RomiChess vs. Oli's 5.29 & 5.30

[Michael Sherwin]

Aren't Gaviota and RomiChess supposed to be much stronger?

OliThink still has the feature, that it's evalation is only mobility. There isn't any further stratetigal information!
If you consider the HUGE eval function of crafty 20.14, I am very happy that it is very close to it on fast blitz games already.

But, on the other site, whatever change on search, hashtables, eval immediately let it drop in strength. It's quite an unstable system...

Older RomiChess versions are not able to score 46% against Crafty 19.19 let alone the stronger 20.14 version. Despite the Crafty 20.14 large eval it still gets as deep or deeper than Olithink. I think it does anyway. Bottom line for Olithink 5.30 is that it has to be a stable and strong engine to score 46% against Crafty 20.14 over a thousand game match. I see no way around that conclusion. Unless you are handicapping Crafty in some way.

Also, I've liked the purity of your programming model from the beginning of Olithink 5.xx. I think that there is a lot of 'chess truth' in rightly discerned mobility factors. After all chess is about moving pieces. And search is about maximizing the benefit of those rightly discerned mobility factors. So I am not surprised that your engine is getting stronger.

The reason that I am using Olithink as a test opponent is simply because, ever since version 5.06 came out Oli has been a supreme nemeses to Romi, doing much better than it should. Testing against nemeses engines has always led to a stronger Romi in a shorter amount of time.

[OliverBr]

Bottom line for Olithink 5.30 is that it has to be a stable and strong engine to score 46% against Crafty 20.14 over a thousand game match. I see no way around that conclusion. Unless you are handicapping Crafty in some way.

No, there is no handicap for Crafty except for the fact that I only have a 32bit version, while OliThink is playing 64bit.


Furthermore I know that on much slower time controls Crafty is performing better mostly because of the complex evaluation.

Note: Crafty 20.14 (I don't know about later versions) starts playing really crappy when there is very little time left. It seems that the search depth is dramastically reduced in order no to forfeit on time.



You don't make/have linux executables for Romichess? Or better some source code that is compilable?

Also, I've liked the purity of your programming model from the beginning of Olithink 5.xx. I think that there is a lot of 'chess truth' in rightly discerned mobility factors. After all chess is about moving pieces. And search is about maximizing the benefit of those rightly discerned mobility factors. So I am not surprised that your engine is getting stronger.

There is still a lot of stuff to improve. I am playing with the idea using the information of mobility for attacking bitsquares which could easily used for other purposes, like checks.

I think all strong engines do this, I have seen something like this in the robbolito (rybka clone???) code, even though they can't spell "attack" (attak_white), lol.

Furthermore I don't have any tree razoring or futility pruning other than at d=0. Every attempt to implement it, failed.

More use (and a more sophisticated use) of hash table is another thing, that could improve olithink more.

My challenge is to create a mobility-only engine as strong as possible.

PS: Sometimes I watch OliThink playing on ICC against some multi-Processor Rybkas. That is not even funny when after 25 moves the game is already lost

[michiguel]

Aren't Gaviota and RomiChess supposed to be much stronger?

Gaviota's search is extremely primitive. It gets outsearched in such a manner that is not even funny. I never had a chance to focus on it. It only has nullmove reduction. That is it. I started work on it now but the usual suspects are not giving me much (yet). LMR is not giving me much at all, but I am still experimenting.

Also, Gaviota's eval is heavily based on mobility since the beginning (I added PST recently but I have them initialized to zero now, IIRC). I added other things, but many times it made gaviota worse. I started to improve it last year when I cleaned up the code and I removed stuff.

If you engine is based only on mobility I need to congratulate you. It plays really strong. Both Romichess and Olithink (5.18 and 5.29) are in my tests. They are really close to Gaviota.

Miguel

OliThink still has the feature, that it's evalation is only mobility. There isn't any further stratetigal information!
If you consider the HUGE eval function of crafty 20.14, I am very happy that it is very close to it on fast blitz games already.

But, on the other site, whatever change on search, hashtables, eval immediately let it drop in strength. It's quite an unstable system...

[michiguel]

Bottom line for Olithink 5.30 is that it has to be a stable and strong engine to score 46% against Crafty 20.14 over a thousand game match. I see no way around that conclusion. Unless you are handicapping Crafty in some way.

No, there is no handicap for Crafty except for the fact that I only have a 32bit version, while OliThink is playing 64bit.


Furthermore I know that on much slower time controls Crafty is performing better mostly because of the complex evaluation.

Note: Crafty 20.14 (I don't know about later versions) starts playing really crappy when there is very little time left. It seems that the search depth is dramastically reduced in order no to forfeit on time.



You don't make/have linux executables for Romichess? Or better some source code that is compilable?

[]
I am testing in Linux, so I use Romi with Wine and it works really well.

Miguel
[]

Also, I've liked the purity of your programming model from the beginning of Olithink 5.xx. I think that there is a lot of 'chess truth' in rightly discerned mobility factors. After all chess is about moving pieces. And search is about maximizing the benefit of those rightly discerned mobility factors. So I am not surprised that your engine is getting stronger.

There is still a lot of stuff to improve. I am playing with the idea using the information of mobility for attacking bitsquares which could easily used for other purposes, like checks.

I think all strong engines do this, I have seen something like this in the robbolito (rybka clone???) code, even though they can't spell "attack" (attak_white), lol.

Furthermore I don't have any tree razoring or futility pruning other than at d=0. Every attempt to implement it, failed.

More use (and a more sophisticated use) of hash table is another thing, that could improve olithink more.

My challenge is to create a mobility-only engine as strong as possible.

PS: Sometimes I watch OliThink playing on ICC against some multi-Processor Rybkas. That is not even funny when after 25 moves the game is already lost

[John_F]

Oliver, I am curious -- are you planning to add knowledge (other than mobility) to your engine in the future? I think an engine with no knowledge other than mobility sounds like it might have an intriguing style of play!

It would be interesting to see how strong you can make it without any other strategic knowledge.

[OliverBr]

Just for info.. The new candiate for OliThink 5.3.1 did this after 26 games:

Code: Select all

olithink - crafty : 21.0/28 20-2-6 75% (1111011001110=11=10110111111)

That looks amazing, but future will show, that 28 games are way too little... I bet xx EURO that it won't reach the 50% after 1000 games either.

[OliverBr]

Oliver, I am curious -- are you planning to add knowledge (other than mobility) to your engine in the future? I think an engine with no knowledge other than mobility sounds like it might have an intriguing style of play!

It would be interesting to see how strong you can make it without any other strategic knowledge.

That was my primary target. I never wanted to put any further knowledge to my engine and check out how strong it gets.

And so I am not planning to do. That's why it's building a lot of double pawns and kings without shelters etc...

[Michael Sherwin]

Just for info.. The new candiate for OliThink 5.3.1 did this after 26 games:

Code: Select all

olithink - crafty : 21.0/28 20-2-6 75% (1111011001110=11=10110111111)

That looks amazing, but future will show, that 28 games are way too little... I bet xx EURO that it won't reach the 50% after 1000 games either.

That is an interesting anomaly! If your right then that makes these next results rather meaningless.

Code: Select all

1: RomiChess96     40.0/65 
2: Gaviota64       35.5/65 
3: RomiChessX64P3k 29.0/65 
4: Olithink64530   25.5/65 

-----------------Gaviota64-----------------
Gaviota64 - Olithink64530       : 16.0/21 15-4-2   76%  +200
Gaviota64 - RomiChess96         : 8.5/22 5-10-7    39%   -78
Gaviota64 - RomiChessX64P3k     : 11.0/22 8-8-6    50%    ±0
-----------------Olithink64530-----------------
Olithink64530 - Gaviota64       : 5.0/21 4-15-2    24%  -200
Olithink64530 - RomiChess96     : 9.0/22 6-10-6    41%   -63
Olithink64530 - RomiChessX64P3k : 11.5/22 8-7-7    52%   +14
-----------------RomiChess96-----------------
RomiChess96 - Gaviota64         : 13.5/22 10-5-7   61%   +78
RomiChess96 - Olithink64530     : 13.0/22 10-6-6   59%   +63
RomiChess96 - RomiChessX64P3k   : 13.5/21 10-4-7   64%  +100
-----------------RomiChessX64P3k-----------------
RomiChessX64P3k - Gaviota64     : 11.0/22 8-8-6    50%    ±0
RomiChessX64P3k - Olithink64530 : 10.5/22 7-8-7    48%   -14
RomiChessX64P3k - RomiChess96   : 7.5/21 4-10-7    36%  -100

If your wrong then you have something very interesting going on!

[OliverBr]

If your wrong then you have something very interesting going on!

I am neither right nor wrong, I just wanted to mention that 30 games are way too little.

Now, after 119 games, it's about 47,5%... I knew it.

[Michael Sherwin]

Code: Select all

1: RomiChess96     87.5/150 
2: Gaviota64       79.0/150 
3: RomiChessX64P3k 71.5/150 
4: Olithink64530   62.0/150 

-----------------Gaviota64-----------------
Gaviota64 - Olithink64530       : 31.0/50 29-17-4    62%   +85
Gaviota64 - RomiChess96         : 21.5/50 14-21-15   43%   -49
Gaviota64 - RomiChessX64P3k     : 26.5/50 20-17-13   53%   +21
-----------------Olithink64530-----------------
Olithink64530 - Gaviota64       : 19.0/50 17-29-4    38%   -85
Olithink64530 - RomiChess96     : 20.5/50 14-23-13   41%   -63
Olithink64530 - RomiChessX64P3k : 22.5/50 17-22-11   45%   -35
-----------------RomiChess96-----------------
RomiChess96 - Gaviota64         : 28.5/50 21-14-15   57%   +49
RomiChess96 - Olithink64530     : 29.5/50 23-14-13   59%   +63
RomiChess96 - RomiChessX64P3k   : 29.5/50 20-11-19   59%   +63
-----------------RomiChessX64P3k-----------------
RomiChessX64P3k - Gaviota64     : 23.5/50 17-20-13   47%   -21
RomiChessX64P3k - Olithink64530 : 27.5/50 22-17-11   55%   +35
RomiChessX64P3k - RomiChess96   : 20.5/50 11-20-19   41%   -63