CCRL Discussion Board

I see that some personalities of Glaurung1.2.1 were tested
I prefer seeing progress 10 10 10 personality in future tournaments.

progress1 progress2 progress3 are uci parameters that it is possible to change in movei.

Uri

Uri Blass wrote:I see that some personalities of Glaurung1.2.1 were tested
I prefer seeing progress 10 10 10 personality in future tournaments.

progress1 progress2 progress3 are uci parameters that it is possible to change in movei.

Uri

Hi Uri,

I'm prepared to give this personality a go.

Where exactly are the changes made and how would you like the engine named?
Movei 0.08.403.10.10.10 gets a bit longwinded.

Regards, Graham.

Graham Banks wrote:

Uri Blass wrote:I see that some personalities of Glaurung1.2.1 were tested
I prefer seeing progress 10 10 10 personality in future tournaments.

progress1 progress2 progress3 are uci parameters that it is possible to change in movei.

Uri

Hi Uri,

I'm prepared to give this personality a go.

Where exactly are the changes made and how would you like the engine named?
Movei 0.08.403.10.10.10 gets a bit longwinded.

Regards, Graham.

You can make the changes simply by changing the uci parameters of movei

progress1 progress2 progress3 are simply engines parameters that can be changed

Movei evaluates every node and the idea of these parameters is to give bonus for progress in the evaluation relative to what happened some plies earlier(the bonus for progress can be at most progress1+progress2+progress3 centi-pawns).

My intuition tells me that bigger value can be productive at longer time control but my intuition can be wrong so I promise nothing about playing strength and I have no time to do serious testing at long time control.

Uri

How about we call it Movei 0.08.403 Pro10?

Graham Banks wrote:How about we call it Movei 0.08.403 Pro10?

You can use that name.
pro10 may be 100 elo better or 100 elo weaker at ccrl time control relative to the default and I simply do not know.

After the last bad results of default 403 I feel that I can lose nothing by trying some significant change in the parameters.

Even if I find that pro10 is weaker at long time control then the default I can learn something from it for future versions that I release and I am afraid that testing a small increase in the progress parameter will give me only difference that is not significant.

Uri

Uri Blass wrote: After the last bad results of default 403 I feel that I can lose nothing by trying some significant change in the parameters.

Even if I find that pro10 is weaker at long time control then the default I can learn something from it for future versions that I release and I am afraid that testing a small increase in the progress parameter will give me only difference that is not significant.

Uri

I'll use Movei 0.08.403 Pro10 in some future tournaments and I'll ask some of my fellow testers to do the same.

Regards, Graham.

Graham Banks wrote:

Uri Blass wrote: After the last bad results of default 403 I feel that I can lose nothing by trying some significant change in the parameters.

Even if I find that pro10 is weaker at long time control then the default I can learn something from it for future versions that I release and I am afraid that testing a small increase in the progress parameter will give me only difference that is not significant.

Uri

I'll use Movei 0.08.403 Pro10 in some future tournaments and I'll ask some of my fellow testers to do the same.

Regards, Graham.

Thanks
I did some testing and first result are good.

I have no computer time for serious testing at long time control but
results at 5 minutes/40 moves were
movei00.8.403 pr10 14-Fruit2.0 8
Of course not enough games were played and I know that movei likes the first positions in the noomen match and it was also leading after 22 games in part of the other matches at different time controls that I played only to lose 100 game match later.

Uri

Uri - when I load Movei in the Shredder Classic II GUI and then open engine options, I do not get any "progress" parameters that I can change.

I haven't looked in Arena or Chessbase GUIs yet.

Graham Banks wrote:Uri - when I load Movei in the Shredder Classic II GUI and then open engine options, I do not get any "progress" parameters that I can change.

I haven't looked in Arena or Chessbase GUIs yet.

Do you use movei as winboard engine or as uci engine?

If you use movei as winboard engine you have no parameters that you can change inside the gui.

If you use movei as uci engine you should have progress parameters that you can change(I did not try it under shredder gui and shredder gui is not installed in the computer that I use now so I will check it later but I see no reason that things should be different under different gui's)

It is also possible to change personality under winboard but in that case you need to generate a special movei_changes403.ini file and write inside it progress 10 10 10 and put that file in the same folder as the exe file.

Note that movei winboard and movei uci have different default values of progress and I simply did a mistake when I change the default value under uci and forgot to do the same for the winboard version..

As winboard engine the default value are 2 3 5 and as uci engine the default values are 2 4 8.

I guess that there is no big difference between the playing strength of both options.

Uri

<snipped>

Uri Blass wrote: It is also possible to change personality under winboard but in that case you need to generate a special movei_changes403.ini file and write inside it progress 10 10 10 and put that file in the same folder as the exe file.

Uri

only one correction

as winboard engine you need to write progress 10 10 10 0 and without 4th number the program will not work.
There is no significance for the 4th number in 403 and I simply forgot about it and did not delete it from the parameters and the program check that it has all the parameters(otherwise it does not run).

Uri

Some Movei results at 40/4

http://www.computerchess.org.uk/ccrl/40 ... _length=30

10 10 10 running

Shaun

Thanks

Note that 403 10 10 10 performed better than 418 10 10 10 in my tests.

I did not play enough games so I do not know if it is because of luck or because of other reasons.

Uri

More results

http://computerchess.org.uk/ccrl/404.li ... _length=30

403 10 10 10 running

Shaun

I will start testing 403 10 10 10 at FRC when I have a machine available

ShaunBrewer wrote:More results

http://computerchess.org.uk/ccrl/404.li ... _length=30

403 10 10 10 running

Shaun

Thanks for your tests.

Uri

It seems that progress1=4 progress2=6 progress3=10 is better than the 10 10 10 personality(based on Jorge tests and also based on my own tests.

I see that testers still did not start to test 10 10 10 so I decided that I prefer testing 4 6 10 if possible.

Uri

In my last post I meant mainly to CCRL 40/40
I understood that 10 10 10 is running in CCRL 40/4

about FRC I did not test 4 6 10 at FRC to say if it is better than 10 10 10
in these conditions.

Uri

Update

this will probably be available view the .live site tonight
Obviously more games would be needed for proof but 418 (10 10 10) looks good at 40/4 time controls!

Shaun

ShaunBrewer wrote:Update

this will probably be available view the .live site tonight

Code: Select all

# Computed on 2007-05-15
# Based on "CCRL-404.[124178].pgn"

# Name                                    =>   ELO (  +    - )  Score   AvOp ( AvDf)  Draws Games (+W-L=D)
  Movei 0.08.418(10 10 10)                =>  2660 ( +27  -27)  48.7%   +7.6 ( 67.0)  27.8%   479 (+167-179=133)
  Movei 0.08.418                          =>  2655 ( +26  -27)  48.0%  +12.6 ( 68.1)  31.5%   480 (+155-174=151)
  Movei 0.08.418(3 7 10)                  =>  2648 ( +26  -26)  47.0%  +19.5 ( 69.9)  34.1%   478 (+143-172=163)
  Movei 0.08.403(10 10 10)                =>  2646 ( +27  -27)  46.6%  +21.6 ( 70.7)  25.2%   476 (+162-194=120)
  Movei 0.08.418(2 4 10)                  =>  2639 ( +27  -27)  45.9%  +28.8 ( 72.4)  27.1%   476 (+154-193=129)
  Movei 0.08.403                          =>  2637 ( +21  -21)  53.0%  -24.2 ( 67.2)  29.5%   828 (+317-267=244)

Obviously more games would be needed for proof but 418 (10 10 10) looks good at 40/4 time controls!

Shaun

Thanks

I believe that 403 10 10 10 was lucky to perform better than 418 10 10 10 in my tests.

I only tested both against fruit2.0 and 403 performed better but maybe 403 was lucky or fruit2.0 is not convenient opponent for it because in head to head match 418 scored slightly more than 50% against 403.

hopefully the version of 421 that I sent you can perform better than both and I will be happy if it can get rating near 2700.

I made the default progress values 10 10 10 inspite of some evidence that 4 6 10 is even better because based on your results of 3 7 10 I cannot believe that 4 6 10 is significantly better.

Uri

Website updated - I will check my emails

Movei at 40/4

Shaun