Plausibility based extension/reduction

Discussion of chess software programming and technical issues.

Moderator: Ras

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 7:43 pm

Plausibility based extension/reduction

Post by sje »

Plausibility based extension/reduction

Back in the 1960s, Richard Greenblatt and some of his colleagues wrote Mac Hack VI, a Shannon type B chess playing program. Mac Hack used about fifty plausibility heuristics for selecting and ordering the best N moves at a node to be chosen for further examination.

Some of these heuristics were path dependent. For example, the program assigned higher plausibility to moves of pieces that had not moved two ply earlier.

My idea is to use path dependent heuristics to assist with search depth extension and reduction management.

For example:

1) Moving the same piece back to its immediately prior square -> reduce

2) Moving the same sweeper (BRQ) piece twice in a row in the same direction -> reduce
Teemu Pudas
Posts: 88
Joined: Wed Mar 25, 2009 12:49 pm

Re: Plausibility based extension/reduction

Post by Teemu Pudas »

sje wrote:For example:

1) Moving the same piece back to its immediately prior square -> reduce

2) Moving the same sweeper (BRQ) piece twice in a row in the same direction -> reduce
Either of those in response to a nullmove -> prune.
Gerd Isenberg
Posts: 2251
Joined: Wed Mar 08, 2006 8:47 pm
Location: Hattingen, Germany

Re: Plausibility based extension/reduction

Post by Gerd Isenberg »

sje wrote:Plausibility based extension/reduction

Back in the 1960s, Richard Greenblatt and some of his colleagues wrote Mac Hack VI, a Shannon type B chess playing program. Mac Hack used about fifty plausibility heuristics for selecting and ordering the best N moves at a node to be chosen for further examination.

Some of these heuristics were path dependent. For example, the program assigned higher plausibility to moves of pieces that had not moved two ply earlier.

My idea is to use path dependent heuristics to assist with search depth extension and reduction management.

For example:

1) Moving the same piece back to its immediately prior square -> reduce

2) Moving the same sweeper (BRQ) piece twice in a row in the same direction -> reduce
If a repetition is the best you can get, 1) should not reduced let say directly into qsearch to eventually miss the (forced) repetition. Also, if after move(A)-move(B, attacking the to-square of move(A)) moving back is forced, but the whole maneuver has weakened let say the pawn structure. For 2) there are exceptions as well, for instance if the sweeper was forced to move twice, because it got attacked.