Help to modify bayeselo.exe

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Vinvin
Posts: 5311
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Help to modify bayeselo.exe

Post by Vinvin »

Ferdy wrote:Sample run from ccrl and cegt games, added diff elo, win % and w-l-d data. I am checking randomly the W-L-D, will be released couple of hours later if I can't find any errors.
Did you change something in the draw calculation ? I made a run on CCRL 40/40 ( http://computerchess.org.uk/ccrl/4040/r ... t_all.html ) and I got a number exactly the half with your new app :
CCRL : 43.0%
1 Houdini 2.0c 64-bit 4CPU 3305.36 0.00 28.01 27.61 407 64.74% 3210.31 21.50% 43.24% 176-56-175

BayesElo (Ferdinand Mosca) : 21.50%
1 Houdini 2.0c 64-bit 4CPU 3305 +28 −28 64.7% −94.8 43.0% 407

Old calculation seems correct : 175/407 = 42.99%
Top
Ferdy
Posts: 4851
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Help to modify bayeselo.exe

Post by Ferdy »

Vinvin wrote:
Ferdy wrote:Sample run from ccrl and cegt games, added diff elo, win % and w-l-d data. I am checking randomly the W-L-D, will be released couple of hours later if I can't find any errors.
Did you change something in the draw calculation ? I made a run on CCRL 40/40 ( http://computerchess.org.uk/ccrl/4040/r ... t_all.html ) and I got a number exactly the half with your new app :
CCRL : 43.0%
1 Houdini 2.0c 64-bit 4CPU 3305.36 0.00 28.01 27.61 407 64.74% 3210.31 21.50% 43.24% 176-56-175

BayesElo (Ferdinand Mosca) : 21.50%
1 Houdini 2.0c 64-bit 4CPU 3305 +28 −28 64.7% −94.8 43.0% 407

Old calculation seems correct : 175/407 = 42.99%
You were right, I changed it to (draw_count/2) / Games. I thought this is a draw score percentage, so this is a draw count percentage.
I will prepare the fix.
Top
Ferdy
Posts: 4851
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Help to modify bayeselo.exe

Post by Ferdy »

Just uploaded v0057.2, seems to be working now :) .

Code: Select all

Rank Name                                     Elo     Diff     +     -      Games  Score    Oppo.   Draws     Win          W-L-D 
   1 Houdini 2.0c 64-bit 4CPU              614.01     0.00  30.56  30.02      353  66.01%  509.33  40.79%  45.61%       161-48-144
   2 Houdini 1.5a 64-bit 4CPU              601.59   -12.42  21.92  21.57      695  67.19%  489.19  41.44%  46.47%       323-84-288
   3 Rybka 4.1 64-bit 4CPU                 563.05   -38.54  20.05  19.89      748  60.56%  498.07  52.41%  34.36%       257-99-392
   4 Stockfish 2.2.2 64-bit 4CPU           562.12    -0.93  30.70  30.43      309  58.90%  509.31  55.02%  31.39%        97-42-170
   5 Rybka 4 64-bit 4CPU                   560.72    -1.40  18.50  18.39      949  65.38%  458.62  46.47%  42.15%       400-108-441
   6 Critter 1.4 64-bit 4CPU               558.12    -2.59  70.06  68.56       57  59.65%  497.07  59.65%  29.82%        17-6-34
   7 Houdini 1.5a 64-bit                   556.50    -1.62  16.81  16.67     1169  64.07%  462.77  42.43%  42.86%       501-172-496
   8 Critter 1.2 64-bit 4CPU               556.23    -0.28  26.63  26.32      433  61.20%  485.46  48.96%  36.72%       159-62-212
   9 Houdini 2.0c 64-bit                   543.67   -12.56  16.48  16.36     1304  67.06%  423.19  36.27%  48.93%       638-193-473
  10 Stockfish 2.0.1 64-bit 4CPU           541.63    -2.04  24.12  24.12      514  56.42%  501.02  48.64%  32.10%       165-99-250
Top
Vinvin
Posts: 5311
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Help to modify bayeselo.exe

Post by Vinvin »

Thanks ! :D

If you want to spend more time, I remember I discussed about a bug when doing "exactdist" on too large spread between strongest and weakest engines (that happend with my very long list, difference is higher than 2500 Elo) ... Rémi told me he used a maximum difference (I don't remember how much). Is it possible to rise this amount to 9000 Elo ? I'll look in my mails tonight to see the current max amount.
Ferdy wrote:Just uploaded v0057.2, seems to be working now :) .
Top
Vinvin
Posts: 5311
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Help to modify bayeselo.exe

Post by Vinvin »

Vinvin wrote:Thanks ! :D

If you want to spend more time, I remember I discussed about a bug when doing "exactdist" on too large spread between strongest and weakest engines (that happend with my very long list, difference is higher than 2500 Elo) ... Rémi told me he used a maximum difference (I don't remember how much). Is it possible to rise this amount to 9000 Elo ? I'll look in my mails tonight to see the current max amount.
Ferdy wrote:Just uploaded v0057.2, seems to be working now :) .
I found the old mail :

Code: Select all

Le 2/11/2009 19:25, Rémi Coulom a écrit :
> OK, merci. J'ai compris le bug. Le problème est dans "exactdist": je ne pensais pas que le niveau des plus faibles pourrait descendre en-dessous de -1500 (avant ajustement par la commande offset).
So, Ferdy, could you change this [-1500, +1500] to [-4500,+4500], please ?
I saw this value in the code ...
Top
Ferdy
Posts: 4851
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Help to modify bayeselo.exe

Post by Ferdy »

Vinvin wrote:
Vinvin wrote:Thanks ! :D

If you want to spend more time, I remember I discussed about a bug when doing "exactdist" on too large spread between strongest and weakest engines (that happend with my very long list, difference is higher than 2500 Elo) ... Rémi told me he used a maximum difference (I don't remember how much). Is it possible to rise this amount to 9000 Elo ? I'll look in my mails tonight to see the current max amount.
Ferdy wrote:Just uploaded v0057.2, seems to be working now :) .
I found the old mail :

Code: Select all

Le 2/11/2009 19:25, Rémi Coulom a écrit :
> OK, merci. J'ai compris le bug. Le problème est dans "exactdist": je ne pensais pas que le niveau des plus faibles pourrait descendre en-dessous de -1500 (avant ajustement par la commande offset).
So, Ferdy, could you change this [-1500, +1500] to [-4500,+4500], please ?
I saw this value in the code ...
Is this what you mean? Found in CEloRatingCUI.cpp.

Code: Select all

eloMin(-1500),
eloMax(1500),
Top
Vinvin
Posts: 5311
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Help to modify bayeselo.exe

Post by Vinvin »

Ferdy wrote:
Vinvin wrote:
Vinvin wrote:Thanks ! :D

If you want to spend more time, I remember I discussed about a bug when doing "exactdist" on too large spread between strongest and weakest engines (that happend with my very long list, difference is higher than 2500 Elo) ... Rémi told me he used a maximum difference (I don't remember how much). Is it possible to rise this amount to 9000 Elo ? I'll look in my mails tonight to see the current max amount.
Ferdy wrote:Just uploaded v0057.2, seems to be working now :) .
I found the old mail :

Code: Select all

Le 2/11/2009 19:25, Rémi Coulom a écrit :
> OK, merci. J'ai compris le bug. Le problème est dans "exactdist": je ne pensais pas que le niveau des plus faibles pourrait descendre en-dessous de -1500 (avant ajustement par la commande offset).
So, Ferdy, could you change this [-1500, +1500] to [-4500,+4500], please ?
I saw this value in the code ...
Is this what you mean? Found in CEloRatingCUI.cpp.

Code: Select all

eloMin(-1500),
eloMax(1500),
Yes !
I hope there will be no side effect ...
Top
Ferdy
Posts: 4851
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Help to modify bayeselo.exe

Post by Ferdy »

From my sample run. Check out the + / - columns. Just tell me if I will upload the new version.

Default at top 10

Code: Select all

Rank Name                                     Elo     Diff     +     -      Games  Score    Oppo.   Draws     Win          W-L-D 
   1 Houdini 2.0c 64-bit 4CPU              849.84     0.00  30.57  30.03      353  66.01%  745.17  40.79%  45.61%       161-48-144
   2 Houdini 1.5a 64-bit 4CPU              837.42   -12.42  21.91  21.59      695  67.19%  725.02  41.44%  46.47%       323-84-288
   3 Rybka 4.1 64-bit 4CPU                 798.88   -38.54  20.06  19.88      748  60.56%  733.90  52.41%  34.36%       257-99-392
   4 Stockfish 2.2.2 64-bit 4CPU           797.96    -0.93  30.71  30.44      309  58.90%  745.15  55.02%  31.39%        97-42-170
   5 Rybka 4 64-bit 4CPU                   796.55    -1.41  18.49  18.40      949  65.38%  694.44  46.47%  42.15%       400-108-441
   6 Critter 1.4 64-bit 4CPU               793.96    -2.59  70.09  68.59       57  59.65%  732.90  59.65%  29.82%        17-6-34
   7 Houdini 1.5a 64-bit                   792.34    -1.61  16.81  16.67     1169  64.07%  698.60  42.43%  42.86%       501-172-496
   8 Critter 1.2 64-bit 4CPU               792.06    -0.28  26.65  26.34      433  61.20%  721.29  48.96%  36.72%       159-62-212
   9 Houdini 2.0c 64-bit                   779.51   -12.56  16.47  16.38     1304  67.06%  659.02  36.27%  48.93%       638-193-473
  10 Stockfish 2.0.1 64-bit 4CPU           777.46    -2.05  24.11  24.12      514  56.42%  736.86  48.64%  32.10%       165-99-250
New version at -4500/4500

Code: Select all

Rank Name                                     Elo     Diff     +     -      Games  Score    Oppo.   Draws     Win          W-L-D 
   1 Houdini 2.0c 64-bit 4CPU              849.84     0.00  31.49  31.58      353  66.01%  745.17  40.79%  45.61%       161-48-144
   2 Houdini 1.5a 64-bit 4CPU              837.42   -12.42  24.08  22.75      695  67.19%  725.02  41.44%  46.47%       323-84-288
   3 Rybka 4.1 64-bit 4CPU                 798.88   -38.54  22.35  22.26      748  60.56%  733.90  52.41%  34.36%       257-99-392
   4 Stockfish 2.2.2 64-bit 4CPU           797.96    -0.93  32.12  31.48      309  58.90%  745.15  55.02%  31.39%        97-42-170
   5 Rybka 4 64-bit 4CPU                   796.55    -1.41  19.83  20.68      949  65.38%  694.44  46.47%  42.15%       400-108-441
   6 Critter 1.4 64-bit 4CPU               793.96    -2.59  70.52  69.04       57  59.65%  732.90  59.65%  29.82%        17-6-34
   7 Houdini 1.5a 64-bit                   792.34    -1.61  19.50  18.05     1169  64.07%  698.60  42.43%  42.86%       501-172-496
   8 Critter 1.2 64-bit 4CPU               792.06    -0.28  28.49  27.39      433  61.20%  721.29  48.96%  36.72%       159-62-212
   9 Houdini 2.0c 64-bit                   779.51   -12.56  18.46  19.22     1304  67.06%  659.02  36.27%  48.93%       638-193-473
  10 Stockfish 2.0.1 64-bit 4CPU           777.46    -2.05  26.20  26.01      514  56.42%  736.86  48.64%  32.10%       165-99-250



Default at last 10

Code: Select all

2024 DarkFUSC 0.9                         -1261.35    -8.45 153.04 114.45       31   9.68%  -772.74  12.90%   3.23%         1-26-4
2025 Gringo 1.4.1                         -1272.43   -11.08 163.53 106.26       30  10.00%  -815.40   0.00%  10.00%         3-27-0
2026 Youk 1.05                            -1291.36   -18.93 174.76  89.55       31   3.23%  -747.24   6.45%   0.00%         0-29-2
2027 Xadreco 5.1                          -1310.60   -19.24 174.93  71.59       30  10.00%  -858.62   6.67%   6.67%         2-26-2
2028 Gargamella 0.0.1                     -1311.96    -1.36 123.98  70.05       67   8.96%  -794.05   2.99%   7.46%         5-60-2
2029 Pyotr Amateur v0.6                   -1435.74  -123.78 241.77 -49.94       35   2.86%  -761.41   0.00%   2.86%         1-34-0
2030 Chad'sChess 0.15                     -1444.26    -8.52 104.57 -55.57      288   4.51%  -795.41   2.78%   3.12%         9-271-8
2031 Pyotr Club Ed 1.0                    -1449.81    -5.55 220.08 -63.40       32   3.12%  -821.58   6.25%   0.00%         0-30-2
2032 Geko 0.4.3                           -1486.88   -37.07 171.23 -98.63      105   3.33%  -863.96   2.86%   1.90%         2-100-3
2033 Xadreco 5.6                          -1548.71   -61.83 322.83 -162.12       31   0.00%  -763.47   0.00%   0.00%         0-31-0
New version at last 10

Code: Select all

2024 DarkFUSC 0.9                         -1261.35    -8.45 150.53 183.14       31   9.68%  -772.74  12.90%   3.23%         1-26-4
2025 Gringo 1.4.1                         -1272.43   -11.08 158.61 216.05       30  10.00%  -815.40   0.00%  10.00%         3-27-0
2026 Youk 1.05                            -1291.36   -18.93 166.48 245.18       31   3.23%  -747.24   6.45%   0.00%         0-29-2
2027 Xadreco 5.1                          -1310.60   -19.24 165.79 213.47       30  10.00%  -858.62   6.67%   6.67%         2-26-2
2028 Gargamella 0.0.1                     -1311.96    -1.36 120.20 145.51       67   8.96%  -794.05   2.99%   7.46%         5-60-2
2029 Pyotr Amateur v0.6                   -1435.74  -123.78 196.14 301.79       35   2.86%  -761.41   0.00%   2.86%         1-34-0
2030 Chad'sChess 0.15                     -1444.26    -8.52  70.25  81.79      288   4.51%  -795.41   2.78%   3.12%         9-271-8
2031 Pyotr Club Ed 1.0                    -1449.81    -5.55 172.55 241.32       32   3.12%  -821.58   6.25%   0.00%         0-30-2
2032 Geko 0.4.3                           -1486.88   -37.07 112.08 148.24      105   3.33%  -863.96   2.86%   1.90%         2-100-3
2033 Xadreco 5.6                          -1548.71   -61.83 225.34 414.84       31   0.00%  -763.47   0.00%   0.00%         0-31-0
Top
Vinvin
Posts: 5311
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Help to modify bayeselo.exe

Post by Vinvin »

Ferdy wrote:From my sample run. Check out the + / - columns. Just tell me if I will upload the new version.
...
I think it's strange 1st line change so clearly +30.57;-30.03 become +31.49;-31.58.
As I don't know the very technical details, I ask the question to Rémi by mail ... I hope a fast response.
Top
Vinvin
Posts: 5311
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Help to modify bayeselo.exe

Post by Vinvin »

Ferdy wrote:From my sample run. Check out the + / - columns. Just tell me if I will upload the new version.
...
Rémi confirmed it's OK.
You can released and thanks again, now, my big list will stay in the same order after little modifications !
Top

Return to “Computer Chess Club: General Topics”