Author: Peter McKenzie
Date: 13:36:54 07/05/01
Go up one level in this thread
On July 05, 2001 at 15:21:22, Randall Shane wrote: >Just a little post to work off frustration with my program, FreeChecking. > >I know it's not very good (around 1700-1800 ICC blitz when not broken), and it >makes up for lack of quality by being slow, and I've been rewriting parts of it >to simplify the spaghetti mess it had become over a couple of years of very >intermittent coding, but still... > >You know something's wrong somewhere when from the starting position, no book, >your search gives something like this (fixed 9-ply search) : > >FreeChecking mode=test ply=9 >Vars : cp(W) ply(9) pos((null)) trace(trace.txt) hash(23) phash(21) spec(TEST) > 1 0.00 plst [-0.25,0.25] [] > 1 0.00[ 21] plen 0.12 Pe2-e4 | Pe2-e4 > 2 0.00 plst [-0.13,0.37] [Pe2-e4] > 2 0.00[ 101] plen 0.00 Pe2-e4 | Pe2-e4 Pe7-e5 > 3 0.00 plst [-0.25,0.25] [Pe2-e4] > 3 0.00[ 721] plen 0.23 Pe2-e4 | Pe2-e4 Pe7-e5 Pd2-d3 > 4 0.01 plst [-0.02,0.48] [Pe2-e4] > 4 0.02[ 3399] plen -0.02 Ng1-f3 | Ng1-f3 Pf7-f5 Pd2-d4 Pe7-e5 > 5 0.02 plst [-0.27,0.23] [Ng1-f3] > 5 0.04[ 5146] plen 0.23 Ng1-f3 | Ng1-f3 Pd7-d5 Nf3-d4 Bc8-d7 Nb1-c3 > 6 0.04 plst [-0.02,0.48] [Ng1-f3] > 6 0.32[ 53609] plen -0.02 Ng1-f3 | Ng1-f3 Nb8-c6 Pe2-e4 Ng8-f6 Bf1-d3 >Pd7-d5 > 7 0.32 plst [-0.27,0.23] [Ng1-f3] > 7 0.61[ 100688] plen 0.23 Ng1-f3 | Ng1-f3 Pd7-d5 Nf3-d4 Pe7-e5 Nd4-b5 >Pg7-g5 Nb1-c3 > 8 0.61 plst [-0.02,0.48] [Ng1-f3] > 8 6.30[1123379] plen -0.02 Ng1-f3 | Ng1-f3 Pe7-e5 Pe2-e3 Nb8-c6 Pb2-b3 >Pd7-d5 Rh1-g1 Ng8-f6 > 9 6.30 plst [-0.27,0.23] [Ng1-f3] > 9 12.06[2139749] plen 0.23 Ng1-f3 | Ng1-f3 Pe7-e5 Nf3xe5 Pf7-f5 Pe2-e4 >Pd7-d6 Ne5-f3 Pf5xe4 Nb1-c3 > 9 12.06[2139749] move 0.23 Ng1-f3 | Ng1-f3 Pe7-e5 Nf3xe5 Pf7-f5 Pe2-e4 >Pd7-d6 Ne5-f3 Pf5xe4 Nb1-c3 > [nps=177469] [reg=697723,qs=1442026,eval=1442026,ill=43503] > [pos hits=370312, cutoffs=741, misses=1452974] > [pawn hits=1255167, misses=186859] >Move is Ng1-f3[0.23] > Ng1-f3 Pe7-e5 Nf3xe5 Pf7-f5 Pe2-e4 Pd7-d6 Ne5-f3 Pf5xe4 Nb1-c3 > > >I can't believe the final PV that my program is generating. I would suspect the >new search, except the old search didn't do any better. Probably something with >hashing, evaluation, or something else. :-) The immediate question that springs to mind is whether this is a search problem, or a PV recording problem. If its a search problem, you really do have serious problems :-) But if its just that you are incorrectly recording (or displaying) the PV, then its not so bad. If you play 1.Nf3 and let your program do an 8ply search, does it really think 1...e5 is the best move?? Some of those PVs above are really screwed up, so if its a search bug I'd guess it was hashing. How about turning hashing off and running the test again? Have fun! cheers, Peter
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.