Movei possible better personality

[Uri Blass]

Latest Movei default personality is progress 0 0 0

I believed that after implementing hash for pruning path dependent evaluation cannot help so I disabled it by default by having the 0 0 0 parameters but testing so far give better result for progress 10 10 10.

I tested both of the personality at 40/40 against 10% rybka2.3.2a

10% rybka2.3.2a probably win both matches but it scored better against
progress 0 0 0.

results are:

10% Rybka2.3.2a 32 bit -Movei default 57-43
10% Rybka2.3.2a 32 bit-Movei 22-16 with 35 draws that is only 54%(match still is not finished)

note also that in direct match at fast time control movei 10 10 10 also scored more than 50% but I also do not have enough games.

Note that I find that there are cases when rybka responds immediatly based on moves that it has in the hash and I wonder if it happens also in regular ponder off games(40/4 is time control that is slow enough not to do it for most programs even if they have some move in the hash but maybe it is good in case of having move with sufficient depth to do it).

Uri

[Ovyron]

Note that I find that there are cases when rybka responds immediatly based on moves that it has in the hash and I wonder if it happens also in regular ponder off games.

Yes, I have seen Rybka doing this. I also noticed it plays a bit faster once she thinks she's winning.

[Kirill Kryukov]

Hi Uri. This is interesting. Do you suggest to use progress 10 10 10 for testing latest Movei in CCRL blitz? Because I don't think I can test both, so which do you think will be stronger? It will take a couple of weeks before I can start testing Movei, so please take your time if you can't answer right now. Thanks!

[Uri Blass]

I am going to do more tests and I am not sure now.

I guess that 10 10 10 is slightly stronger but I guess that the difference is only 10 elo or something in that enviroment.

result against 10% of rybka after 88 games is now 29-19 with 20 draws (still better than 43% the previous version but the difference is small)


Uri

[Uri Blass]

correction:I meant 40 draws and not 20 draws in the previous post.

Uri

[Kirill Kryukov]

Thanks, Uri! I will look forward to your updated results and conclusion after more games! Good luck!

[Uri Blass]

Final result of the match

32-21-47 for rybka that is 55.5-45.5

Note that it is better than the loss 57-43

Next match against rybka is going to be played at faster time control 40/10 for movei against 40/1 of rybka.

Default personality lost 51-49 in these conditions.

Uri