HAwkeye 1.03 Released

[THyer]

MikeB: I also have a Gull variant, which I hope could be the base for further development. I have ported the Gull 3 source to more standard C++, removing all goto statements except those in gen_kpk (which will be superseded by tablebases anyway).

Would you consider merging our efforts?

Tom Hyer

[basil00]

MikeB: I also have a Gull variant, which I hope could be the base for further development. I have ported the Gull 3 source to more standard C++, removing all goto statements except those in gen_kpk (which will be superseded by tablebases anyway).

Would you consider merging our efforts?

I'd suggest making your code publicly available first, e.g. as a github repo.

[Ozymandias]

MikeB: I also have a Gull variant, which I hope could be the base for further development. I have ported the Gull 3 source to more standard C++, removing all goto statements except those in gen_kpk (which will be superseded by tablebases anyway).

Would you consider merging our efforts?

I'd suggest making your code publicly available first, e.g. as a github repo.

Second that, I've read of even more people involved in Gull development. but everyone is waging war on their own. That's the big difference with SF.

[MikeB]

MikeB: I also have a Gull variant, which I hope could be the base for further development. I have ported the Gull 3 source to more standard C++, removing all goto statements except those in gen_kpk (which will be superseded by tablebases anyway).

Would you consider merging our efforts?

Tom Hyer

Sure - let me see what you have.

[THyer]

One thing that has been bothering me in the Gull code is the calls to see<me> with a hardcoded 50-cp margin.

Since two pieces minus a rook and pawn is exactly 50 cp, a class of exchanges is just barely ignored (I think).

Are you able and willing to test changing this to either 40 or 60 cp? I think 60 and 50 give the same strength, and 40 is stronger, but my significance levels aren't good.

[MikeB]

One thing that has been bothering me in the Gull code is the calls to see<me> with a hardcoded 50-cp margin.

Since two pieces minus a rook and pawn is exactly 50 cp, a class of exchanges is just barely ignored (I think).

Are you able and willing to test changing this to either 40 or 60 cp? I think 60 and 50 give the same strength, and 40 is stronger, but my significance levels aren't good.

Sure - send me me the code that needs to be changed.

[THyer]

In 5 places Gull source contains 'see<me>(move,-50)'. I recommend global-replacing that with 'see<me>(move,-SeeThreshold)' and then setting SeeThreshold=40.

[Dan Cooper]

I have a mystery for the bird devs to solve.

Gull doesn't use its assigned hash under ChessGUI (and IIRC, Arena). If I assign 2GB, it will report ~70MB or so under resource monitor. With cutechess, it uses the entire 2GB.

Vadim told Graham that nothing was wrong, but I definitely get better results for Gull under cutechess.

[THyer]

The code is now public; see http://talkchess.com/forum/viewtopic.php?t=61195

[Ozymandias]

I'm not sure if opening another project is the best course of action. Maybe if all these efforts went towards adding to basil's (the first one to retake Gull) we'd have something better how.