Dann Corbit wrote:Change macro:
#define HSIZE 0x800000
to be:
#define HSIZE (0x800000>>1)
and you will get 64 MB.
Change macro:
#define HSIZE 0x800000
to be:
#define HSIZE (0x800000>>2)
and you will get 32 MB.
And then recompile
You have to be careful for this, because in the hash table the stored data contains the upper bits of the complete hash key and in the lower bits there is stored either the value for lower/upper bound (X)OR a complete move (21bits). so, if you make the HMASK too small, there is no place for the 21bit move.
Unfortunately it is not that easy. you have to handle two different HMASK for Hashtables smaller than 21bit. I am working to allow this.
Last edited by OliverBr on Sat Jan 26, 2008 5:41 am, edited 2 times in total.
Engine Score Cr Ol He Ai GN Ol S-B
1: Crafty 26,5/30 ······ 1110=1 111111 111011 11=1=1 111111 331,50
2: OliThink508 20,0/30 0001=0 ······ 0=1111 1=0=11 011111 111101 236,25
3: Hermann 1.5 17,0/30 000000 1=0000 ······ 011111 011111 11=111 166,00
4: Aice 0.92 12,5/30 000100 0=1=00 100000 ······ 11111= 001101 143,00
5: GNUChes5 7,0/30 00=0=0 100000 100000 00000= ······ 101=10 94,25
6: Olithink411 7,0/30 000000 000010 00=000 110010 010=01 ······ 83,50
90 games played / Tournament finished
Name of the tournament: Arena tournament11
Site/ Country: TARN-PC, Germany
Level: Blitz 3/0
Hardware: with 1.534 MB Memory
Operating system: Microsoft Windows NT Professional (Build 6000)
Due to the -just mobility- evaluation OliThink 5.0.8 plays very risky and very aggressive. The won games are sometimes stunning.
Can anyone calculate the correct ELO for this performance?
You could get an elo estimate by using Elostat that is available with your arena. Olithink does seem very strong. I wish you would release a version that uses less memory.
Edsel Apostol wrote:If you add UCI, SEE and quiescent checks move generation, then Olithink would really have a great potential.
UCI is that important? I though xboard protocol is fine...
I had some tests with SEE and I am not happy. The improved move ordering only bring some percent.
Yes. Quiescent Checks, this is one I was planning for long time. Didn't know what it can bring as I think crafty is not using it...
Engine Score Cr Ol He Ai GN Ol S-B
1: Crafty 26,5/30 ······ 1110=1 111111 111011 11=1=1 111111 331,50
2: OliThink508 20,0/30 0001=0 ······ 0=1111 1=0=11 011111 111101 236,25
3: Hermann 1.5 17,0/30 000000 1=0000 ······ 011111 011111 11=111 166,00
4: Aice 0.92 12,5/30 000100 0=1=00 100000 ······ 11111= 001101 143,00
5: GNUChes5 7,0/30 00=0=0 100000 100000 00000= ······ 101=10 94,25
6: Olithink411 7,0/30 000000 000010 00=000 110010 010=01 ······ 83,50
90 games played / Tournament finished
Name of the tournament: Arena tournament11
Site/ Country: TARN-PC, Germany
Level: Blitz 3/0
Hardware: with 1.534 MB Memory
Operating system: Microsoft Windows NT Professional (Build 6000)
Due to the -just mobility- evaluation OliThink 5.0.8 plays very risky and very aggressive. The won games are sometimes stunning.
Can anyone calculate the correct ELO for this performance?
You could get an elo estimate by using Elostat that is available with your arena. Olithink does seem very strong. I wish you would release a version that uses less memory.
Now I am running a 5/0 tournament and OliThink has much less success (( Can there be such a difference between 3/0 and 5/0.
Next release I am planning to have a switch for adjusting Hashsize
UCI is that important? I though xboard protocol is fine...
Yes, UCI is important. It is easier to configure for people like me.
The Analyze Mode is important too. I rarely play engine matches. If OliThink is "something different" then I would certainly look forward to using it in analysis.
UCI is that important? I though xboard protocol is fine...
Yes, UCI is important. It is easier to configure for people like me.
The Analyze Mode is important too. I rarely play engine matches. If OliThink is "something different" then I would certainly look forward to using it in analysis.
I wanted to hold the code small and neat. UCI will blow it up a little bit... but I see that people want it.
Yes, it plays completely different than other engines. This is the most significant reason to use OliThink. the 300+ other free engines could mostly been some weaker crafty clones. Hardly any difference.
UCI is that important? I though xboard protocol is fine...
Yes, UCI is important. It is easier to configure for people like me.
The Analyze Mode is important too. I rarely play engine matches. If OliThink is "something different" then I would certainly look forward to using it in analysis.
I wanted to hold the code small and neat. UCI will blow it up a little bit... but I see that people want it.
Yes, it plays completely different than other engines. This is the most significant reason to use OliThink. the 300+ other free engines could mostly been some weaker crafty clones. Hardly any difference.
I disagree.
I do not think that many programmers look at Crafty today when there are toga or strelka or glaurung.
I also do not believe that most programs are clones.
(( Can there be such a difference between 3/0 and 5/0.