hgm wrote:Well, so claiming that it has "XBoard protocol v2 compatibility" seems an over-statement. It seems most of XBoard protocol would not work, (i.e. is ignored), and you have to use command-line options in stead.
No worries, all the essential commands (timecontrol settings, switching opening book, pondering, xboard mode, analyze mode, position setup, ...) are supported according to the CECP, only some of them can be switched via commandline as well.
Randomize moves and nps are exceptions here, random play implementation is different from that of gnuchess and nps feature i noticed too late in the protocol definition. The advantage of this is that Phalanx can support nps limit in any GUI and also can have both nps limit and pondering on. Also, both the nps limit and randomizing is used internally in the easy levels.
It also seems to ignore the hash-size setting, which I would consider also pretty important.
And even if your randomixation is not just an on/off switch, but needs a value, it would be much more user friendly if you provided an option for it so that it could be controlled interactively from the Engine Settings dialog.
I am not critizising the fact that you provide command-line options as a work-around for obsolete GUIs to activate the features; it is good to have those. But, as some of the options (such as ponder) already show, one doesn't need to exclude the other. I really think it would improve the user experience if hash size, random magnitude, book directories, learning files and such could be set from the GUI.
hgm wrote:It also seems to ignore the hash-size setting, which I would consider also pretty important.
And even if your randomixation is not just an on/off switch, but needs a value, it would be much more user friendly if you provided an option for it so that it could be controlled interactively from the Engine Settings dialog.
I am not critizising the fact that you provide command-line options as a work-around for obsolete GUIs to activate the features; it is good to have those. But, as some of the options (such as ponder) already show, one doesn't need to exclude the other. I really think it would improve the user experience if hash size, random magnitude, book directories, learning files and such could be set from the GUI.
Thanks for checking the engine and for the hints. The 'memory' command support should be fairly simple, this will be part of the XXIV release. The other features i will think about, there are more ideas related to easy levels and the root moves evaluation feature (now used for randomizing only).
Werner wrote:I wonder,
what is the difference to Phalanx XXIII JA from 2012 ?
regards
Werner
I was not involved in 2012, but as far as i can guess based on the sourceforge repository history, the Jim Ablett's 2012 version is just XXII tweaked for a good and fast native windows build, plus there's amended FEN parsing, but the engine inside is identical to XXII. Should have been called 'XXIII beta'.
Phalanx was always one of my favorite programs, and to a large part I learned chess programming by studying your source code. Even today, you can see a clear influence from Phalanx in Stockfish's king safety evaluation.
If you are interested, I can help you bring your program to iOS, using Stockfish's iOS GUI. All I would need from your side is UCI protocol support in the engine.
Tord Romstad wrote:Thanks Dusan, this is awesome news!
Phalanx was always one of my favorite programs, and to a large part I learned chess programming by studying your source code. Even today, you can see a clear influence from Phalanx in Stockfish's king safety evaluation.
If you are interested, I can help you bring your program to iOS, using Stockfish's iOS GUI. All I would need from your side is UCI protocol support in the engine.
Hi Tord, thanks a lot, i'm glad to read this. Phalanx had the reputation of somewhat hard to read code, but maybe people just got confused with all the magic formulas in the fractional ply extension routines. Other than that i still believe the basic data structures are easy to grasp. You had even read important parts of the large static evaluation function and found a bug in the king safety years ago, now this is finally fixed in XXIII, thanks again.
I haven't checked the UCI protocol yet, at this time i'm finally almost ok with CECP2 (the current CECP version is a very good protocol, too) and would like to play with reductions and prunings first, so maybe then.
BTW Stockfish is an amazing thing. Not only it's a good piece of chess software, it's also a well organized opensource project, using the advantages of open source approach to maximum. Often I use it for post game analysis of my OTB games, some months ago it surpised me with a positional sac of a piece for 2 pawns + initiative when i was analysing a game where i was not brave enough to play that sac. A nice surprise indeed.
Now that's a hypermodern opening style, Phalanx ignores the dogmata and plays 7. Kd2!?
Arnaud, how come AnMon 5.75 is rated below Phalanx in your list, I believe it should be still stronger by 100 elo points at least. You might want to check AnMon setup, does it get the correct amount of hash table? Or is it just it cannot handle openings that well without the opening book?