Some RomiChess progress

[Michael Sherwin]

RomiChessP3n CCRL 2424

RomiChessX

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws 

  1 RomiChess                      : 2446   27  27   500    62.9 %   2354   23.8 % 
  2 Yace                           : 2415   60  60   100    45.5 %   2447   25.0 % 
  3 Tcb0052                        : 2365   60  61   100    38.5 %   2447   25.0 % 
  4 Horizon_4_4                    : 2354   61  62   100    37.0 %   2447   24.0 % 
  5 OliThink532_x64                : 2343   61  62   100    35.5 %   2447   25.0 % 
  6 Bitfoot-1.0.65acfcb-win64      : 2291   65  67   100    29.0 %   2447   20.0 % 

All the individual results are better than before and gives a CCRL equivalent of 2502. And following is a match with Graham's last D7 winner Jumbo and is also within the error margins confirming an improvement.

RomiChess - Jumbo64-0.6.10-bb : 50.5/100 36-35-29 (011101=01=000==01=1==1100=11111110=1=11==10=1010=0=100=00=10=1=0101=1010010=001==1100011==0====10000) 51% +7

[CMCanavessi]

So is a new version about to be released? Next week CCLS Season 3 will start

[Michael Sherwin]

Hi Carlos, Not likely as this is only the second round of testing. Besides, I thought you were interested in seeing if Romi's learning would help her to climb through the ranks. If you replace the old Romi with the new then that goes out the window.

[Michael Sherwin]

RomiChessP3n CCRL 2424

RomiChessX

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws 

  1 RomiChess                      : 2446   27  27   500    62.9 %   2354   23.8 % 
  2 Yace                           : 2415   60  60   100    45.5 %   2447   25.0 % 
  3 Tcb0052                        : 2365   60  61   100    38.5 %   2447   25.0 % 
  4 Horizon_4_4                    : 2354   61  62   100    37.0 %   2447   24.0 % 
  5 OliThink532_x64                : 2343   61  62   100    35.5 %   2447   25.0 % 
  6 Bitfoot-1.0.65acfcb-win64      : 2291   65  67   100    29.0 %   2447   20.0 % 

All the individual results are better than before and gives a CCRL equivalent of 2502. And following is a match with Graham's last D7 winner Jumbo and is also within the error margins confirming an improvement.

RomiChess - Jumbo64-0.6.10-bb : 50.5/100 36-35-29 (011101=01=000==01=1==1100=11111110=1=11==10=1010=0=100=00=10=1=0101=1010010=001==1100011==0====10000) 51% +7

As per request to keep Jumbo informed here is an update. Personally I think Jumbo has a bad case of puppy love for Romi.

All I did for this test was give the knights their own custom table for king safety. And it netted Romi one more point ...

RomiChess - Jumbo64-0.6.10-bb : 51.5/100 39-36-25 (1100=1101=0=1100001=11110111011110=====01==111000000=10=100=10===10010=01000=1=11=0110=001=0110=1=10) 52% +14

... unless it is just noise.

[Michael Sherwin]

RomiChessP3n CCRL 2424

RomiChessX

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws 

  1 RomiChess                      : 2446   27  27   500    62.9 %   2354   23.8 % 
  2 Yace                           : 2415   60  60   100    45.5 %   2447   25.0 % 
  3 Tcb0052                        : 2365   60  61   100    38.5 %   2447   25.0 % 
  4 Horizon_4_4                    : 2354   61  62   100    37.0 %   2447   24.0 % 
  5 OliThink532_x64                : 2343   61  62   100    35.5 %   2447   25.0 % 
  6 Bitfoot-1.0.65acfcb-win64      : 2291   65  67   100    29.0 %   2447   20.0 % 

All the individual results are better than before and gives a CCRL equivalent of 2502. And following is a match with Graham's last D7 winner Jumbo and is also within the error margins confirming an improvement.

RomiChess - Jumbo64-0.6.10-bb : 50.5/100 36-35-29 (011101=01=000==01=1==1100=11111110=1=11==10=1010=0=100=00=10=1=0101=1010010=001==1100011==0====10000) 51% +7

As per request to keep Jumbo informed here is an update. Personally I think Jumbo has a bad case of puppy love for Romi.

All I did for this test was give the knights their own custom table for king safety. And it netted Romi one more point ...

RomiChess - Jumbo64-0.6.10-bb : 51.5/100 39-36-25 (1100=1101=0=1100001=11110111011110=====01==111000000=10=100=10===10010=01000=1=11=0110=001=0110=1=10) 52% +14

... unless it is just noise.

Just some minor tweaking and cleanup of the eval. The 500 game 3rd test is underway. The 100 game Jumbo 3rd test has finished.

RomiChess - Jumbo64-0.6.10-bb : 53.0/100 44-38-18 (11=11010001100001110=011=111==111100=0000001000111010001==1=0==0=1111=0100==1100011=11110=10111=0100) 53% +21

I do not know what the 4th test will be as I have not thought that far ahead. And the reason for that is because I've never had 3 successful test in a row.

[Michael Sherwin]

This is why I hate testing new versions of Romichess.

My second test.

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws 

  1 RomiChess                      : 2446   27  27   500    62.9 %   2354   23.8 % 
  2 Yace                           : 2415   60  60   100    45.5 %   2447   25.0 % 
  3 Tcb0052                        : 2365   60  61   100    38.5 %   2447   25.0 % 
  4 Horizon_4_4                    : 2354   61  62   100    37.0 %   2447   24.0 % 
  5 OliThink532_x64                : 2343   61  62   100    35.5 %   2447   25.0 % 
  6 Bitfoot-1.0.65acfcb-win64      : 2291   65  67   100    29.0 %   2447   20.0 %

Plus Jumbo.
RomiChess - Jumbo64-0.6.10-bb : 51.5/100 39-36-25 52% +14 

My third test started with Jumbo. RomiChess - Jumbo64-0.6.10-bb : 53.0/100 44-38-18 53% +21 but was no good for the rest so it was aborted. Then the 4th test included Jumbo in the main event as well.

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  1 Yace                           : 2458   57  57   100    52.5 %   2441   31.0 %
  2 RomiChess                      : 2440   25  25   600    61.3 %   2361   23.8 %
  3 OliThink532_x64                : 2381   62  63   100    41.5 %   2441   19.0 %
  4 Jumbo64-0.6.10-bb              : 2367   60  61   100    39.5 %   2441   25.0 %
  5 Bitfoot-1.0.65acfcb-win64      : 2345   64  65   100    36.5 %   2441   17.0 %
  6 Tcb0052                        : 2318   60  61   100    33.0 %   2441   28.0 %
  7 Horizon_4_4                    : 2290   63  65   100    29.5 %   2441   23.0 %

Despite doing really well against Jumbo I have to consider it a fluke because the changes from test 3 to test 4 were very minor.

RomiChess - Jumbo64-0.6.10-bb : 61.0/100 48-26-26 61% +78

So 2 results were very good but the rest of the results were substantially worse. So I'm done posting updates for now and maybe I will just post the results for the release version whenever it is done.