Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Storage idea (maybe Mr Corbit is interested)

Author: Dann Corbit

Date: 14:06:14 10/23/01

Go up one level in this thread


On October 23, 2001 at 16:11:32, Angrim wrote:

>On October 23, 2001 at 13:35:53, Dann Corbit wrote:
>
>>On October 23, 2001 at 02:17:15, Nino wrote:
>>[snip]
>>>Hi Angrim,
>>>
>>>How are egtb's done if the position is not stored? (just curious)
>>>Anyway I really dont know much about egtb's but I do know a little bit about
>>>storing of epd positions and that the current method I think requires 192 bits
>>>per position.  I have just put together a table using Les's concept but for
>>>storing chess positions.  The table follows:
>>>
>>>        Bits
>>>Square  Saved   % saved
>>>
>>>1x3	50	78.1
>>>1x4	49	76.6
>>>1x5	48	75.0
>>>1x6	47	73.4
>>>1x7	46	71.9
>>>1x8	45	70.3
>>>2x3	47	73.4
>>>2x4	45	70.3
>>>2x5	43	67.2
>>>2x6	41	64.1
>>>2x7	39	60.9
>>>2x8	37	57.8
>>>3x1	50	78.1
>>>3x2	47	73.4
>>>3x3	44	68.8
>>>3x4	41	64.1
>>>3x5	38	59.4
>>>3x6	35	54.7
>>>3x7	32	50.0
>>>3x8	29	45.3
>>>4x1	49	76.6
>>>4x2	45	70.3
>>>4x3	41	64.1
>>>4x4	37	57.8
>>>4x5	33	51.6
>>>4x6	29	45.3
>>>4x7	25	39.1
>>>4x8	21	32.8
>>>5x1	48	75.0
>>>5x2	43	67.2
>>>5x3	38	59.4
>>>5x4	33	51.6
>>>5x5	28	43.8
>>>5x6	23	35.9
>>>5x7	18	28.1
>>>5x8	13	20.3
>>>6x1	47	73.4
>>>6x2	41	64.1
>>>6x3	35	54.7
>>>6x4	29	45.3
>>>6x5	23	35.9
>>>6x6	17	26.6
>>>6x7	11	17.2
>>>6x8	 5	 7.8
>>>7x1	46	71.9
>>>7x2	39	60.9
>>>7x3	32	50.0
>>>7x4	25	39.1
>>>7x5	18	28.1
>>>7x6	11	17.2
>>>7x7	 4	 6.3
>>>7x8	-3	-4.7
>>>8x1	45	70.3
>>>8x2	37	57.8
>>>8x3	29	45.3
>>>8x4	21	32.8
>>>8x5	13	20.3
>>>8x6	 5	 7.8
>>>8x7	-3	-4.7
>>>8x8    -11     -17.2
>>>
>>>It is interesting to see the gains using this method.  Remember to throw out the
>>>7x8, 8x7 and 8x8 sets.  Now those numbers might be interesting to ponder when it
>>>comes to storing chess positions.  Unless my math is wrong can someone tell me
>>>if they see anything wrong with this approach?  Mr Corbit??
>>
>>I think it is a good idea, which compresses the 64 bit board image pretty well.
>>My problem is I don't really know what Eugene does internally.  To me, the
>>tablebase is just a black box that spits out answers.  Probably Eugene can
>>address whether this notion will save anything or not.
>
>Here is a simplified version of the algo for calculating the index into
>the tablebase file.  This index could be used to store board positions,
>since the calculation is reversible.
>1. use the filename or a few bytes at the start of the file to store which
>pieces are present.
>2. for each piece, calculate that pieces index from 0-63,
>  notation i[N] is index of piece #N
>3. for 3 piece table, index is i[1]*64*64+i[2]*64+i[3]
>
>This is a number between 0 and 63*64*64+63*64+63=262143 so could
>be stored in 18 bits.  Note that this is used as an index into the file
>and so does not need to be stored at all. Also note that the actual
>algo used takes board symetries and legal positions for kings into
>account, and so is much more efficient.

Here are the statistics for KQK, which is a 3 piece file.  It is a lot less than
262,143 entries (by a factor of about 4.8).

wtm: Mate in  10:	1
wtm: Mate in   9:	375
wtm: Mate in   8:	1936
wtm: Mate in   7:	4102
wtm: Mate in   6:	4101
wtm: Mate in   5:	3313
wtm: Mate in   4:	2546
wtm: Mate in   3:	1157
wtm: Mate in   2:	649
wtm: Mate in   1:	312
wtm: Broken positions:	7137
btm: Lost in   0:	46
btm: Lost in   1:	175
btm: Lost in   2:	397
btm: Lost in   3:	949
btm: Lost in   4:	1823
btm: Lost in   5:	3230
btm: Lost in   6:	5077
btm: Lost in   7:	6883
btm: Lost in   8:	5603
btm: Lost in   9:	1500
btm: Lost in  10:	8
btm: Draws:	2953
	54273

This brings a lot of questions to my mind.  If (for instance) we are actually
representing 262,143 (less illegal) positions, how many wins/losses/draws are
actually contained in the tablebase file?

Is the compression based solely on the rotations of the board, or are mirror
reflections used also?



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.