🚀 go-pugleaf

Folks-

In the interests of adding BW to this group, and waking up
all the computer/chess types out there,  I am posting the
following questions, suggestions and modifications to GNUchess.

1. GNUchess (at least the version I got from MIT, I don't
remember which version # it is) evaluates nights as only 5 points less
than a bishop in quiessent state.  Since a pawn is nominally 100 points,
this leads to GNUchess giving up the 2B's just to double its opponent's
pawns.  Now, IMHO the doubled pawns are dynamic and can be undoubled
with proper play, leading to positions where GNUchess is simply down
the B pair with no compensation.  By modifying the value of a knight to
be 300 (lowering it by 1/2 pawn), GNUchess is much more protective of
the 2 B's, and plays slightly stronger.  In a G/15 match between the
modified GNUchess and the unmodified GNUchess, the modified version
went +7 -2 =12.

2. Another obnoxious piece of behavior with GNUchess is that its
book does not handle transpositions or do any evaluations on the
positions.  While I am still thinking about how to build a hashed
transposition table in GNUchess (I've been kicking ideas around for
about a year, but haven't managed to get the propagation of position
values to move backword through the history of positions properly), I
am wondering if anyone out there has solved this problem.

3. A minor quip:  has somebody out there fiqured out how to use
shorthand algebriac (ie. e4 c5 nf3, rather than e2e4 c7c5, g1f3)?

4.  Lastly, and where I think the place where the largest improvment
as far as the programs speed is concerned, is that GNUchess uses a linear
tree for move generation in its searches.  This is fine, except that it
calculates moves for all ply>1 repeatedly.  Has anyone thought of a way
to keep a tree of all the legal moves around after a particular move
has been played, so that at the next ply, the program is only required
to play the moves and not generate them.  An alternative (although
I doubt it could be made to fit in any reasonably sized machine) is
to maintain a record (list, whatever) of all positions arising
from the tree, and conducting the next ply of the search by loading
the position and doing a 1 ply search on it.

Ryan Moats
AT&T Bell Laboratories
Holmdel, NJ

PS.  Another thing.  Has anybody figured out a good way to improve
GNUchess's admittly gross endgame play?  Like, any ideas for encoding
most of the Averbakh series on endgames?