Search found 62 matches

by 64_bit_checkers_engine
Sun Aug 02, 2009 18:50
Forum: Draughts, Computer, Internet
Topic: handling db blocks
Replies: 7
Views: 4276

Buffer diagrams

Another thing I did to help me understand how my buffer was performing, and how I might optimize it, was to create a utility for drawing the contents of my buffer. A snapshot from solving the 5-piece slice: <----------TOP---------> --------------------------------------------------------------------...
by 64_bit_checkers_engine
Sun Aug 02, 2009 17:37
Forum: Draughts, Computer, Internet
Topic: handling db blocks
Replies: 7
Views: 4276

I think your concern is, how to find a block in your most-recently-seen buffer. Here is how I do it for my uncompressed databases, 10-bits per position, with distance-to-win probed in RAM during the search: total blocks (g_buffer_node_count) defined for current db solving run = 0 DB 2 slice # 1 requ...
by 64_bit_checkers_engine
Sun Aug 02, 2009 17:33
Forum: Draughts, Computer, Internet
Topic: handling db blocks
Replies: 7
Views: 4276

(see next post)
by 64_bit_checkers_engine
Sun Aug 02, 2009 17:25
Forum: Draughts, Computer, Internet
Topic: handling db blocks
Replies: 7
Views: 4276

Re: handling db blocks

Hi Ed, Let's suppose that the 8 pc db need about 400Gb of disk space. With about 4kb by block that means that we have to handle about 100M block of data. For the 7pc db I needed 96bits per block to describe where is the block on the disk, where is the block in memory and which positions are found i...
by 64_bit_checkers_engine
Sat Aug 01, 2009 22:47
Forum: Draughts, Computer, Internet
Topic: EndGame Databases
Replies: 289
Views: 125287

For the larger databases (>6 pieces), they have for every database subdivision (wm, wk, bm, wk, wr, br tuple) a plain text index file with for every 4K block of data in the database the 32-bit index of the position that starts the block. Suppose you build the full 8 pc databases and ultimately have...
by 64_bit_checkers_engine
Wed Jul 29, 2009 17:03
Forum: Draughts, Computer, Internet
Topic: Verification of Endgame Databases
Replies: 1
Views: 2198

More on verification

I might have to do an index-by-index comparison of the databases. I am not looking forward to this! What the database actually looks like, byte-by-byte (8 bytes per row, sorted by index, NOT position) ***************************************************************************************** * 1111111...
by 64_bit_checkers_engine
Wed Jul 29, 2009 15:35
Forum: Draughts, Computer, Internet
Topic: Verification of Endgame Databases
Replies: 1
Views: 2198

Verification of Endgame Databases

This might not bode well for all endgame database programmers. While what I am posting concerns something I discovered about 8x8 checkers, it could potentially apply equally well for International Draughts. I've computed many endgame databases before: for chess, my 10x8 chess variant Gothic Chess (s...
by 64_bit_checkers_engine
Sun May 10, 2009 01:41
Forum: Draughts, Computer, Internet
Topic: perft for 8x8 checkers (again)
Replies: 4
Views: 3676

I did up to perft(18) on a single machine. What kind of cluster did you use Aart? Did you have each machine do a move (7 boxes)? Or did each machine do 1 ply for each side (49 boxes)? move 1 of 7 from parent position, white to move nodes below after 17 plies = 550829166472 XXXXXXXXXXXXXXXXXXXXXXXXXX...
by 64_bit_checkers_engine
Sun May 03, 2009 06:03
Forum: Draughts, Computer, Internet
Topic: EndGame Databases
Replies: 289
Views: 125287

Mine is set up like this: XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX X----------X----------X----------X----------X----------X----------X----------X----------X----------X----------X X----------X----------X----------X----------X-----...
by 64_bit_checkers_engine
Sat May 02, 2009 21:05
Forum: Draughts, Computer, Internet
Topic: EndGame Databases
Replies: 289
Views: 125287

Look at figure A-1 in the appendix of the first paper: here is his 35-bit bitboard layout with 3 ghost squares! For years, IBM had a line of 36-bit machines that allowed Samuel to implement his clever bitboard scheme. The Samuel layout works great for checker moves, but not jumps. For example, jump...
by 64_bit_checkers_engine
Sat May 02, 2009 20:40
Forum: Draughts, Computer, Internet
Topic: Endgame Database Logic Question
Replies: 43
Views: 17896

If the winning side follow the DTC db it is sure to win. It will be less efficient (no doubt at all about that!) than following a DTW db but the win is 100% sure. My definition of conversion is the promotion of a checker to a king, or the capture of any piece. Just moving a checker is not a convers...
by 64_bit_checkers_engine
Sat May 02, 2009 00:53
Forum: Draughts, Computer, Internet
Topic: Endgame Database Logic Question
Replies: 43
Views: 17896

This is the type of position that can confuse a DTC checkers program *************************************************************************************************************** *----------*----------*----------*----------*----------*----------*----------*----------*----------*----------* *------...
by 64_bit_checkers_engine
Fri May 01, 2009 23:43
Forum: Draughts, Computer, Internet
Topic: Endgame Database Logic Question
Replies: 43
Views: 17896

If you take the above example, after the small mistake at the 11th white move how a program with only the DTW can win ? With a DTC (or DTZ) we saw it is possible but it might not be possible with only a DTW db. The DTW is the proven best possible move, period. You should read the paper I published ...
by 64_bit_checkers_engine
Fri May 01, 2009 05:00
Forum: Draughts, Computer, Internet
Topic: Open Dutch Championship Computerdraughts
Replies: 38
Views: 16766

Too soon for me this year. Maybe next year! That rule with "must take the most number of pieces" is killing me right now.
by 64_bit_checkers_engine
Fri May 01, 2009 00:59
Forum: Draughts, Computer, Internet
Topic: Endgame Database Logic Question
Replies: 43
Views: 17896

When setting out to build a DTZ database, for whatever your "Z" value is, you need: A. (2 * Z) - 1 values for won and lost positions. B. 1 value for a theoretical DRAWN result C. 1 value for an UNKNOWN position D. 1 value for a "TOO LONG TO LOSE" DRAW. E. 1 value for a "TOO LONG TO WIN" DRAW. The re...