version of Wed Dec 4 15:33:10 EST 1996
If you find an error, please let me know at plaat@theory.lcs.mit.edu
-
page 14, 5th line from bottom of page
"denotred" should be "denoted"
- page 23, figure 2.13
The figure shows a version of NegaScout that is not always fail-soft,
because of line 7 and 15. The figure should have shown a fail-soft
NegaScout. The code, upon failing low or high
sometimes returns alpha or beta, instead of the value of the child
that caused the fail low/high. Non-fail-soft algorithms are a thing
of the past. (And mixtures of the two make no sense at all.)
The algorithms in chapter 3 and 4 all need fail-soft versions of
Alpha-Beta for their Null-Window searches. If you need a
fail-soft NegaScout, look here
or in [116] (not [115], which contains an
error) or [81], [84]. (The rest of the thesis does not use NegaScout,
but MT or Alpha-Beta. The code for these algorithms is the correct
fail-soft variant.)
-- Thanks to Dennis Breuker for finding this bug and the one in [115].
- page 43, paragraph starting with "Note that while..."
You can save space by not storing both bounds at both node types.
Alpha-Beta with a transposition table only has to store upper bounds,
f^+, at min nodes, and lower bounds, f^-, at max nodes, and the
mechanism still works, give or take a few extra node expansions to
retrieve the bounds of the children.
- page 133, column "Alpha-Beta" and "NegaScout" position 10 of the table
The number 1869436 is bold, whereas is 1763143 is not. Since the
lowest number should be in bold, this is wrong.
-- Thanks to Roel van der Goot for finding this bug.
- page 147, references [54], [55], [56]
Feng-hsiung Hsu's first name is usually spelled with a hyphen.
Back to Aske Plaat