Hi Mr Muller, any news for the 2016 Chinese Chess Olympiad ?

[hgm]

Actually a 4th program joined: Shiga. The past month I did a lot of work on my Xiangqi engine; it was in fact the first time in many years that I touched it. Last year I finished 4th/5th place tied with XieXie, and I knew for sure Pascal Tang would have been working on his engine. So I implemented speculative pondering (instead of thinking for the opponent), simple shared-hash-table SMP, an on-the-fly calculated (but hashed) material table (instead of a small pre-calculated table indexed by attacking material of one side and defensive material of the other), some end-game heuristics (like KHKP = draw), and some simple (partial) end-game tables (KRKd, KHHKd, KHPKd, KPPKd, KCPEKd, where d = any combination of A and E).

Unfortunately this was not enough, as Pascal had improved XieXie by over 400 Elo. So I lost all games. The final ranking after a doubleround-robin was:

Code: Select all

1. Shiga    6 pt
2. XieXie   3.5 pt
3. Chimo    2.5 pt
4. HaQiKi D 0 pt

Shiga is a monster, which was running on 12 cores, and has won all its games in the Olympiads where it participated since I first joined Xiangqi in 2009.

I could have had one easy draw against Chimo, and possibly a draw against XieXie if I had filled the material table more sensibly. But it was still containing the score corrections based on the difference of the entries in the small tables, or special codes that would invoke a small routine to evaluate the position based on the placement of that material (e.g. in KRCKR, to see whether the defending King occupies the central file, or in endings with Pawns to test the advance of the Pawn).

As it was, it chose in KRPEEKRAA to capture 2E rather than P against Chimo, and somewhat less clearly capture 2A instead of P against XieXie in KRPPAAKEAA. The material table in principle should allow me to cancel the piece value of the E or A in such situations to make them effectively worthless, and highly upgrade the effective value of the Pawn that makes the difference between overwhelming the defense or not.

Not that any of this would have changed the final ranking, but it just feels bad to have an engine that is so stupid that it misses things that even I (as someone who played perhaps 2 Xiangqi games in my entire life) see instantly...

[edwardyu]

Thanks HG! I don't know if you were aware that a Computer Xiangqi tournament has taken place in China in May 2016. All the strong Chinese Xiangqi engines have participated and 名手(CHESS NADE) has beaten 旋風(CYCLONE) 4-2 in the final to take the champion. Here is the sina news link :-
http://sports.sina.com.cn/go/2016-05-11 ... 1146.shtml

[hgm]

No, I did not know that. Thank you for the info. Xiangqi has been off my radar for a very long time; the past years I was completely absorbed by Shogi and its many variants. This year there were no opponents for Shogi at the Olympiad, so it was cancelled, and for mini-Shogi there were no opponents that I had to worry about. That gave me some time to spend on Xiangqi.

I enjoyed building the tablebases. Especially KCPEKd is a pretty interesting end-game. The implementation of these is not such that I could distribute them in the current form, however. (I did not pay any attention to size whatsoever, and because the machine I was using at the Olympiad had plenty of DRAM I just loaded them all in memory before the start of the game.) So I guess I will have to do some more work this month before I release a new version of HaQiKi D. I guess I will just package it with a generator for the EGTs.

[Evert]

Shame I couldn't be there with SjaakII this year, although I dont't think there would have been any danger on it missing out on the bottom spot.

Maybe next year.