Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hashkey collisions (typical numbers)

Author: Renze Steenhuisen

Date: 08:44:26 04/07/04

Go up one level in this thread


>>>>>>How important is it to hash in the castling rights and en-passant fields?
>>>>>
>>>>>Very important; it will go wrong at a tournament otherwise.
>>>>
>>>>Yes, I read of some (your?) tournament issue?
>>>>
>>>>But is it important for my results, now? Because I am using the starting
>>>>position as test position...
>>>
>>>If you search above a ply or 4 sure, as then you can already start capturing en
>>>passant.
>>
>>Should I put it in first, could that really be the problem?
>
>Depends upon how you measures errors/collissions.

>Define collision first.
>
>(1) two different positions produce the same hash signature...
>
>(2) two different hash signatures address the same table entry...
>
>(1) should not happen with 64 bit signatures.  (2) is common and is why the
>replacement strategy is so important.

I defined collision somewhere in this thread but it is the same as (1):

from code of DarkSight which is comparable with next.c in Crafty
/********************/

    case HASH_MOVE:
        if( a tt_move is provided )
        {
            tree->stats.hashkey_requests++;
            if( provided tt_move is valid move )
                return tt_move;
            else
                tree->stats.hashkey_collisions++;
        }

/********************/

When I search from the starting position I get:
   tree->stats.hashkey_collisions  =    34
   tree->stats.hashkey_requests    = 38503

Or to give the complete output...

nodes searched-> 4651507
qnodes/nodes_searched-> 0.24
capture moves-> 0
evaluations done-> 4150454
first fail highs-> 220020
fail highs-> 257405
move-ordering: 85.48
NULL-move fail-highs-> 903676
tt_probes-> 1367771
tt_hits-> 139286
tt hit-ratio: 10.18
TT overwrites-> 87645
  number of LOWERbounds-> 257405
  number of EXACTbounds-> 499
  number of UPPERbounds-> 115898
hashkey collisions: 34 / 38503 = (0.0009)
ptt_probes-> 4150454
ptt_hits-> 3627474
ptt hit-ratio: 87.40
PTT overwrites-> 481427
PTT key collisions-> 82819
Futility Prunings-> 287703
EXTENSIONS:
    Check        Ext: 39065
    Mate Threat  Ext: 27
    Single Reply Ext: 0
    Pushed Pawn  Ext: 1086
    Recapture    Ext: 31824
    Singular     Ext: 0
captures: 374765
normals : 228993
norm/caps: 0.61



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.