New Toga version: +200 elo??

[Marc MP]

Check this: http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=5240

[Eelco de Groot]

Check this: http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=5240

Hi Marc,

A bit too early I think for such a big jump. Nice name though, Mara de bois is French for strawberries I believe? No 200 or more elojumps but the Toga team is making some progress with new Toga versions that you do not hear about much in the general fora for the unwashed masses.

One of the latest to be found on the Toga Developers Forum is a learning Toga! It is by Bill Mowery I believe. Not sure if anyone can register in all the fora there, that is up to Shaun. http://www.computerchess.info/tdbb/phpBB3/

I don't know any details of the latest versions, and Betas will probably not be released but if you are interested in beta-testing you could ask about this in the general forum there maybe, after registering.


Regards, Eelco

[gerold]

Check this: http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=5240

Right on Toga.
Seeing is believing.

I would be happy to test it as well as anyone else if it is 200 elo
improved.
That is a great improvement for a clone.

[Marc MP]

Check this: http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=5240

Hi Marc,

A bit too early I think for such a big jump. Nice name though, Mara de bois is French for strawberries I believe? No 200 or more elojumps but the Toga team is making some progress with new Toga versions that you do not hear about much in the general fora for the unwashed masses.

One of the latest to be found on the Toga Developers Forum is a learning Toga! It is by Bill Mowery I believe. Not sure if anyone can register in all the fora there, that is up to Shaun. http://www.computerchess.info/tdbb/phpBB3/

I don't know any details of the latest versions, and Betas will probably not be released but if you are interested in beta-testing you could ask about this in the general forum there maybe, after registering.


Regards, Eelco

Thank you very much for the link Eelco, I didn't know about and I don't think I would have found it otherwise. Good luck for Ancalagon!

[Marc MP]

... Nice name though, Mara de bois is French for strawberries I believe?

...


Regards, Eelco

I speak french but I'm not French so maybe I just never heard of that idiom.

[Eelco de Groot]

... Nice name though, Mara de bois is French for strawberries I believe?

...


Regards, Eelco

I speak french but I'm not French so maybe I just never heard of that idiom.

Well I just put it in Google and then I saw some pictures of strawberries, but my high school Wolters Noordhoff French dictionary, twelfth edition, 1972 has never heard of "mara de bois" Maybe it's a bit out of date my dictionary, and maybe there is some African influence in the use? Swahili? I don't know.

Regards, Eelco

[Eelco de Groot]

Check this: http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=5240

Hi Marc,

A bit too early I think for such a big jump. Nice name though, Mara de bois is French for strawberries I believe? No 200 or more elojumps but the Toga team is making some progress with new Toga versions that you do not hear about much in the general fora for the unwashed masses.

One of the latest to be found on the Toga Developers Forum is a learning Toga! It is by Bill Mowery I believe. Not sure if anyone can register in all the fora there, that is up to Shaun. http://www.computerchess.info/tdbb/phpBB3/

I don't know any details of the latest versions, and Betas will probably not be released but if you are interested in beta-testing you could ask about this in the general forum there maybe, after registering.


Regards, Eelco

Thank you very much for the link Eelco, I didn't know about and I don't think I would have found it otherwise. Good luck for Ancalagon!

Thanks! I'm bughunting at the moment, Jon Dart's SSDF test-position brought some strange behaviour to light in the following position Ancalagon plays 4. Nh1?? and then refuses to take the Knight for as yet mysterious reasons playing moves like 4... Qc2?! first .

[d]b7/P5p1/R4pk1/4p3/1Bp1P3/2P2P1P/4KN2/1q6 w - -

So now I have gone back from Ancalagon version 108 crossed with Glaurung Mjolnir to Ancalagon Build 65 that does not show this behaviour, at least in the following position things look more or less okay, I named this version Ancalagon 0.65:


[d]b7/p5p1/R1p2pk1/2P1p3/1Bp1P3/2P2P1P/4K3/1q5N b - -

Engine: Ancalagon 0.65 (64 MB)
by Tord Romstad, Eelco de Groot

2.00 0:00 +6.11 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 (235) 3

3.00 0:00 +6.11 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 (1.031) 13

4.00 0:00 +6.13 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qg1+
4.Ke2 (3.404) 43

5.00 0:00 +5.86 1...Qxh1 2.Rxa7 Qg2+ 3.Ke1 Qg1+
4.Ke2 Qh2+ 5.Kd1 (9.682) 103

6.00 0:00 +5.68 1...Qxh1 2.Rxa7 Qg2+ 3.Ke1 Qg1+
4.Ke2 Qh2+ 5.Ke1 Qg3+ 6.Ke2 (23.552) 188

7.00 0:00 +5.94 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qf2+
5.Kc1 Qe1+ 6.Kb2 (57.086) 303

8.00 0:00 +5.76 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qf2+
5.Kc1 Qe1+ 6.Kb2 Qd2+ 7.Ka3 (93.941) 375

9.00 0:00 +6.19 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Ra2 Qg1+
5.Rf2 Qg5+ 6.Ke2 Qc1 7.Rg2+ Kf7 (159.454) 444

10.00 0:00 +6.23 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qf2+
5.Kc1 Qxf3 6.Rxa8 Qxh3 7.Rd8 Qe3+
8.Kd1 Qxe4 (247.166) 494

11.00 0:00 +6.23 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qf2+
5.Kc1 Qxf3 6.Rxa8 Qxh3 7.Rd8 Qe3+
8.Kd1 Qxe4 (409.071) 523

12.01 0:01 +6.23 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qf2+
5.Kc1 Qxf3 6.Rxa8 Qxh3 7.Rd8 Qg2
8.Kd1 Qxe4 (690.692) 545

13.01 0:02 +6.29 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qf2+
5.Kc1 Qe3+ 6.Kc2 Qxf3 7.Rxa8 Qxe4+
8.Kd2 Qd3+ 9.Ke1 Qxh3 10.Kf2 Qf5+
11.Ke2 Qd3+ 12.Kf2 (1.500.086) 551

14.01 0:04 +6.45 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qxh3
5.Rxa8 Qxf3 6.Rd8 Qf4+ 7.Ke2 Qxe4+
8.Kf2 f5 9.Rd6+ Kg5 10.Rd7 g6 11.Rd2 (2.450.750) 568

15.01 0:08 +6.64 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qxh3
5.Rxa8 Qxf3 6.Rd8 Qf4+ 7.Kd1 Qxe4
8.Rd6 Qe3 9.Kc2 Kf5 10.Rd2 (4.646.143) 571

16.01 0:15 +6.72 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qxh3
5.Rxa8 Qxf3 6.Ra1 Qd3+ 7.Ke1 Qxe4+
8.Kf2 Qc2+ 9.Kf3 Qd3+ 10.Kf2 Qd2+
11.Kf3 Kf5 12.Kg3 Qe3+ 13.Kg2 (8.973.491) 583

17.01 0:28 +6.86 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qxh3
5.Rxa8 Qxf3 6.Rd8 Qxe4 7.Kd1 Kg5
8.Rd2 f5 9.Rd7 Qf3+ 10.Kc2 g6 11.Rd6 Qe2+
12.Rd2 (16.758.652) 583

18.01 0:50 +6.94 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Qf1 4.Kd2 Qxh3
5.Rxa8 Qxf3 6.Rd8 Qxe4 7.Rd6 Qf3
8.Ke1 Kf5 9.Rd2 Qg3+ 10.Ke2 Kf4
11.Kd1 g5 (29.438.553) 578

19.01 3:02 +8.62 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Kh5 4.Rd7 Kh4
5.Rd2 Qxh3 6.Re2 Kg3 7.Kd2 Qf1 8.Re3 Kf4
9.Ba3 Qf2+ 10.Re2 Qxf3 11.Bb4 g5
12.Ke1 (83.918.780) 459

20.01 4:10 +8.88 1...Qxh1 2.Rxa7 Qg2+ 3.Ke3 Kh5 4.Rd7 Kh4
5.Rd2 Qf1 6.Re2 Kxh3 7.Kd2 Kg3 8.Re1 Qd3+
9.Kc1 Kxf3 10.Ba5 Kf2 11.Rd1 Qe3+
12.Kb2 Qe2+ 13.Kc1 Qxe4 (115.799.142) 462

best move: Qb1xh1 time: 5:05.156 min n/s: 463.408 nodes: 141.390.000

The search is from Build 108 so that is a relief, that the new search works still in the code of early Ancalagon 0.65. I have no idea where the bug could be hiding then at the moment, if it's not in search.cpp, I narrowed it down to the eval.cpp hopefully, that should be a little less hard to debug

Eelco

[gerold]

... Nice name though, Mara de bois is French for strawberries I believe?

...


Regards, Eelco

I speak french but I'm not French so maybe I just never heard of that idiom.

Well I just put it in Google and then I saw some pictures of strawberries, but my high school Wolters Noordhoff French dictionary, twelfth edition, 1972 has never heard of "mara de bois" Maybe it's a bit out of date my dictionary, and maybe there is some African influence in the use? Swahili? I don't know.

Regards, Eelco

Hello Eelco. Mara des bois was developed in France in the 1990s.
To bring back the taste of wild strawberrys.I have wild strawberrys
here. They are small and really has a sweet and better taste than
the hybrids in stores.

Best to you,

Gerold.

[ernest]

Check this:

So far, it looks like another crackpot...

[kranium]

here's the original post:

--------------
Hi All,

I've been having this toga for some months now, I could not release it for obvious reasons. Now that there could be a worthy competitor, I feel the gap needs to be bridged, but I'll make sure it is not as strong as Rybka 3 . The +200 is when compared with other strong toga versions. Against 64 and 32 bit versions of Rybka 2.3.2a and Naum 3.1, the gain is far much greater I no longer test against them (to contain my excitement). Surprisingly Naum 3.0 could hold off better than 3.1. I don't use opening books when running the tournaments.

80% of the gain is from an algorithm I developed for awarding....., 20% from additional evaluation terms. I've barely touched the search. I lost major elos (it's killing me) trying to avoid draws, now Toga Mara loses some games to other togas resulting to only over +200 elo gain. I'm yet to recall the initial code. I'll be releasing the pgns soon.

Regards,
Tam.
--------------
these claims are encouraging, albeit fantastic, but i'm skeptical.

if Tam 'lost major elos' while implementing code in an attempt to avoid draws, why doesn't he simply revert to the previous version?
he said 'im yet to recall the initial code', does this mean he's not using some sort of change control or version management software (as any experienced developer would...)?

what is 'awarding'?

he also said- 'I could not release it for obvious reasons'
why not?

i'm puzzled.

regards-
Norm