evalDevelopment function pb

iamLovingJerk / pychess

Automatically exported from code.google.com/p/pychess

GNU General Public License v3.0

0 stars 0 forks source link

evalDevelopment function pb #122

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago

The game in prev iussue(121) shows that the eval or maybe evalDevelopment
function has problems.
1. e4,c5
2. d4,cxd4
3. Qxd4,Nc6
4. Qd1,Nb8

The last move of Black contradict the goal of evalDevelopment use.

Original issue reported on code.google.com by gbtami on 24 Dec 2006 at 1:35

GoogleCodeExporter commented 9 years ago

Oh yeah, nasty one. I actually think PyChess grow quite a few ELO's there.
Thanks.

Original comment by lobais on 26 Dec 2006 at 11:41

Changed state: Fixed

GoogleCodeExporter commented 9 years ago

The problem still exist. See the attached log file.
Played the game 2 times. First was ok, but second produced the bogus Nc6-g8.

Original comment by gbtami on 28 Dec 2006 at 9:59

Attachments:

pychess-log.txt

GoogleCodeExporter commented 9 years ago

Original comment by lobais on 1 Jan 2007 at 7:01

Changed state: Reopened
Added labels: Milestone-Release0.6, Priority-Critical
Removed labels: Priority-Medium

GoogleCodeExporter commented 9 years ago

Wasn't caused by the transposition table directly, but a aplhabeta speedup, 
that gave
illegal moves.

Original comment by lobais on 2 Jan 2007 at 3:33

GoogleCodeExporter commented 9 years ago

Original comment by lobais on 2 Jan 2007 at 3:34

Changed state: Fixed

GoogleCodeExporter commented 9 years ago

Tested with svn r326, pychess takes the bogus 4.---, Nc6-b8 :(

Original comment by berad...@gmail.com on 2 Jan 2007 at 4:13

GoogleCodeExporter commented 9 years ago

And the stacktrace?

Original comment by lobais on 2 Jan 2007 at 4:46

Changed state: Reopened

GoogleCodeExporter commented 9 years ago

There is no traceback with r326.

Original comment by gbtami on 2 Jan 2007 at 7:35

GoogleCodeExporter commented 9 years ago

So it is not a crash?

Is it the evaluation problem then? That black gains points instead of loosing 
them,
at Nc6-b8?

Original comment by lobais on 2 Jan 2007 at 7:43

GoogleCodeExporter commented 9 years ago

There is no crash with svn head now (it was before).
The remaining bug with test game above is the wrong 4. move of black.
Seems the eval function is ok. It evaluatess Nc6-b8 
with score 44, and f.e. Ng8-f6 with score -52. So it seems OK.
I feel the pb exist somewhere in alphaBeta().

Original comment by gbtami on 3 Jan 2007 at 3:19

GoogleCodeExporter commented 9 years ago

Oh, so the problem simply is that it makes a move that could have been better?

Original comment by lobais on 3 Jan 2007 at 3:48

GoogleCodeExporter commented 9 years ago

It evaluates moves in a correct way, but decides to chose not the best scored 
move,
but one of the worst.

Original comment by gbtami on 3 Jan 2007 at 7:03

GoogleCodeExporter commented 9 years ago

I see. Yes it is probably a seek problem. But don't be sure, perhaps it has just
calculated that this will be the best. You should not look at the score of 
following
position, but at the score of the position it ends with.

Original comment by lobais on 5 Jan 2007 at 9:06

Changed state: Need-Info
Added labels: Priority-Low
Removed labels: Milestone-Release0.6, Priority-Critical

GoogleCodeExporter commented 9 years ago

Seems the patch sended with issue 141 fixes this issue too.

Original comment by gbtami on 13 Jan 2007 at 7:48

GoogleCodeExporter commented 9 years ago

Original comment by lobais on 7 Feb 2007 at 9:00

Changed state: Fixed