); accumulated:  59.66 %.
From   43000 to   43999 games:   1933 simulations (  1.93 %); accumulated:  61.59 %.
From   44000 to   44999 games:   1739 simulations (  1.74 %); accumulated:  63.33 %.
From   45000 to   45999 games:   1657 simulations (  1.66 %); accumulated:  64.99 %.
From   46000 to   46999 games:   1523 simulations (  1.52 %); accumulated:  66.51 %.
From   47000 to   47999 games:   1406 simulations (  1.41 %); accumulated:  67.92 %.
From   48000 to   48999 games:   1378 simulations (  1.38 %); accumulated:  69.29 %.
From   49000 to   49999 games:   1248 simulations (  1.25 %); accumulated:  70.54 %.
From   50000 to   50999 games:   1264 simulations (  1.26 %); accumulated:  71.81 %.
From   51000 to   51999 games:   1159 simulations (  1.16 %); accumulated:  72.96 %.
From   52000 to   52999 games:   1109 simulations (  1.11 %); accumulated:  74.07 %.
From   53000 to   53999 games:   1084 simulations (  1.08 %); accumulated:  75.16 %.
From   54000 to   54999 games:    929 simulations (  0.93 %); accumulated:  76.09 %.
From   55000 to   55999 games:    952 simulations (  0.95 %); accumulated:  77.04 %.
From   56000 to   56999 games:    851 simulations (  0.85 %); accumulated:  77.89 %.
From   57000 to   57999 games:    811 simulations (  0.81 %); accumulated:  78.70 %.
From   58000 to   58999 games:    833 simulations (  0.83 %); accumulated:  79.53 %.
From   59000 to   59999 games:    769 simulations (  0.77 %); accumulated:  80.30 %.
From   60000 to   60999 games:    741 simulations (  0.74 %); accumulated:  81.04 %.
From   61000 to   61999 games:    739 simulations (  0.74 %); accumulated:  81.78 %.
From   62000 to   62999 games:    705 simulations (  0.71 %); accumulated:  82.49 %.
From   63000 to   63999 games:    658 simulations (  0.66 %); accumulated:  83.15 %.
From   64000 to   64999 games:    602 simulations (  0.60 %); accumulated:  83.75 %.
From   65000 to   65999 games:    634 simulations (  0.63 %); accumulated:  84.38 %.
From   66000 to   66999 games:    571 simulations (  0.57 %); accumulated:  84.95 %.
From   67000 to   67999 games:    567 simulations (  0.57 %); accumulated:  85.52 %.
From   68000 to   68999 games:    538 simulations (  0.54 %); accumulated:  86.06 %.
From   69000 to   69999 games:    520 simulations (  0.52 %); accumulated:  86.58 %.
From   70000 to   70999 games:    519 simulations (  0.52 %); accumulated:  87.10 %.
From   71000 to   71999 games:    471 simulations (  0.47 %); accumulated:  87.57 %.
From   72000 to   72999 games:    468 simulations (  0.47 %); accumulated:  88.04 %.
From   73000 to   73999 games:    471 simulations (  0.47 %); accumulated:  88.51 %.
From   74000 to   74999 games:    427 simulations (  0.43 %); accumulated:  88.93 %.
From   75000 to   75999 games:    391 simulations (  0.39 %); accumulated:  89.32 %.
From   76000 to   76999 games:    361 simulations (  0.36 %); accumulated:  89.69 %.
From   77000 to   77999 games:    389 simulations (  0.39 %); accumulated:  90.07 %.
From   78000 to   78999 games:    357 simulations (  0.36 %); accumulated:  90.43 %.
From   79000 to   79999 games:    318 simulations (  0.32 %); accumulated:  90.75 %.
From   80000 to   80999 games:    364 simulations (  0.36 %); accumulated:  91.11 %.
From   81000 to   81999 games:    323 simulations (  0.32 %); accumulated:  91.44 %.
From   82000 to   82999 games:    322 simulations (  0.32 %); accumulated:  91.76 %.
From   83000 to   83999 games:    319 simulations (  0.32 %); accumulated:  92.08 %.
From   84000 to   84999 games:    261 simulations (  0.26 %); accumulated:  92.34 %.
From   85000 to   85999 games:    268 simulations (  0.27 %); accumulated:  92.61 %.
From   86000 to   86999 games:    260 simulations (  0.26 %); accumulated:  92.87 %.
From   87000 to   87999 games:    286 simulations (  0.29 %); accumulated:  93.15 %.
From   88000 to   88999 games:    235 simulations (  0.24 %); accumulated:  93.39 %.
From   89000 to   89999 games:    240 simulations (  0.24 %); accumulated:  93.63 %.
From   90000 to   90999 games:    239 simulations (  0.24 %); accumulated:  93.87 %.
From   91000 to   91999 games:    216 simulations (  0.22 %); accumulated:  94.08 %.
From   92000 to   92999 games:    238 simulations (  0.24 %); accumulated:  94.32 %.
From   93000 to   93999 games:    204 simulations (  0.20 %); accumulated:  94.52 %.
From   94000 to   94999 games:    199 simulations (  0.20 %); accumulated:  94.72 %.
From   95000 to   95999 games:    205 simulations (  0.21 %); accumulated:  94.93 %.
From   96000 to   96999 games:    174 simulations (  0.17 %); accumulated:  95.10 %.
From   97000 to   97999 games:    169 simulations (  0.17 %); accumulated:  95.27 %.
From   98000 to   98999 games:    152 simulations (  0.15 %); accumulated:  95.42 %.
From   99000 to   99999 games:    158 simulations (  0.16 %); accumulated:  95.58 %.
From  100000 to  100999 games:    161 simulations (  0.16 %); accumulated:  95.74 %.
From  101000 to  101999 games:    161 simulations (  0.16 %); accumulated:  95.90 %.
From  102000 to  102999 games:    152 simulations (  0.15 %); accumulated:  96.06 %.
From  103000 to  103999 games:    155 simulations (  0.16 %); accumulated:  96.21 %.
From  104000 to  104999 games:    140 simulations (  0.14 %); accumulated:  96.35 %.
From  105000 to  105999 games:    141 simulations (  0.14 %); accumulated:  96.49 %.
From  106000 to  106999 games:    110 simulations (  0.11 %); accumulated:  96.60 %.
From  107000 to  107999 games:    126 simulations (  0.13 %); accumulated:  96.73 %.
From  108000 to  108999 games:    127 simulations (  0.13 %); accumulated:  96.85 %.
From  109000 to  109999 games:    108 simulations (  0.11 %); accumulated:  96.96 %.
From  110000 to  110999 games:     99 simulations (  0.10 %); accumulated:  97.06 %.
From  111000 to  111999 games:    102 simulations (  0.10 %); accumulated:  97.16 %.
From  112000 to  112999 games:    112 simulations (  0.11 %); accumulated:  97.28 %.
From  113000 to  113999 games:    106 simulations (  0.11 %); accumulated:  97.38 %.
From  114000 to  114999 games:     96 simulations (  0.10 %); accumulated:  97.48 %.
From  115000 to  115999 games:     86 simulations (  0.09 %); accumulated:  97.56 %.
From  116000 to  116999 games:     97 simulations (  0.10 %); accumulated:  97.66 %.
From  117000 to  117999 games:     73 simulations (  0.07 %); accumulated:  97.73 %.
From  118000 to  118999 games:     97 simulations (  0.10 %); accumulated:  97.83 %.
From  119000 to  119999 games:     82 simulations (  0.08 %); accumulated:  97.91 %.
From  120000 to  120999 games:     92 simulations (  0.09 %); accumulated:  98.00 %.
From  121000 to  121999 games:     75 simulations (  0.08 %); accumulated:  98.08 %.
From  122000 to  122999 games:     68 simulations (  0.07 %); accumulated:  98.15 %.
From  123000 to  123999 games:     55 simulations (  0.06 %); accumulated:  98.20 %.
From  124000 to  124999 games:     63 simulations (  0.06 %); accumulated:  98.27 %.
From  125000 to  125999 games:     56 simulations (  0.06 %); accumulated:  98.32 %.
From  126000 to  126999 games:     78 simulations (  0.08 %); accumulated:  98.40 %.
From  127000 to  127999 games:     59 simulations (  0.06 %); accumulated:  98.46 %.
[...]
From  250000 to  250999 games:      0 simulations (  0.00 %); accumulated:  99.97 %.
From  251000 to  251999 games:      0 simulations (  0.00 %); accumulated:  99.97 %.
From  252000 to  252999 games:      2 simulations (  0.00 %); accumulated:  99.98 %.
From  253000 to  253999 games:      0 simulations (  0.00 %); accumulated:  99.98 %.
From  254000 to  254999 games:      2 simulations (  0.00 %); accumulated:  99.98 %.
From  255000 to  255999 games:      1 simulation  (  0.00 %); accumulated:  99.98 %.
[...]
From  314000 to  314999 games:      0 simulations (  0.00 %); accumulated: 100.00 %.
From  315000 to  315999 games:      0 simulations (  0.00 %); accumulated: 100.00 %.
From  316000 to  316999 games:      0 simulations (  0.00 %); accumulated: 100.00 %.
From  317000 to  317999 games:      1 simulation  (  0.00 %); accumulated: 100.00 %.

I hope no typos. What I want to say with all this stuff is that this test was unlikely to pass (only a 5.1% of chances with the current status) but there is a probability of 98.5% that this test is finished with less than 128k games, which is the original limit of games, although it can be raised in any moment. I include the percentiles just to have an idea of the estimated length of the run in the framework. For example, that this run will last more than 78k games with a probability of 9.9% more less.

Anyway, I am not fan of stopping SPRT. There was a heavy bug in the changed code?

Regards from Spain.

Ajedrecista.

[Eelco de Groot]

Hello:

My ic diff

LLR: -1.47 (-2.94,2.94) [0.00,5.00]
Total: 26127 W: 4737 L: 4688 D: 16702

sprt @ 10+0.1 th 1 Use intelligent contempt: increase/decrease our contempt for every fail high/low then reset like delta window.

I ran 100000 SPRT(0, 5) simulations (alpha = 5%, beta = 5%) with the starting point 4737 wins, 4688 loses and 16702 draws. I used the parameters taken from this sample of 26127 games: drawelo ~ 262.9849; Elo gain ~ 1.1019 Bayeselo. I know that the Elo gain has uncertainties (± some Bayeselo) but anyway, I took the central point. Then:

Code: Select all

Passes:  5108/100000 ~  5.11%.
Fails:  94892/100000 ~ 94.89%.

----------------------------------------------------------------------------------------

There are  83645 simulations with score > 50% that failed SPRT.
There are   1390 simulations with score = 50% that failed SPRT.

----------------------------------------------------------------------------------------

Shortest simulation:
FAIL after  26699 games (+ 4806 - 4818 = 17075).      {   572 games to go}.

Median simulations (there are three):
FAIL after  38744 games (+ 6968 - 6929 = 24847).      { 12617 games to go}.
FAIL after  38744 games (+ 7073 - 7033 = 24638).      { 12617 games to go}.
FAIL after  38744 games (+ 7049 - 7011 = 24684).      { 12617 games to go}.

Average number of games ~ 47579 games.                { 21452 games to go}.
With 95% confidence intervals ~ [47431, 47728] games. {From 21304 to 21601 games to go}.

Longest simulation:
FAIL after 317368 games (+57786 -56562 =203020).      {291241 games to go}.

----------------------------------------------------------------------------------------

Summary of the distribution of the length of simulations:

From   26000 to   26999 games:     41 simulations (  0.04 %); accumulated:   0.04 %.
From   27000 to   27999 games:   2062 simulations (  2.06 %); accumulated:   2.10 %.
From   28000 to   28999 games:   5028 simulations (  5.03 %); accumulated:   7.13 %.
From   29000 to   29999 games:   6027 simulations (  6.03 %); accumulated:  13.16 %.
From   30000 to   30999 games:   5993 simulations (  5.99 %); accumulated:  19.15 %.
From   31000 to   31999 games:   5532 simulations (  5.53 %); accumulated:  24.68 %.
From   32000 to   32999 games:   4985 simulations (  4.99 %); accumulated:  29.67 %.
From   33000 to   33999 games:   4399 simulations (  4.40 %); accumulated:  34.07 %.
From   34000 to   34999 games:   3990 simulations (  3.99 %); accumulated:  38.06 %.
From   35000 to   35999 games:   3590 simulations (  3.59 %); accumulated:  41.65 %.
From   36000 to   36999 games:   3331 simulations (  3.33 %); accumulated:  44.98 %.
From   37000 to   37999 games:   2906 simulations (  2.91 %); accumulated:  47.88 %.
From   38000 to   38999 games:   2777 simulations (  2.78 %); accumulated:  50.66 %.
From   39000 to   39999 games:   2491 simulations (  2.49 %); accumulated:  53.15 %.
From   40000 to   40999 games:   2396 simulations (  2.40 %); accumulated:  55.55 %.
From   41000 to   41999 games:   2085 simulations (  2.09 %); accumulated:  57.63 %.
From   42000 to   42999 games:   2024 simulations (  2.02 %); accumulated:  59.66 %.
From   43000 to   43999 games:   1933 simulations (  1.93 %); accumulated:  61.59 %.
From   44000 to   44999 games:   1739 simulations (  1.74 %); accumulated:  63.33 %.
From   45000 to   45999 games:   1657 simulations (  1.66 %); accumulated:  64.99 %.
From   46000 to   46999 games:   1523 simulations (  1.52 %); accumulated:  66.51 %.
From   47000 to   47999 games:   1406 simulations (  1.41 %); accumulated:  67.92 %.
From   48000 to   48999 games:   1378 simulations (  1.38 %); accumulated:  69.29 %.
From   49000 to   49999 games:   1248 simulations (  1.25 %); accumulated:  70.54 %.
From   50000 to   50999 games:   1264 simulations (  1.26 %); accumulated:  71.81 %.
From   51000 to   51999 games:   1159 simulations (  1.16 %); accumulated:  72.96 %.
From   52000 to   52999 games:   1109 simulations (  1.11 %); accumulated:  74.07 %.
From   53000 to   53999 games:   1084 simulations (  1.08 %); accumulated:  75.16 %.
From   54000 to   54999 games:    929 simulations (  0.93 %); accumulated:  76.09 %.
From   55000 to   55999 games:    952 simulations (  0.95 %); accumulated:  77.04 %.
From   56000 to   56999 games:    851 simulations (  0.85 %); accumulated:  77.89 %.
From   57000 to   57999 games:    811 simulations (  0.81 %); accumulated:  78.70 %.
From   58000 to   58999 games:    833 simulations (  0.83 %); accumulated:  79.53 %.
From   59000 to   59999 games:    769 simulations (  0.77 %); accumulated:  80.30 %.
From   60000 to   60999 games:    741 simulations (  0.74 %); accumulated:  81.04 %.
From   61000 to   61999 games:    739 simulations (  0.74 %); accumulated:  81.78 %.
From   62000 to   62999 games:    705 simulations (  0.71 %); accumulated:  82.49 %.
From   63000 to   63999 games:    658 simulations (  0.66 %); accumulated:  83.15 %.
From   64000 to   64999 games:    602 simulations (  0.60 %); accumulated:  83.75 %.
From   65000 to   65999 games:    634 simulations (  0.63 %); accumulated:  84.38 %.
From   66000 to   66999 games:    571 simulations (  0.57 %); accumulated:  84.95 %.
From   67000 to   67999 games:    567 simulations (  0.57 %); accumulated:  85.52 %.
From   68000 to   68999 games:    538 simulations (  0.54 %); accumulated:  86.06 %.
From   69000 to   69999 games:    520 simulations (  0.52 %); accumulated:  86.58 %.
From   70000 to   70999 games:    519 simulations (  0.52 %); accumulated:  87.10 %.
From   71000 to   71999 games:    471 simulations (  0.47 %); accumulated:  87.57 %.
From   72000 to   72999 games:    468 simulations (  0.47 %); accumulated:  88.04 %.
From   73000 to   73999 games:    471 simulations (  0.47 %); accumulated:  88.51 %.
From   74000 to   74999 games:    427 simulations (  0.43 %); accumulated:  88.93 %.
From   75000 to   75999 games:    391 simulations (  0.39 %); accumulated:  89.32 %.
From   76000 to   76999 games:    361 simulations (  0.36 %); accumulated:  89.69 %.
From   77000 to   77999 games:    389 simulations (  0.39 %); accumulated:  90.07 %.
From   78000 to   78999 games:    357 simulations (  0.36 %); accumulated:  90.43 %.
From   79000 to   79999 games:    318 simulations (  0.32 %); accumulated:  90.75 %.
From   80000 to   80999 games:    364 simulations (  0.36 %); accumulated:  91.11 %.
From   81000 to   81999 games:    323 simulations (  0.32 %); accumulated:  91.44 %.
From   82000 to   82999 games:    322 simulations (  0.32 %); accumulated:  91.76 %.
From   83000 to   83999 games:    319 simulations (  0.32 %); accumulated:  92.08 %.
From   84000 to   84999 games:    261 simulations (  0.26 %); accumulated:  92.34 %.
From   85000 to   85999 games:    268 simulations (  0.27 %); accumulated:  92.61 %.
From   86000 to   86999 games:    260 simulations (  0.26 %); accumulated:  92.87 %.
From   87000 to   87999 games:    286 simulations (  0.29 %); accumulated:  93.15 %.
From   88000 to   88999 games:    235 simulations (  0.24 %); accumulated:  93.39 %.
From   89000 to   89999 games:    240 simulations (  0.24 %); accumulated:  93.63 %.
From   90000 to   90999 games:    239 simulations (  0.24 %); accumulated:  93.87 %.
From   91000 to   91999 games:    216 simulations (  0.22 %); accumulated:  94.08 %.
From   92000 to   92999 games:    238 simulations (  0.24 %); accumulated:  94.32 %.
From   93000 to   93999 games:    204 simulations (  0.20 %); accumulated:  94.52 %.
From   94000 to   94999 games:    199 simulations (  0.20 %); accumulated:  94.72 %.
From   95000 to   95999 games:    205 simulations (  0.21 %); accumulated:  94.93 %.
From   96000 to   96999 games:    174 simulations (  0.17 %); accumulated:  95.10 %.
From   97000 to   97999 games:    169 simulations (  0.17 %); accumulated:  95.27 %.
From   98000 to   98999 games:    152 simulations (  0.15 %); accumulated:  95.42 %.
From   99000 to   99999 games:    158 simulations (  0.16 %); accumulated:  95.58 %.
From  100000 to  100999 games:    161 simulations (  0.16 %); accumulated:  95.74 %.
From  101000 to  101999 games:    161 simulations (  0.16 %); accumulated:  95.90 %.
From  102000 to  102999 games:    152 simulations (  0.15 %); accumulated:  96.06 %.
From  103000 to  103999 games:    155 simulations (  0.16 %); accumulated:  96.21 %.
From  104000 to  104999 games:    140 simulations (  0.14 %); accumulated:  96.35 %.
From  105000 to  105999 games:    141 simulations (  0.14 %); accumulated:  96.49 %.
From  106000 to  106999 games:    110 simulations (  0.11 %); accumulated:  96.60 %.
From  107000 to  107999 games:    126 simulations (  0.13 %); accumulated:  96.73 %.
From  108000 to  108999 games:    127 simulations (  0.13 %); accumulated:  96.85 %.
From  109000 to  109999 games:    108 simulations (  0.11 %); accumulated:  96.96 %.
From  110000 to  110999 games:     99 simulations (  0.10 %); accumulated:  97.06 %.
From  111000 to  111999 games:    102 simulations (  0.10 %); accumulated:  97.16 %.
From  112000 to  112999 games:    112 simulations (  0.11 %); accumulated:  97.28 %.
From  113000 to  113999 games:    106 simulations (  0.11 %); accumulated:  97.38 %.
From  114000 to  114999 games:     96 simulations (  0.10 %); accumulated:  97.48 %.
From  115000 to  115999 games:     86 simulations (  0.09 %); accumulated:  97.56 %.
From  116000 to  116999 games:     97 simulations (  0.10 %); accumulated:  97.66 %.
From  117000 to  117999 games:     73 simulations (  0.07 %); accumulated:  97.73 %.
From  118000 to  118999 games:     97 simulations (  0.10 %); accumulated:  97.83 %.
From  119000 to  119999 games:     82 simulations (  0.08 %); accumulated:  97.91 %.
From  120000 to  120999 games:     92 simulations (  0.09 %); accumulated:  98.00 %.
From  121000 to  121999 games:     75 simulations (  0.08 %); accumulated:  98.08 %.
From  122000 to  122999 games:     68 simulations (  0.07 %); accumulated:  98.15 %.
From  123000 to  123999 games:     55 simulations (  0.06 %); accumulated:  98.20 %.
From  124000 to  124999 games:     63 simulations (  0.06 %); accumulated:  98.27 %.
From  125000 to  125999 games:     56 simulations (  0.06 %); accumulated:  98.32 %.
From  126000 to  126999 games:     78 simulations (  0.08 %); accumulated:  98.40 %.
From  127000 to  127999 games:     59 simulations (  0.06 %); accumulated:  98.46 %.
[...]
From  250000 to  250999 games:      0 simulations (  0.00 %); accumulated:  99.97 %.
From  251000 to  251999 games:      0 simulations (  0.00 %); accumulated:  99.97 %.
From  252000 to  252999 games:      2 simulations (  0.00 %); accumulated:  99.98 %.
From  253000 to  253999 games:      0 simulations (  0.00 %); accumulated:  99.98 %.
From  254000 to  254999 games:      2 simulations (  0.00 %); accumulated:  99.98 %.
From  255000 to  255999 games:      1 simulation  (  0.00 %); accumulated:  99.98 %.
[...]
From  314000 to  314999 games:      0 simulations (  0.00 %); accumulated: 100.00 %.
From  315000 to  315999 games:      0 simulations (  0.00 %); accumulated: 100.00 %.
From  316000 to  316999 games:      0 simulations (  0.00 %); accumulated: 100.00 %.
From  317000 to  317999 games:      1 simulation  (  0.00 %); accumulated: 100.00 %.

I hope no typos. What I want to say with all this stuff is that this test was unlikely to pass (only a 5.1% of chances with the current status) but there is a probability of 98.5% that this test is finished with less than 128k games, which is the original limit of games, although it can be raised in any moment. I include the percentiles just to have an idea of the estimated length of the run in the framework. For example, that this run will last more than 78k games with a probability of 9.9% more less.

Anyway, I am not fan of stopping SPRT. There was a heavy bug in the changed code?

Regards from Spain.

Ajedrecista.

Hello Jesús,

I do not know why exactly it was stopped. There are no comments to the patch and the 'Actions' list is already past that date. You would have to ask Jonathan himself. But as Jonathan later started tuning test against Stockfish DD I can understand he had enough information from the intermediate result and it (code as it was at that point) would not be committed anyway (even if it had passed both STC and LTC!) unless it was clear that the contempt actually works against a weaker opponent, and preferably tested against more different opponents etc. It is a tricky subject and, just my five cents, not my favorite idea (because, without clear information about who your opponent is, it is in my opinion better to assume the worst and "Play against the board" as Jan Timman would put it. Like Fischer did, no tricks or 'Schwindling', just the strongest moves should be enough )

Regards,
Eelco

[Eelco de Groot]

Jean-Paul pointed out to me that Ipman has made Stockfish compiles available that include Intelligent Contempt Maybe of interest to Neel Basant.
http://immortalchess.net/forum/showpost ... tcount=372

Thank you Jean-Paul!

I trust that sources were included but I'm not going to check all that right now.
I just wanted to make use of this opportunity to say that the other contribution from Rocky640 is the really interesting one, although I have not studied his patch yet. And no I hope that all the other Stockfish programmers are not going to be put out of work to leave everything to him as Damir would like. Thank you very much Rocky