Twisted Logic 0.065e35 and 0.065e37 released!

[Edsel Apostol]

Hi everyone,

This versions doesn't support winboard anymore. It now supports UCI. Pondering and Analyse in infinite mode now is possible. It is still slow in NPS, I have not rewritten the data structures yet since I learned and started programming a year ago. I tried to make up with this by introducing very aggresive pruning methods.

I still don't have my own PC so I could not test all of my ideas and I could not run a comphrehensive testing against other programs but I think it is stronger than the previous versions. I only do my programming on internet cafe's, you know, I'm just a poor student from the Philippines. One of the reasons why I release two versions because I don't have much time testing. I'm paying every minute of computer time out of my allowance in school.

Anyway, I will try to make this program very strong. I will try to support SMP and 64 bit compiles. I just have a few questions? Can I compile an SMP version or a 64 bit version on a 32 bit machine? 32 bit machines are the only computers that I could find here.

My web page is on the link below.

Edsel

[Graham Banks]

Thanks Edsel.

Twisted Logic 0.065e37 will replace Twisted Logic 0.057e3 in my tournement.

Regards, Graham.

[Guenther Simon]

Hi Edsel,

Did you fix the 40th move playing as Black bug for time controls
40 moves/x in this version? I noticed that bug was still there
in the last version.

Regards,
Guenther

[Tony Thomas]

I do have one little problem with this version of Twisted, it doesnt clear its hashtables even after I turn on the clear hash option. The hash fills up after 24-30 moves and then it plays weird moves. The total memory usage of Twisted is 37MB, I guess it is probably using about 16Mb of that for hash.

[Edsel Apostol]

Hi Guenther,

I think it was fixed now. It was a bug in my winboard implementation. I am using UCI now.

Edsel

[Edsel Apostol]

Hi Tony,

I will try to review my source code for the problem of that Clear Hash. Maybe I misinterpreted how to implement it. I will read again the UCI protocol to fix it.

My UCI implementation needs a "ucinewgame" command every new game for it to clear and initialize its hash tables. Don't worry about the hash being full, I am displaying the percentage of the total number of entries being used over the total number of entries. It will usually fill up after a few moves. Even if it is already 100% full, the program will just overwrite the oldest entries. I will try to change the display of that "hashfull" display command for the next versions to avoid confusion.

If the total memory usage is 37Mb, it means that you have set the hash table size to 32Mb - 63Mb. If the size is below 64Mb and >= 32Mb, it will set it to 32Mb. Im using "power of two" for the hash size. It is using 2Mb for the pawn hash table and 2Mb for the static eval cache. The remaining are used for the program variables.

About it playing weird moves, maybe that was just it's playing style. I often don't understand the logic of its moves, and that's what its name implies "Twisted Logic".

Edsel

[Tony Thomas]

Thanks for helping me out Ed. When I said weird moves, I meant that Twisted's eval started droping slowly few moves after its hashtable reached 99%. It was just few games, most of the games ended up being draws. I have similar problems with Little Goliath and Capture, but both of those programs tend to exit once the hash is full. Since you said Twisted rewrites the hash entries, I guess I would go home and enter it in my 2nd division bullet tournament, it is currently in clear lead in 3rd division using the previous version. Also, it would be nice if you support the winboard protocol like you used to, and also if you included a configuration file.