Alaric 707 Quick Results

[Peter Fendrich]

I set learn depth = 0 in the uci-options to set off learning. Is that the way to do it? What is the learning about? (own) book learning or something else?

This is learning out of the book. It is a simple "avoid bad lines again" learning. I don't want to look at 10 identical games losing...

Together with a real book it wont have much influence at all. Probably not even visible if the book is acceptable.

If you delete the learn.bin and set learndepth = 0 no learning will be created or used for sure. A null-content learn.bin will be created that have no information. I haven't tested this and I am not sure how Alaric will react in the next game with a null content file but you apparently tried it, so it should be fine.
/Peter

[Eelco de Groot]

I finished Alaric Deep IIb gauntlet.

Code: Select all

Rank Name                 Elo    +    - games score oppo. draws 
   1 Strelka 1.8         2843   25   24   500   67%  2724   34% 
   2 Rybka 1.0 Beta      2836   28   28   400   67%  2721   31% 
   3 Gambit Fruit 4bx    2778   28   27   400   61%  2703   31% 
   4 Spike 1.2           2762   27   27   400   54%  2731   35% 
   5 Chess Tiger 2007.1  2756   24   24   500   55%  2723   30% 
   6 Alaric 707          2720   23   23   600   51%  2714   29% 
   7 Naum 2.0            2716   27   27   400   52%  2698   33% 
   8 Glaurung 2 - eps/4  2709   22   21   650   50%  2707   30% 
   9 Alaric Deep IIb     2703   22   23   600   48%  2714   29% 
  10 CM9 Tribute         2666   28   28   400   45%  2697   24% 
  11 CMX Descent         2660   27   28   400   42%  2718   28% 
  12 Alaric 704          2657   23   23   600   42%  2714   27% 
  13 Delfi 5.1           2655   26   26   450   43%  2704   27% 
  14 Pharaon 3.5.1       2618   28   28   400   39%  2697   24% 
  15 Colossus 2007b      2571   28   29   400   34%  2688   25%

Code: Select all

   9 Alaric Deep IIb     2703 600.0 (289.0 : 311.0)
                               50.0 ( 15.5 :  34.5) Strelka 1.8         2843
                               50.0 ( 16.0 :  34.0) Rybka 1.0 Beta      2836
                               50.0 ( 15.5 :  34.5) Gambit Fruit 4bx    2778
                               50.0 ( 25.0 :  25.0) Spike 1.2           2762
                               50.0 ( 17.5 :  32.5) Chess Tiger 2007.1  2756
                               50.0 ( 26.5 :  23.5) Naum 2.0            2716
                               50.0 ( 24.5 :  25.5) Glaurung 2 - eps/4  2709
                               50.0 ( 25.5 :  24.5) CM9 Tribute         2666
                               50.0 ( 28.0 :  22.0) CMX Descent         2660
                               50.0 ( 30.5 :  19.5) Delfi 5.1           2655
                               50.0 ( 31.5 :  18.5) Pharaon 3.5.1       2618
                               50.0 ( 33.0 :  17.0) Colossus 2007b      2571

Thanks very much for testing the Deep IIb version Marc. If I see it correctly it finished 17 elo behind the default Alaric 707. It is a pity that Deep IIb lost, but on the other hand the difference is not big! The confidence in both results can be greater now in reality because a) the ratings are very close and b) the two versions do not differ much in playing style etc. so for the big picture it can be counted as one 1200 game match too, that's quite a match!

It is possible that the Deep IIb would do a little better at longer timecontrols, it did not lose by that big a margin compared to the default so it is possible, but I would not want to make a prediction about that. I would have to play more games first. Actually I was experimenting with another version too, that would probably be Deep III.

The settings as they are possible now do not change much about the style of play, that is, probably, also one of the reasons that the Alaric default and Deep IIb achieved very similar results in your matches against most of the opponents, and Deep IIb usually just a point or point and a half less. Maybe Peter will offer more options in future versions! Then we can experiment a bit with different playing styles, and not just with the pruning parameters.

I hope Alaric Deep III will play a little better against Shredder at 6 minutes + 6 seconds but it is just too early to say.

I hope you saw not a lot of crashes Marc in your matches? It can still happen I think.

I do plan on playing some more matches with Deep III, I will post later. Maybe a match against a Fruit Beta! But Fruit Beta is a bit stronger I think... I have got several settings for Fruit 051103 too, don't know if they are any good.. Suddenly a lot of new programs to try! And the bugfixed Togas are on their way too!

Regards, Eelco

[Marc MP]

Thanks very much for testing the Deep IIb version Marc. If I see it correctly it finished 17 elo behind the default Alaric 707. It is a pity that Deep IIb lost, but on the other hand the difference is not big! The confidence in both results can be greater now in reality because a) the ratings are very close and b) the two versions do not differ much in playing style etc. so for the big picture it can be counted as one 1200 game match too, that's quite a match!

It is possible that the Deep IIb would do a little better at longer timecontrols, it did not lose by that big a margin compared to the default so it is possible, but I would not want to make a prediction about that. I would have to play more games first. Actually I was experimenting with another version too, that would probably be Deep III.

The settings as they are possible now do not change much about the style of play, that is, probably, also one of the reasons that the Alaric default and Deep IIb achieved very similar results in your matches against most of the opponents, and Deep IIb usually just a point or point and a half less. Maybe Peter will offer more options in future versions! Then we can experiment a bit with different playing styles, and not just with the pruning parameters.

I hope Alaric Deep III will play a little better against Shredder at 6 minutes + 6 seconds but it is just too early to say.

I hope you saw not a lot of crashes Marc in your matches? It can still happen I think.

I do plan on playing some more matches with Deep III, I will post later. Maybe a match against a Fruit Beta! But Fruit Beta is a bit stronger I think... I have got several settings for Fruit 051103 too, don't know if they are any good.. Suddenly a lot of new programs to try! And the bugfixed Togas are on their way too!

Regards, Eelco

Given that the difference is rather small and that Deep IIb seems a bit better at tactics as you shown, it remains an interesting set up.

I had exactly no crash in the 1800 games played by the three Alaric versions here. Maybe that is because I use reliable software like winboard and polyglot?

I didn't follow the games too much, but I noticed once that Alaric Deep IIb lost a drawn endgame (both engine initially gave a draw score) KB vs KR against Gambit Fruit (which had bitbases on). Maybe it is due to my super-fast time control, but you may want to check it out for this kind of simple endgames.

Good luck for Deep III ,

Marc