Thursday, April 12, 2012

Evaluation Output

The latest changes to iCE did not gain any ELO or made things worse so I threw them away. To see a bit of progress I decided to spend some time into something that does not affect playing strength so I can keep it for sure. I had a reformatting of the evaluation output on my todo list for some time already and I now decided to get it done. The eval output of the stockfish engine looks really nice and helpful so I decided to mirror that format just using the evaluation terms of iCE instead of the stockfish ones of course.

So it now looks like this and I hope it will me help to fine tune eval in the later development of iCE.


position fen r2q1rk1/pp2np2/5P2/3bp2Q/3pN2b/1B1P4/PPP3PP/R4RK1 w
Total eval of the position : -52
Evaluation:
===========

            Eval term |    White    |    Black    |     Total
                      |   MG    EG  |   MG    EG  |   MG     EG
 ---------------------+-------------+-------------+--------------
      Static Material |   ---   --- |   ---   --- |   141    141
     Dynamic Material |    36    36 |     2     2 |    38     38
             Mobility |   -73   -73 |    66    66 |    -7     -7
   Pattern Recognizer |     0     0 |     0     0 |     0      0
        King Pressure |   -20   -20 |     1     1 |   -19    -19
   Static Pawn Struct |   ---   --- |   ---   --- |     6      0
 Dynamic. Pawn Struct |   -33   -38 |    -4    -4 |   -37    -42
               Pieces |   -10   -10 |    -8     0 |   -18    -10
              Threats |     0     0 |   -52   -87 |   -52    -87
 ---------------------+-------------+-------------+--------------
                Total |   ---   --- |   ---   --- |    52     14

Scaling : 100% MG (52)  0% EG (0)  = 52
NegaMax Adjustment for Side To Move (White): -52


In order to be able to gather the sub terms of eval I had to rewrite parts of it and a minor functionality change related to the sub scores rounding (which is not done anymore) was unavoidable. But tests show it does not seem to hurt. In theory it should even help.

But it now shows also a weakness of iCE. In this position Black is in big trouble but static eval does not indicated that. The king pressure evaluation definitely needs some adjustment. iCE finds its way through anyway once it reaches that position (... but it might never get so far)

info depth 11 seldepth 26 time 2734 nodes 7865064 pv h5h6 e7f5 f1f5 h4f6 e4f6 d8f6 h6f6 f8d8 f5h5 d5b3 h5h8  nps 2876760 score mate 6 hashfull 133 tbhits 0
bestmove h5h6 ponder e7f5