WebSVN – Games.Chess Giants – /engines/Crafty/readme.html

<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html> <head>
<title> Crafty Command Documentation (version 18)
</title>
</head>
<body>
<h1> Crafty Command Documentation
</h1>
<!--- Note that I have put "Crafty" into <cite></cite>. This is
logical, as a computer program with source code provided is not less
than a book. On the other hand, maybe the document looks a bit garish
as a result? A matter of taste. ---JM. --->
<cite>Crafty</cite> is nothing more than a long-time hobby of mine,
dating back to <cite>Blitz</cite> and later <cite>Cray Blitz</cite>.
People ask me how I keep doing this, and that is the one question that
generally leaves me at a loss for words.
Perhaps the most common question I'm asked is "is this version of
<cite>Crafty</cite> some dumbed-down version of what you play on ICC
or what you use at a computer chess event?" The answer is a
resounding *NO*. The current version is
exactly what is running on ICC under this version number.
Note that a new version can, on occasion, introduce weaknesses or
outright bugs that were not present in previous "gold" versions. As a
result, you should be careful to back up your "favorite" before trying
the latest and greatest. If you aren't satisfied with the new
version, you can then go back to what you believe is a better version.
If you are looking for the strongest playing computer chess
program available, you should likely look to <cite>Fritz</cite>,
<cite>Rebel</cite>, <cite>Tiger</cite>, and the other commercial
entries. There you will find strong opponents with polished
interfaces that have been tested in a systematic and careful way. If
you are looking for a program that plays good chess, has a reasonable
set of features for you to use, is available in source form, and one
where the author welcomes feedback, code or suggestions, then you are
at the right place. I welcome comments and suggestions, and also
feedback from ideas you try yourself that seem to work.
<cite>Crafty</cite> is a state-of-the-art computer chess program,
and uses all of the search algorithms you have probably read about,
negascout search, killer/history move ordering, SEE (Static Exchange
Evaluation) quiescence move ordering and pruning, hash
(transposition/refutation) tables as well as evaluation caches,
selective extensions, recursive null-move search, and a host of other
features that have been used and are still being used in most computer
chess programs. If it's not in <cite>Crafty</cite>, either it is on
the "to do" list, or it has been tried, found wanting, and discarded.
Chess knowledge is growing, and suggestions (or even better, real
code) are welcome. This is the best place to contribute your ideas,
because knowledge can be used to supplant search and make it play
better. The evaluation is probably the easiest place to start
studying <cite>Crafty</cite> because of the comments and simplicity of
using bitmaps, *once* you get "into" them.
My purpose for doing this is an exercise in computer chess
efficiency. I can't begin to count the number of people I know that
started from scratch to write a chess program. Even larger is the
group that started from scratch, and gave up before finishing, because
of the basic size of the project.
<cite>Crafty</cite> offers everyone a very clean starting point,
if you are fascinated by the bitmap chess board implementation (as I
am). The search and quiescence code is reasonably straightforward, as
is the evaluation.
It offers a great starting point, so that if you are interested
in trying a new search extension, you can be testing tomorrow, rather
than next year, because you start with a fully functional chess engine
that is not a "toy" application, but is a functional and "dangerous"
chess player. It offers a rapid start, although you can certainly
replace it piece by piece until it is "yours" if you want. It also
offers a fairly complete set of commands and an interface for a GUI as
well as support for chess server play, so that testing and debugging
your new ideas is greatly simplified.
If you'd like more information, please check out the
<code>read.me</code> document and the <code>crafty.FAQ</code> that are
distributed with <cite>Crafty</cite>. These contain recent news and
specific instructions for commonly asked questions, like "where can I
obtain tablebase files and how do I use them?"
<hr>
<h2>How to play a game.</h2>
When you execute crafty, you will immediately be
greeted by the prompt string white(1): and
<cite>Crafty</cite> will wait for commands. This prompt means it is
White on move, and we are at move #1 for White. You can first use any
of the commands from the alphabetic command listing below to tailor
the game to your liking (time control, hash table size, book
randomness, etc.) and then you have two choices. If you want to play
White, just enter your move, and <cite>Crafty</cite> will take it from
there and make a move in response. You will then be prompted by
white(2): and it is your move again. If you would
prefer to play Black, just enter either move or
go at the prompt and <cite>Crafty</cite> will move
for that side rather than accepting a move from you. After it makes
its move for White, you will then see the prompt
black(1): indicating it is now time for Black's first
move. You can enter a move, or you can once again enter
move or go and <cite>Crafty</cite>
will again move for the current side, change sides, and prompt you for
what to do next.
If you find yourself continually using a set of commands to
configure <cite>Crafty</cite> to play as you want, you can put these
commands in a startup file called <code>.craftyrc</code>
(<cite>Unix</cite>) or <code>crafty.rc</code>
(<cite>DOS</cite>/<cite>Windows</cite>). The format for this file is
just like you would type the commands at the keyboard, with the
requirement that the last line of the file must be
exit on a line by itself. Using this, each time you
start <cite>Crafty</cite>, it will first execute the commands from
this file before it prompts you for input.
While <cite>Crafty</cite> is running, you can control what it
displays, but here's a couple of samples to explain what it is saying
and why:
<pre><tt>
depth time score variation (1)
book moves {d4, c3, Nc3, d3, b3, c4, g3, b4, Be2, Bb5}
book 0.0s 70% d4
White(3): d4
time used: 0.01
</tt></pre>
This is the normal output for those cases where <cite>Crafty</cite> is
in book. The book moves line gives the set of book moves that made
the first selection cut (see the book selection explanation given
later), followed by the move actually played, in this case d4.
If <cite>Crafty</cite> is out of book, then the output looks
somewhat different as given below:
<pre><tt>
depth time score variation (1)
4-> 0.81 2.09 6. dxe4 Bxe4 7. Rad8 Qf2 8. Qb5
5 1.37 2.41 6. dxe4 Bxe4 7. Ne5 Qf4 8. Bxe4+ Qxe4 9. f5
5-> 1.88 2.41 6. dxe4 Bxe4 7. Ne5 Qf4 8. Bxe4+ Qxe4 9. f5
6 7.38 -- 6. dxe4
6 11.90 1.97 6. dxe4 Bxe4 7. Rab8 Qf2 8. Qc7 Nc5 9. Qe5
6 12.92 ++ 6. Ne5
6 13.71 2.23 6. Ne5 Qg2 7. Ng6 h5 8. Nh4 Qg4
6-> 15.59 2.23 6. Ne5 Qg2 7. Ng6 h5 8. Nh4 Qg4
time: 15.60 cpu:99% mat:1 n:246565 nps:15927
ext-> checks:4706 recaps:1336 pawns:0 1rep:301
nodes full:45951 quiescence:200614 evals:104657
endgame tablebase-> probes done: 0 successful: 0
</tt></pre>
Let's take this stuff one line at a time.<ul>
<li> Lines that have something like <samp>4-></samp> in the depth
column are printed when that iteration (depth) is completely finished.
<li> The time and score columns should be obvious as to their
meaning as is the PV, the sequence of moves that led to this score.
One note about the "score" column. As of version 18,
<cite>Crafty</cite> displays the score with + values good for White,
- values good for Black, no matter which side it is playing in the
game. All output now follows this convention, from playing, to
analysis mode, to annotating your games, to whispering/kibitzing on
the chess servers, and so forth. This is unlike other engines, but
once you get used to it, it is much less confusing when you remember
that negative scores are good for Black and bad for White, and
vice-versa.
<li> The line that has <samp>--</samp> in the score column means that
when we started depth 6, dxe4 turned out to be worse than we thought
(notice score dropped from 2.411 last search to 1.972 for this move
this search.) To resolve this, <cite>Crafty</cite> lowers the lower
search bound (alpha) and re-searches the move to find the score.
<li> The line with <samp>++</samp> means that this move
(<samp>Ne5</samp>) is better than the best move so far, so
<cite>Crafty</cite> raises the upper search bound (beta) and
re-searches this move to find the new score.
<li> The first line of statistics gives the total time taken for
this search, the cpu percentage which should stay at 98-100%
unless your machine is heavily loaded or unless <cite>Crafty</cite>
is in an endgame that is having lots of contact with endgame
databases. If this drops below 98%, it means that
<cite>Crafty</cite> is not getting full CPU usage and will be
playing weaker than normal. The mat:1 is simply the true material
score, since <cite>Crafty</cite>'s positional scores are often
larger than a pawn.
</ul>
<hr>
<h2>Alphabetic Listing of Commands</h2>
<dl>
<!-- This is formatted as a `definition list', quite properly.
However, there is a problem about what tags to wrap around the
`definiendum'. The general style, `dfn', is not quite suitable for
computer commands. The logical choice, `code', does not show up well
in Lynx, nor in the default settings of Mozilla. So I have cheated
slightly, and used `strong'. Further, it all looks better with the
crude expedient of putting an empty paragraph at the end of each
definition. ---JM. --->
<a name="alarm"></a>
<dt> alarm on|off
<dd> This command is used to control <cite>Crafty</cite>'s "beep"
after it makes a move. Turning this off will make <cite>Crafty</cite>
"quiet" when it plays, but also makes it easy to miss a move if you
are using <cite>Crafty</cite> to play in a tournament. This is
primarily designed to make <cite>Crafty</cite> tolerable during late
night matches.
<a name="analyze"></a>
<dt> analyze
<dd> This command puts <cite>Crafty</cite> into analyze mode. In this
mode, <cite>Crafty</cite> starts computing for whichever side is on
move, and it continues computing and showing its analysis until a move
is entered. This move is made, <cite>Crafty</cite> changes sides, and
starts thinking and printing analysis all over, but for the other side
now.
This command is useful to play through a game, because you get
instant feedback when you try a move. If you want to try a different
move from the one you just entered, use the back
command to back up one move, or use back <n> to
back up <n> moves. Note that one move is a single move for the
last player, not a move for both sides. To unmake the most recent 2
moves (one for Black, one for White) use back 2.
<a name="annotate"></a>
<dt> annotate | annotateh
<var><filename> <colors</var> | <var>name> <moves>
<margin> <time></var>
<dd> This command is used to annotate (make comments in) a game that
has already been played.<dl>
<dt> annotate
<dd> produces a file with the .can extension added to the original
name. This file will contain pure ascii information from the
annotation pass.
<dt> annotateh
<dd> produces an HTML file instead (with the .html extension). This
includes the normal output, plus a nice bitmapped graphical board
display for every position where <cite>Crafty</cite> had 'something to say'.
<dt> <var><filename></var>
<dd> is the name of the file that has the game moves stored in it.
This should be a PGN-compatible file, although <cite>Crafty</cite> can
read nearly any file with chess moves and convert it to pgn using the
<a href="#read">read</a> and <a
href="#savegame">savegame</a> commands to perform the
conversion.
<dt> <var><colors</var> | <var>name></var>
<dd> indicates which side <cite>Crafty</cite> will annotate. The
valid choices are w, b, and
wb/bw for White only, Black only,
and both, respectively. <cite>Crafty</cite> will search and produce
results for the indicated color only, making moves for the other side
silently as they are read in. Alternatively, you can specify the
player's name (useful if you want to annotate several of your own
games in one large pgn file, for example, and you alternated colors so
that you can't pick the right one easily). <cite>Crafty</cite> will
then figure out which side to annotate for in each game. Note that
the name is case-sensitive, but that you only have to enter a string
that is unique in the name field. IE if one name is "Anatoly Karpov"
and the other is "unknown" then specifying Karpov as the name would be
sufficient. If the same 'string' appears in both names,
<cite>Crafty</cite> will complain.
<dt><var><moves></var>
<dd> indicates the moves that should be annotated. If this is a
single integer, annotation starts at this move number (for the color
given above) and proceeds for the rest of the game. If a range is
given, as (20-33), then only moves 20-33 inclusive are annotated. To
annotate the complete game, you can use 1-999.
<dt> <var><margin></var>
<dd> gives a score "window" that controls whether
<cite>Crafty</cite> will produce comments (see below). The larger
this number this number, the fewer annotations <cite>Crafty</cite>
will produce. A negative number will result in an annotation for
every move selected.
<dt> <var><time></var>
<dd> indicates the time limit for each search. Since each move
selected requires two searches, you can take the number of moves,
double this number and multiply by <time> to determine how long
the annotation process will take. This time is in seconds.
</dl>
How it works. Suppose you use the command annotate game1
w 1-999 1.000 30. This asks <cite>Crafty</cite> to read the
file "game1", and annotate the white moves for the entire game. The
margin is 1 pawn and the search time limit is 30 seconds. The output
for the annotate command is found in <filename>.can, in this
case this is game1.can.
<cite>Crafty</cite> first searches the move actually played in
the game to determine the score for it. <cite>Crafty</cite> then
searches the same position, but tries all legal moves. If the score
for the best move found in this search is greater than the score for
the move actually played plus the margin, then a comment is added to
the output file. This output file is quite short, with all the game
moves (plus any PGN tags in the original, for identification purposes)
plus the brief comments.
An annotation looks like this: {real_value
(depth:best_value PV moves)}
<ul>
<li> real_value is the score for the move actually
played.
<li> depth is the depth <cite>Crafty</cite>
searched to produce the best_value and PV for what it thinks is the
best sequence of moves for both sides.
</ul>
If you set <var><margin></var> to 1.000, you are asking
<cite>Crafty</cite> to only annotate moves that either lost a pawn or
more, or moves that failed to win a pawn or more. If you set
<var><margin></var> to .300, you are asking for annotations for
any move that makes the score drop about 1/3 of a pawn below the value
for the best move <cite>Crafty</cite> found.
If you have other moves you would like to see analyzed during
this annotate process, at the point where the move can be played,
insert it into the PGN file as an analysis comment, surrounded by ()
or {} characters. <cite>Crafty</cite> will produce analysis for this
move as well. If more than one move appears inside a single set of
delimiters, only the first will be analyzed. To force
<cite>Crafty</cite> to analyze more than one move, enter them like
this: (move1) (move2) as though they were two separate comments.
<a name="ANSI"></a>
<dt>ANSI on | off
<dd> This command is used to control whether or not
<cite>Crafty</cite> attempts to display its move in reverse video or
not. For PC's, Linux, and most Unix boxes, this works fine. Should
you find yourself playing <cite>Crafty</cite> via a dumb terminal,
this might hose the terminal and interfere with your ability to see or
input moves. If moves are not displayed in reverse video, it's
probably wise to turn this off to avoid hanging the terminal you are
using.
<a name="black|white"></a>
<dt> black | white
<dd> This command simply toggles the side on move. if it is White to
move, and you enter white, nothing happens. If it is White to move
and you enter black, then it becomes Black's turn to move immediately
from the same position. Used only infrequently.
<a name="book"></a>
<dt> book
<dd> (See the <a href="#Opening Book">Opening Book Setup and Usage</a>
section near the end of this document for a full explanation of this
command and its many options.) Note that there are special commands
available (*only* on the command line, *not* in the
<code>crafty.rc</code>/<code>.craftyrc</code> files) to direct
<cite>Crafty</cite> to specific directories for the book files
(<code>bookpath=/a/b/c</code>), the tablebase files
(<code>tbpath=/i/j/k</code>) and the log files
(<code>logpath=/x/y/z</code>). Note that these commands can *only* be
used on the command line, because they must be executed before the
engine is initialized. Putting them in the
<code>crafty.rc</code>/<code>.craftyrc</code> file will produce error
messages without affecting how the files are opened.
If you need to specify multiple directories (tbpath only) you may
do so by using "<code>tbpath=path1:path2:path3:etc</code>" or else use
the more Unix-like "<code>tbpath=(path1:path2:path3:etc)</code>"
instead. The paths are separated by ":" (colon) characters and
everything is case-sensitive as usual. For
<cite>DOS</cite>/<cite>Windows</cite> users, the separator can be a
semi-colon (;) or a comma (,) to avoid the drive designation
ambiguity.
<a name="cache"></a>
<dt> cache [=] <var><N></var>
<dd> This command is used to alter the cache size used for <a
href="#egtb">endgame database</a> probes. <var><N></var> can be
a simple integer, representing the number of bytes to use or it can be
specified as nK or nM representing n * 1024 bytes or n * 1024 * 1024
bytes. This should be in multiples of the database "chunk" size,
which might vary. Using the nM form guarantees that you will have a
reasonable number.
<a name="clock"></a>
<dt> clock <var><ctime> <otime></var>
<dd> This command is primarily intended for use when
<cite>Crafty</cite> is playing in a tournament, such as the WMCCC or
WCCC events. If the operator is somewhat slow in entering moves, or
forgets to stop the clock after making a move for <cite>Crafty</cite>,
the chess clock for the game will drift from the values that
<cite>Crafty</cite> maintains internally. <var><ctime></var> is
the time (in minutes or hh:mm format) <cite>Crafty</cite> has left
until the next time control, and <var><otime></var> is the
opponent's remaining clock time. This command can be used at any
time, but will only affect the time per move *after*
<cite>Crafty</cite> makes the current move and starts to think about
what the opponent might do next.
<a name="computer"></a>
<dt> computer
<dd> This command usually comes from <a
href="#xboard">XBoard/WinBoard</a>, but can be used at any time to
tell <cite>Crafty</cite> it is playing a computer. This will prevent
some things from happening, such as a draw score that varies, as well
as adjusting the book selection code to be more selective in what it
plays.
<a name="display"></a>
<dt> display
<dd> This command is used to display the game board. This board is
displayed using the ICS style #1 type of ASCII display, with White
always at the bottom of the screen, Black at the top. Very unusable
to play a game with, but good to verify a position after it has been
set up.
This command is also used to display the various piece/square
tables, by typing display <var><piece></var>
where <var><piece></var> is replaced by pawn,
knight, bishop,
rook, queen or
king. The board is oriented in the same way as the
display board with a one-to-one correspondence between the squares.
Perhaps useful for curiosity, but not for anything else. These values
can not be modified by the user.
The final version of this command is used to control what kind of
output you will see when <cite>Crafty</cite> runs. Currently the
following options are available.
<dl>
<dt> display time:
<dd> this will make <cite>Crafty</cite> display the amount of
time each side takes after making a move.
<dt> display changes:
<dd> this will make <cite>Crafty</cite> display the PV each time
it changes during the search, including when a move fails high or
becomes a new best move.
<dt> display variation:
<dd> this will make <cite>Crafty</cite> display the PV at the end of each
iteration, but it will only show the best PV for the entire
iteration, not all of the changes.
<dt> display stats:
<dd> this enables basic search statistics output including time,
nodes and so forth.
<dt> display extstats:
<dd> this enables extended search stats including the hashing
statistics, search extension statistics and so forth.
<dt> display movenum:
<dd> causes all PV output to have move numbers embedded in them to
make the PV possibly easier to read. This causes the PV to look
like this: 12. ... Nxe4 13. Bxe4 h6 rather than
simply Nxe4 Bxe4 h6. This is very helpful when
playing on a server and whispering or kibitzing analysis. It
will also be useful when <cite>Crafty</cite> is run from within a
database program as the move numbers will sync up with the actual
game.
<dt> display moves:
<dd> will display each root move as it is searched, along with
updating the search time at the bottom of the screen, so you can
see what move is currently being analyzed.
<dt> display general:
<dd> will display general information messages whenever
<cite>Crafty</cite> wants to tell you something (ie "clearing
hash tables" or other such things like "Mate in n moves."
</dl>
If you put a no in front of any of these options,
that will disable that particular type of output.
<a name="draw"></a>
<dt> draw
<dd> offers <cite>Crafty</cite> a draw. It generally will look at the
value returned by the last search, and compare it with the value
returned by an internal function DrawScore(). If the search value is
not above this result, then <cite>Crafty</cite> will accept the draw.
If the search value is above the theoretical value for a draw,
<cite>Crafty</cite> will decline the draw. Note that
<cite>Crafty</cite> will offer draws based on internal analysis. When
it offers a draw, you can respond with "draw"
although the game does not really end until you exit
<cite>Crafty</cite>.
<a name="drawscore"></a>
<dt> drawscore <var>N</var>
<dd> sets the draw score (or contempt factor) to <var>N</var>. If you
want <cite>Crafty</cite> to avoid draws, set this number to something that is
negative. IE -50 says that a repetition (draw) is the same as being
1/2 pawn down. Setting it to +100 will make it try very hard to draw
because that looks like it is winning a pawn when it does so. Note
that this is dangerous (either way) as a -50 in a king and pawn ending
is very likely dead lost... and a repetition is better.
<a name="echo"></a>
<dt> echo <var><text></var>
<dd> This command is normally used inside a command file that you are
going to use to "feed" <cite>Crafty</cite> some positions for analysis
or whatever. Since <cite>Crafty</cite> depends on the operating
system to echo commands as they are typed, commands read in from a
file are "invisible." This gives you the ability to insert commands
into such a file so that <cite>Crafty</cite> displays a message on the
screen to give you an idea of where it is in processing the file.
<a name="edit"></a>
<dt> edit
<dd> This command has been "cloned" from <cite>GnuChess</cite> to
provide an interface with <a href="#xboard">Xboard</a>. After
entering the edit command, you are in "edit.white"
mode, where any piece/square combination you enter will add the
indicated white piece on the given square. Piece/squares are entered
as "qa3", or "bc4" for example. This puts a white queen on a3 and a
white bishop on c4. Once all white pieces are entered, typing a "c"
changes to "edit.black" mode where piece/square combinations now place
black pieces. Typing a "." character exits edit mode. To clear the
board initially, you use the "#" character.
Here's a sample to set up the original starting position, after
White has played 1. e4, but no other moves have been played.
<pre>
edit
#
ra1 nb1 bc1 qd1 ke1 bf1 ng1 rh1
pa2 pb2 pc2 pd2 pe4 pf2 pg2 ph2
c
ra8 nb8 bc8 qd8 ke8 bf8 ng8 rh8
pa7 pb7 pc7 pd7 pe7 pf7 pg7 ph7
.
</pre>
Note that input is free form, so piece/squares can be entered one per
line or all on one line. Ditto for the #, c, and . special
characters. Note also that there is no way to enter castling status
here. It is far better to use the <a href="#setboard">
setboard</a> command which uses a FEN-like syntax and
allows you to set both castling and en passant status.
<a name="egtb"></a>
<dt> egtb
<dd> This command enables the endgame databases. <cite>Crafty</cite>
will use the "tbpath" directory (if provided) to locate and register
all of the databases you have downloaded. It will report the largest
class it found, as in "5 piece tablebase files found" if you
downloaded at least one 5-piece file. If you use this command to
enable databases, you should also consider using the <a
href="commands.html#cache">cache</a> command to
specify the egtb cache size.
<a name="quit"></a> <dt> end | quit
<dd> These commands are used to terminate <cite>Crafty</cite>. Note
that you can resume a game later without having to replay the moves,
by starting <cite>Crafty</cite> using the crafty c
command. It will immediately read in the moves for the last game,
although you will have to set the time controls and clock time
remaining yourself.
<a name="evaluation"></a>
<dt> evaluation <var>option <value></var>
<dd> This command is used to modify the evaluation scores.
<ul>
<li> The option asymmetry is used to make
<cite>Crafty</cite> evaluate king safety differently for each side.
evaluation asymmetry 25 will increase the king
safety scores for the opponent only, meaning it will pay less
attention to its own king safety than to that of its opponent. This
will make it play more aggressively.
evaluation asymmetry -25 will reduce the king
safety scores for the opponent by 25%, making it care more about
its own king safety than that of its opponent. This will make it
play more defensively.
<li> The bscale option will adjust the scores for blocked
pawns. The default value is 100. Increasing this will tend to make
<cite>Crafty</cite> dislike blocked pawn positions more, which will
lead to more open positions. Note that this only affects moves
_after_ the opening book has been followed, which means that the
position might be blocked before the evaluation term has a chance to
affect the game.
<li> The kscale option will adjust all king safety
scores based on the 'value' entered. For example,
<dl>
<dt> evaluation kscale 50
<dd> will reduce all king safety scores to 50% of their normal
value.
<dt> evaluation kscale 133
<dd> will increase all king safety scores to 133% of their normal
values.<dd></dl>
<li> The option tropism is used to scale king
tropism scores. This will attract pieces toward kings. A value of
100 means no change. other values are treated as a percentage (like
scale) to increase (> 100) or decrease (<100) the king
tropism scores.
</ul>
When you use this command, you will see something like this:
<pre>
modified king-safety values:
white: 0 4 16 26 39 45 58 77 87 90 93 96 100 103 106 109
112 116 119 122 125 128 128 128 128 128 128 128 128 128 128 128
black: 0 5 20 32 48 56 72 96 108 112 116 120 124 128 132 136
140 144 148 152 156 160 160 160 160 160 160 160 160 160 160 160
</pre>
Those values represent the king-safety evaluation as the king gets
more and more exposed. This is always based on the fast that "crafty"
will be the side on move when the search starts. In the above, it was
White's move when this was typed, meaning that it appears that
<cite>Crafty</cite> will be playing Black. Notice that White's king
safety numbers are scaled by 20% to make it slightly more cautious
about its own king. If you type go in this position,
the scores get reversed as <cite>Crafty</cite>'s scores are always
left alone (with the asymmetry option) and the opponent's scores are
scaled down as indicated.
You will see similar numbers (but not black and white sets) that
represent the actual scores produced for king tropism. Note that
pieces interact to choose which element of this vector is used, but in
general, the more pieces are close to the king, the larger the element
from this array.
<ul>
<li> The pscale option is used to scale normal pawn
structure scoring in the same way as the other scaling options. 100
is the default. Values less than 100 reduce this term, values over
100 inflate it.
<li> The ppscale option is used to scale some
passed pawn scoring in the same way as the other scaling options.
100 is the default. Values less than 100 reduce this term, values
over 100 inflate it. This mainly effects outside passed
pawns/protected passed pawns. The normal pawn scoring computes the
value of a passed pawn. This term is then used to scale those terms
that modify this value further, such as two connected passed pawns
on the 6th, or a passed pawn with the king supporting it in an
endgame.
</ul>
<a name="extensions"></a>
<dt> extensions <var><type> <value></var>
<dd> This command is used to control the extension depth for the
various extensions done in <cite>Crafty</cite>'s search. The
extensions are set as decimal numbers that represent plies (or
fractions of plies) to extend for each particular reason. Most
default to 1.0 and .75, but any value can be used. Note that values
> 1.0 are _very_ dangerous as they can cause the search to become
non-terminating (<cite>Crafty</cite> will stop when it runs out of
time for the move, but it might not be able to get anything useful
from such a search).
These extensions are presently limited to a maximum of one ply of
extensions at any point in the tree. IE no matter what you set these
values to, they can not exceed one ply at present.
<dl>
<dt> incheck
<dd> This is the amount to extend when the side on move makes a move
that leaves the opponent in check. Note that <cite>Crafty</cite>
extends on the ply where the check is played, not on the next ply
where the opposite side is in check.
<dt> onerep
<dd> This is the one-reply-to-check extensions, and is done at the
point where one side is in check and has exactly one legal move to
escape from the check.
<dt> pushpp
<dd> This is the extension used for certain passed pawn pushes in
the endgame.
<dt> recapture
<dd> This is the recapture extension, and is applied when the
current move recaptures an equal-valued piece that made a capture at
the previous ply. IE BxN, PxB. Note that this can only be applied
once for every two plies so that BxN, BxB, NxB, NxN won't look like
three recaptures.
<dt> mate
<dd> This is the mate threat extensions and is applied when a null
move search returns -MATED, which means that doing nothing gets the
side on move mated. The opponent must have some sort of serious mate
threat in such a position.
</dl>
<a name="flag"></a>
<dt> flag on|off
<dd> This command is used to force <cite>Crafty</cite> to end a game
where the opponent runs out of time with <a
href="#xboard">WinBoard/XBoard</a> (on) or to ignore
this (off) if desired.
<a name="force"></a>
<dt> force [<var>move</var>]
<dd> This command is used to force <cite>Crafty</cite> to play a move
that is different from the one chosen and played by the tree search.
If [<var>move</var>] is given, and it is a legal move,
<cite>Crafty</cite> will retract its last move and make this move
instead. It does not change the side on move, but does change the
position of course. If [<var>move </var>]is not given,
<cite>Crafty</cite> will prompt you for a move to make.
<a name="hash"></a>
<dt> hash=<var>x</var> and
hashp=<var>x</var>
<dd> These commands are used to adjust the size of the hash tables in
<cite>Crafty</cite>: hash modifies the size of the
transposition/refutation table, while hashp modifies
the size of the pawn structure/king safety hash table. The size
<var>x</var> may be entered as one of the following two types of
values: nnnK where nnn is an integer indicating how many Kbytes
<cite>Crafty</cite> should use for this hash table; nnnM where nnn is
an integer indicating how many Mbytes <cite>Crafty</cite> should use.
The transposition/refutation table is the more critical of the
two, because it directly affects search efficiency, particularly in
the endgame. For this reason this should be maximized. The most
effective size for this hash table is 3/4 of your available memory.
If you don't know how to figure this out, but know that you have 16
megs for example, they you can say hash=16M and
<cite>Crafty</cite> will round that down to 12M, which is 3/4 of a
power of two size. If you study the sizes that are possible, you will
find 3M, 6M, 12M, 24M, 48M, and so forth. Anything up to, but not
including, the next size will be rounded down to the next lower size.
hashp should be set to approximately 1/2 of what is left. For
example, the P6 <cite>Crafty</cite> runs on when playing on ICC often
uses hash=48M and hashp=8M. The
only thing to watch for is that if you make this too large,
particularly under <cite>Windows</cite>, performance will suffer badly
because of paging I/O overhead. When <cite>Crafty</cite> is searching
in a normal (non-book, non-endgame database) position, the disk light
should *not* be on, indicating lots of I/O.
There is no danger in making this table too large, although you
have to watch out because if <cite>Crafty</cite> barely fits in
memory, doing something else on the machine can cause
<cite>Crafty</cite> to be swapped out completely or partially,
depending on the operating system you are using. If you are going to
use the machine for anything else while <cite>Crafty</cite> is
running, it is better to "pretend" that the machine only has 1/2 of
the memory it actually does when computing the size of the hash tables
you want to use.
<a name="help"></a>
<dt> help
<dd> This command displays multiple pages of one-line help, one
command per line. If a line ends with [help], then you can use help
followed by the specific command to get detailed help.
<a name="history"></a>
<dt> history
<dd> This command displays the history in a vertical column with one
move for White and one per Black per line. There are other ways to
display the current game moves and also to save them in files that are
explained later.
<a name="import"></a>
<dt> import <var><filename></var>
[clear]
<dd> This command is used to import any sort of learning data that
<cite>Crafty</cite> supports, which currently includes book learning
data and position learning data. This command reads the appropriate
<var><filename></var> and imports that learned data, just as
though <cite>Crafty</cite> had learned it by playing the games. The
[clear] option, if specified, causes all old learned
results to be cleared before the import operation, otherwise the
imported data is simply added to what is already present.
<a name="info"></a>
<dt> info
<dd> This command is used to display information about
<cite>Crafty</cite> and the current game. Such things as the time
control, the time left on the clocks and other information is
shown.
<a name="input"></a>
<dt> input <var><filename></var>
<dd> This command is used to redirect the console input I/O stream
from the keyboard to a file. <cite>Crafty</cite> will then read
commands from this file, rather than from the keyboard, and execute
them just as though they were typed in. Such a command file *must* be
terminated by an "exit" command (no quotes) as the
last command in the file. This reverts the input stream back to the
keyboard, and prompts you for another command or move.
This command might be used to configure <cite>Crafty</cite> for a
specific time control, by putting the appropriate time control
commands in the file, or to customize the hash table sizes as
needed.
<a name="learn"></a>
<dt> learn <var>n</var>
<dd> controls the learning facilities in <cite>Crafty</cite>.
Currently this is a 3-bit boolean switch, bit 1 (001) controls book
learning, bit 2 (010) controls position learning, and bit 3 (100)
controls result learning. learn=0 disables all
learning, learn=1 enables book learning only,
learn=2 enables position learning only, and
learn=4 enables result learning. Add the values
together to turn on more than one type of learning (default=7 to
enable everything).
<a name="level"></a>
<dt> level <var><m> <t> <inc></var>
<dd> This command was taken directly from <cite>GnuChess</cite> so
that the <cite>Xboard</cite>/<cite>WinBoard</cite> interface would
interface with <cite>Crafty</cite>. There are other better ways to
set the time, but this one is well-known. The three parameters are
<var><m></var> (number of moves in the game)
<var><t></var> initial time on the clock. After
<var><m></var> moves are made, this amount of time is added to
the clock again. <var><inc></var> is the Fischer-Clock
increment that is added back to each clock after a move is made. It
may be zero for a non-increment game.
Examples:
<pre>
level 0 5 0 (ICS 5 0 game)
level 0 5 3 (ICS 5 3 game)
level 0 15 30 (ICS 15 30 game)
</pre>
<a name="list"></a>
<dt> list GM | IM |
C | AK | S +<var>name</var>
[+<var>name</var> ...]
-<var>name</var> [-<var>name</var>
...]
<dd> This command is used to maintain the internal "lists"
<cite>Crafty</cite> uses to auto-tune itself when playing on a chess
server. There are three lists, GM,
IM and C. If <cite>Crafty</cite>'s
opponent is in any of these lists, <cite>Crafty</cite> adjusts
internal controls that affect how/when it resigns or offers draws, and
how randomly it will choose moves from the opening book. For example,
<cite>Crafty</cite> resigns much sooner against a GM, because it
assumes he knows how to win a rook-up ending, regardless of how much
time is left. By the same token, when playing against computers,
<cite>Crafty</cite> will always assume that a draw is 0.000, so that
it doesn't wreck its position trying to avoid repeating a position.
The AK list will automatically kibitz
scores/PV's if the opponent is in this list. The S
list will turn on special scoring for opponents in this list. The
only current member is "mercilous".
The syntax +name1 +name2 simply adds these players to the
specified list. To remove a name, use -name1 -name2. You can use one
command per name to remove or add, or you can use one command to add
and remove multiple names. Note that all names must be entered in
lowercase characters, using any uppercase characters will break the
matching algorithm.<dd>
<a name="log"></a>
<dt> log off | on |
<var><n></var>
<dd> This command is used to disable logging. The default is
log on, which causes <cite>Crafty</cite> to produce a
new log.nnn file for each game played. If you are running
<cite>Crafty</cite> on a server, you might use log
off, which disables creating these files as well as the
game.nnn files used to restart a game after you exit
<cite>Crafty</cite> and come back later. If you use the form
log <var>n</var> <cite>Crafty</cite> will simply
display the last <var>n</var> lines of the log on the screen. If you
use log <var>n file</var> <cite>Crafty</cite> will
copy the last n lines of the log to <var>file</var> which could be
your hard drive, or a floppy.
Note that if you run with log off, you will be
unable to find out what <cite>Crafty</cite> was thinking about since
there is no other record of the game. You will always see a game.001
because as <cite>Crafty</cite> plays a game, this contains all the
real moves played so far so that you can back up if needed. you will
also see a log.001 file, but it will be empty.
<a name="ls"></a>
<dt> ls <var><filename></var>
<dd> will list all the files that match the
<var><filename></var> wildcard (the wildcards depend on the
system you are using, but generally *, ? will work fine). you can
also supply path information in the filename if you want to list the
contents of a different directory. Just use the same syntax you would
if you were using "ls" under <cite>Unix</cite> or "dir" under
<cite>Windows</cite>.
<a name="mode"></a>
<dt> mode tournament | normal
<dd> This command is primarily used to put <cite>Crafty</cite> into
"tournament" mode, which is intended for use when <cite>Crafty</cite>
is playing in computer chess events. It accomplishes two things: (1)
makes all draws return a score of 0.000, and (2) makes
<cite>Crafty</cite> issue a message after each move showing the
internal chess clock time, and requesting that that operator check and
adjust as needed using the <a href="#clock">clock</a>
command. This primarily makes <cite>Crafty</cite> comply with
computer chess rules that say the operator can't do anything not
specifically requested by the program.
<a name="name"></a>
<dt> name <var><name></var>
<dd> This command is an ICS-play specific command.
<cite>Xboard</cite>/<cite>WinBoard</cite> uses this to inform
<cite>Crafty</cite> of the opponent's name. <cite>Crafty</cite> uses
the <var><name></var>, and looks it up in its GM/IM/C lists, and
if found, adjusts itself accordingly. This is not used by the PGN
code and this will not cause the players <var><name></var> to
show up in the PGN tag section.
<a name="new"></a>
<dt> new
<dd> This command wipes everything out and starts a brand new game.
It closes the old log-file and game-file, and opens the next
sequential numbered file. It also resets the game to the beginning
and prepares to start a brand new game. This was added for Xboard,
but it turns out that Xboard does not use this, rather it starts
<cite>Crafty</cite> fresh for each new game by first terminating the
old copy then starting a new one. Not nearly as efficient as using
new but likely safer it a program can't be sure of
resetting everything back to the initial state.
<a name="noise"></a>
<dt> noise <var><n></var>
<dd> This command sets the "noise" level in <cite>Crafty</cite>.
Basically, until <var><n></var> nodes have been searched,
<cite>Crafty</cite> will be silent and not display analysis.
This is useful in two ways. First, in end-games, 20+ ply
searches are not uncommon, and the search analysis for the first few
plies arrives so quickly that it is distracting. Second, when
observing games (new interface only) on ICS servers, this can be used
to avoid having <cite>Crafty</cite> generate too many analysis kibitzes. A value
of 100000 will basically shut off any output for the first second or
so (on a P6/200). Similarly, 1000000 will eliminate any output for
about the first 10 seconds. When watching and kibitzing games like
the World Championship games on ICC, I generally use 5000000, which is
almost one minute of silence so that the first PV it kibitzes is a
pretty deep search.
noise 0 will cause *all* analysis to be
displayed, which on a fast machine causes no problems. On a slower
machine, or over a slow phone connection, this might cause a big
communication backlog. The default is roughly one second on a P6/200
(100000) but can be modified by this command.
<a name="operator"></a>
<dt> operator <var><n></var>
<dd> Another command intended for use when <cite>Crafty</cite> is
playing in a tournament, operated by a human. This tells
<cite>Crafty</cite> to "hide" <var><n></var> minutes of time and
not use them. This time is basically allocated to the operator to
make up for the time it takes to type in moves and/or correct
mistakes. At the WMCCC events, the normal value we use is 5. Playing
on a server, this is not needed, as it is not needed if you are
playing <cite>Crafty</cite> yourself.
<a name="perf"></a>
<dt> perf
<dd> This command is primarily used in optimizing <cite>Crafty</cite>,
or to test the speed of the move generator and
<code>MakeMove()</code>/<code>UnMakeMove()</code> on different
platforms. It produces two results, the moves it can generate per
second, and the moves is can generate and make/unmake per second.
While this is not a perfect performance indicator, it does give an
"approximation" for how fast <cite>Crafty</cite> might run. In
general, the higher the numbers, the better the program will play,
although machines are certainly different. It's not uncommon to find
a machine that searches slower than another, but has a higher "perf"
value.
<a name="perft"></a>
<dt> perft <var><depth></var>
<dd> This command is generally used to confirm that the move generator
and bitmap operators are working properly. It simply takes the
current position, and generates/makes/unmakes moves and counts them.
Many programs use this from a "standard" position to make sure that
their move generator does not miss generating odd moves like
enpassant/promotions and also to confirm that the make/unmake code
correctly updates the board so that the totals remain constant across
different machines and programs, since there is no alpha/beta or
evaluation things done. if <var><depth></var> is greater than 5
or 6, it will take a *long* time, since this is basically a minimax
tree traversal that will visit *every* node within the
<var><depth></var> search horizon.
<a name="pgn"></a>
<dt> pgn <var><tag> <value></var>
<dd> This command is used to set the usual PGN tags to meaningful
values. The recognized tags are Event,
Site, Round, Date,
White, WhiteElo,
Black, BlackElo, and
Result, and the tags *are* case sensitive. The
<var><value></var> can be any valid input and blanks and special
characters are allowed. Note that the date is clearly specified in
the PGN standard and must be yyyy.mm.dd with no variance. Valid
results are 1-0 (White won), 0-1
(Black won), 1/2-1/2 (drawn) and *
(unknown). Some examples:
<pre>
pgn Event 14th World MicroComputer Chess Championship
pgn Date 1996.10.8
pgn Site Jakarta, Indonesia
pgn Round 1
pgn White Crafty
pgn WhiteElo 2400
pgn Black assassin
pgn BlackElo 2400
pgn Result 1-0
</pre>
Setting these values will result in a proper PGN file when using the
<a href="#savegame">savegame</a> command. Note that if you
use the <a href="#read">read</a> command to input a PGN game,
these values will be extracted from that game if they are given.
<a name="ponder"></a>
<dt> ponder off | on |
<var><move></var>
<dd> This command serves two purposes. First, it can be used to
disable (off) or enable (on)
thinking on the opponent's time (or pondering as it is called in many
programs including <cite>Crafty</cite>.) Turning it off will weaken
<cite>Crafty</cite> since it will not use any machine time while
waiting on the opponent to move. It is sometimes useful, however,
when playing <cite>Crafty</cite> against another computer and the
machines are not equal. If <cite>Crafty</cite> is on a faster
machine, and you attempt to adjust for this by giving the opponent
more time than <cite>Crafty</cite>, it doesn't work quite as expected,
because while the opponent is thinking, so is <cite>Crafty</cite>,
which lets it use the extra opponent time in an unexpected way. In
such a case, it's best to stop pondering in both programs.
If <var><move></var> is given, it directs
<cite>Crafty</cite> to use that <var><move></var> to ponder,
rather than the one from the previous search. Most commonly this is
used to set the right move to ponder after <cite>Crafty</cite> has
been stopped and then restarted, assuming that it is the opponent's
turn to move when this happens. Otherwise, it is probably better to
not try to influence things, although if you are watching and suddenly
wonder "what would <cite>Crafty</cite> do if the opponent plays move
'X'?", you can answer this by simply typing ponder X
and then watching the analysis. You should reset the correct ponder
move after you do this of course.
<a name="read"></a>
<dt> read | reada
[<var><filename></var>]
<dd> This command will read input, and extract the chess moves and
make them to set up the position at the end of the game. It first
resets the chess board to the initial position (read command only) and
then extracts the PGN tags (if present) from the front of the input.
The rest of the input is parsed for chess moves (comments and similar
things are culled automatically) and the moves are made and added to
the game history. Once this is done, you can back up, go forward, or
play from any point in the game. If you specify a
<var><filename></var> everything is read from the file,
otherwise it is read from the console keyboard.
The reada command reads moves, but appends them
to the current game history/position rather than resetting to the
initial chess position. This lets you read in a game, then use reada
to manually add some more moves to see the resulting position.
<a name="reset"></a>
<dt> reset <var><n></var>
<dd> This command lets you back up in the current game to any move of
your choice. reset <var><n></var> backs up the
game to move <var><n></var> with the same side on move. If you
want to first change the side to move, use the <a href="#black|white"
>white|black</a> command, then use
the reset command to back up to the right move. Note
that you can also go forward as well, just so there are moves in the
current game history.
<a name="resign"></a>
<dt> resign <var><n></var>
<dd> This command sets the resign threshold. When running on ICC I
typically use resign 9 which will make
<cite>Crafty</cite> resign roughly five moves after the score drops
below -9.000. For IM's I change this to 6, and for GM's I often use
3, so that it will resign quicker and not drag a lost game out
unnecessarily.
<a name="savegame"></a>
<dt> savegame <var><filename></var>
<dd> This command is used to save the current game in a PGN-compliant
file with the PGN tags included. Note that the default TAG values
might not be what you want if you do not either use the <a
href="#pgn">pgn</a> command to set them or else input
a valid PGN file with the tags already filled in.
Be aware that this command doesn't check the filename for
legality since anything goes in UNIX. In DOS, you might produce a bad
filename with either too many characters, too many periods, or
whatever, so be careful with the name you choose. Note also that this
file will be overwritten if it already exists, so be sure to choose a
name that is not the name of a file that has something you consider
important in it.
<a name="savepos"></a>
<dt> savepos <var><filename></var>
<dd> This command writes a single line into
<var><filename></var> in FEN-like notation. This lets you save
a position, and then come back later to re-examine it. You would use
the <a href=#input>input
<var><filename></var></a> command to input this file and set the
position up.
<a name="search"></a>
<dt> search <var><move></var>
<dd> This command allows you to specify one particular move for the
side on move, and then when you tell <cite>Crafty</cite> to search
this position, this is the only move that will be searched. This is
used internally by the annotate command, but can be used to
investigate one specific move. If the move is not the best move, a
normal search won't show you why it is bad, but this will. It is also
quite a bit faster since the other moves in the position are not
searched at all.
<a name="settc"></a>
<dt> settc <var><moves> <ctime>
<otime></var>
<dd> This command is primarily used in tournaments, and is an
error-recovery command. If the machine crashes and corrupts the game
history file, frequently the operator will have to simply set the
position using the <a href=#setboard">setboard</a>
command, and then use the settc command to restore
the time control values.
<dl>
<dt> <var><moves></var>
<dd> is moves until the next time control (from <cite>Crafty</cite>'s
perspective, be careful and don't look at the opponent's moves to time
control by accident.)<dd>
<dt> <var><ctime></var>
<dd> is minutes left on <cite>Crafty</cite>'s clock,
<var><otime></var> is minutes left on the opponent's clock.
</dl>
<a name="setboard"></a>
<dt> setboard <var><FEN input></var>
<dd> This command is used to set a chess position up for analysis and
is the preferred way to do this, rather than using the gnu <a
href=#edit> edit</a> interface. It uses a classic
Forsythe-like notation to encode the position and also has provisions
for castling status and en passant capture status.
Standard piece codes p, n, b, r, q, k are used to denote the type
of piece on a square, upper/lower case are used to indicate the color
of the piece (uppercase=white pieces, lowercase=black pieces).
Pieces are entered from the classic chess diagram's orientation
of a8 being the upper-left-hand corner of the board, and this square
is entered first, followed by the remainder of the 8th rank left to
right. To indicate empty squares, use a number between 1 and 8 to
indicate how many adjacent squares are empty. Use a slash (/) to
terminate each rank after all of the pieces for that rank have been
entered. Note that you do not have to account for all 8 squares on a
given rank, although many test suites do this for clarity.
The input k2r/ppp////Q/5PPP/7K/ B
will setup the board position that is given below:
<pre><tt>
-k * * -r * * * *
-p -p -p * * * * *
* * * * * * * *
* * * * * * * *
* * * * * * * *
q * * * * * * *
* * * * * p p p
* * * * * * * k *
</tt></pre>
This assumes that k represents a white king and -q represents a black
queen.
The field after the final "/" should be either b or w to indicate
which side is "on move." After this side-to-move field any of the
following characters can appear to indicate the following: KQ: White
can castle king-side/queen-side/both; kq: same for Black; a1-h8:
indicates the square occupied by a pawn that can be captured en
passant.
<a name="score"></a>
<dt> score
<dd> This command simply gives the positional score for the current
position. This score is from White's perspective, so a + score is
good for White, a - score is good for Black. <cite>Crafty</cite> also
breaks the score down into major categories (from Evaluate()) to
indicate things like pawn structure, piece evaluation, passed pawns,
development, and so forth. Note that some of <cite>Crafty</cite>'s
evaluation is asymmetric, so that if you simply change sides with the
<a
href="#black|white">white/black</a>
command and then enter "score" again, you may get a different value.
This is *not* a bug. :)
<a name="sd"></a>
<dt> sd <var><n></var>
<dd> This command lets you specify a specific search depth limit that
<cite>Crafty</cite> can not exceed. It still pays attention to the
clock, however, so often you will use the <a
href=#st>st</a> command (below) in conjunction with
this if the search is going to take an extended amount of time.
<var><n></var> is the depth (in plies or 1/2 moves) that the
search must reach. Note that if <cite>Crafty</cite> is pondering, it
still honors this limit and will stop a ponder search after this depth
has been completed as well. This is *not* the way to make
<cite>Crafty</cite> play weaker, although this will be covered in a
later section of this document.
<a name="show"></a>
<dt> show <var><category></var>
<dd> This command forces <cite>Crafty</cite> to display additional
information about certain actions it takes. Currently the only
<var><category></var> is book which will make
<cite>Crafty</cite> display information about all the book moves it
found in the database. More is given about this information in the
the <a href="#Opening Book">Opening Book Setup and Usage</a> section
later in this file.
<a name="smpmt"></a>
<dt> smpmt=<var>n</var>
<dd> This command is used to set the number of threads to use on a
machine with more than one processor. For optimal performance,
"<var>n</var>" should be set to the number of processors you have,
although using fewer will reduce the load on your machine. For this
command to work, <cite>Crafty</cite> must have been compiled with SMP
defined. When compiled with SMP enabled, mt=0
effectively disables the SMP code completely.
This command also has two that are closely related:
smpmin and smpmax. Both accept
single numerical arguments.<dl>
<a name="smpmin"></a>
<dt> smpmin
<dd> is used to control the minimum tree depth required at a node
for it to be eligible for parallel searching. IE smpmin
2 says don't split unless at least two more plies are left
to search below this node.<dd
<a name="smpmax"></a>
<dt> smpmax
<dd> sets the maximum for the same idea; that is, smpmax
10 says don't split if more than 10 plies are remaining
below this node.
</dl>
<a name="sn"></a>
<dt> sn <var><n></var>
<dd> This command is similar to the <a
href="#sd">sd</a> command, but instead of setting a
specific search depth, it sets a number of nodes to search. Once the
search has searched this number of nodes (+ maybe one more second of
searching to check the time) the search will exit.
<a name="st"></a>
<dt> st <var><n></var>
<dd> This command lets you specify a specific search time limit for
<cite>Crafty</cite>. Again, this is not the preferred way to set a
time per move, because this limit is absolute and <cite>Crafty</cite>
will never go over this limit, even if it sees that it is losing or
getting mated. Setting the time control with the usual <a
href="#time">time</a> or <a
href="#level">level</a> command is *much* better.
<var><n></var> is given in seconds, although it may also be
entered as mm:ss if preferred.
<a name="swindle"></a>
<dt> swindle on|off
<dd> This command gives you control over "swindle mode." When on, and
playing a game, <cite>Crafty</cite> will try to win drawn endings
(according to the tablebases) if it has winning chances (like KR vs
KB, for example). This will put up very stiff "resistance" to
accepting the draw, while with this mode off, it may be very easy to
draw a position once the tablebases say "drawn." This mode is
automatically turned "off" during <a href="#analyze">analysis</a> or
when <a href="#annotate">annotating</a> a game, and is only used when
actually playing a game against an opponent. If there are <a
href="#egtb">no tablebases</a> then this has no effect on the game at
all.
<a name="tags"></a>
<dt> tags
<dd> This command will simply display the current PGN tags (you can
edit them with the various <a href="#pgn">pgn</a>
commands).
<a name="test"></a>
<dt> test <var><filename></var> [<var>n</var>]
<dd> This command will run a suite of positions (the file must be in
"<cite>Crafty</cite>" format as explained below) and produce a summary
of how many it got right, wrong, etc. It uses the time per move you
set with the (typically) st <var><n></var>
command. The optional parameter [<var>n</var>] is the "early exit"
counter. If <cite>Crafty</cite> finds, and holds the solution move
for n iterations, it will terminate the search. I use this to make a
"<cite>Win at Chess</cite>" run take < 15 minutes, even though the
time per position is set to 1 minute, by setting n to 2. After two
correct iterations, <cite>Crafty</cite> goes on to the next problem.
For absolutely correct results, this is not advisable as it could
obviously change its mind later on, but for performance analysis, this
saves a lot of time.
The test suite contains the following lines: (this is a sample
from my test suite for <cite>Win At Chess</cite>.)
<pre>
title wac299
setboard 1n2rr2/1pk3pp/pNn2p2/2N1p3/8/6P1/PP2PPKP/2RR4 w
solution Nca4
title wac300
setboard b2b1r1k/3R1ppp/4qP2/4p1PQ/4P3/5B2/4N1K1/8 w
solution g6
end
</pre>
<ul>
<li> The title command simply displays this message
in the log file so you can look at the output and figure out which
position it goes with. This is optional, but useful.
<li> The setboard command sets the position as
explained above.
<li> The solution command gives the set of solution
moves (one or more moves that are separated by blanks) and/or a set
of "anti-solution" moves (moves that must not be played to
count the position as correct.) "anti-solution" moves are
simply followed by a "?" character, for example:
<center> solution Bxa7?</center>
<center> </center>
The solution command supplies a set of key moves, and
then starts the search. If, after the search terminates, one of the
key solution moves was chosen (or none of the anti-solution moves
were chosen) the position is counted as correct.
<li> The final line should be end although
end-of-file or EOF will also be detected in this case.</ul>
<a name="time"></a>
<dt> time
CPU | elapsed | <var><values></var>
<dd> In the forms time CPU and time
elapsed, this command controls whether the program uses CPU
time or wall-clock time for timing. For tournament play, it is safer
to use wall-clock timing. For testing it may be more consistent to
use CPU timing if the machine is used for other things concurrently
with the tests being run.
The form time <var><values></var> is used
to set the basic search timing controls. The general form of the
command is as follows:
<center>time
<var>nmoves</var> / <var>ntime</var>
[/ <var>nmoves</var> /
<var>ntime</var>]
[/ <var>increment</var>]</center>
<ul>
<li> <var>nmoves</var>/<var>ntime</var> represents a
traditional first time control when nmoves is an integer representing
the number of moves and ntime is the total time allowed for these
moves.
<li> the optional
[<var>nmoves</var>/<var>ntime</var>] is a traditional
secondary time control.
<li> <var>increment</var> is a feature related to ICS play and
emulates the Fischer clock where <var><increment></var> is added
to the time left after each move is made.
</ul>
Instead of an integer, nmoves can be sd which
represents a sudden death time control of the remainder of the game
played in ntime. The optional secondary time control can be a
sudden-death time control, as in the following example:
<center>time 60/30/sd/30;</center>
this sets 60 moves in 30 minutes, then game in 30 additional
minutes. An increment can be added if desired.
One final example is the following:
<center>time sd/15</center>
which is a simple game/15 setting. This command can also be used to
perform the same function as the "level" command. For example, to set
up a classic ICS 2 12 game, the following would would work:
<center>time sd/2/12.</center>
<a name="trace"></a>
<dt> trace <var><n></var>
<dd> This command is used to make <cite>Crafty</cite> display the tree as it
searches a position. Due to the incredible speed at which this
program can search, however, this is less than useful since it can
swamp any known display driver and make things scroll impossibly fast.
Also note that this command usually is disabled, because
<cite>Crafty</cite> is generally compiled with the -DFAST flag, which
removes the trace output code from the search to make things slightly
faster. You will have to recompile, without the -DFAST, if you want
to use this. Its utility is limited, except for debugging, anyway.
<a name="usage"></a>
<dt> usage <var><n></var>
<dd> is simply a way to modify <cite>Crafty</cite>'s time usage to fit
your tastes. You can "suggest" a time limit with any of the options
discussed previously, but if you use anything other than the <a
href="#st">st</a> command, <cite>Crafty</cite> will
do its best to use time as you suggest, but it also anticipates that
it will save some time by pondering, etc., and will therefore be more
aggressive at trying to use time. If <var><n></var> is a
positive integer, it is taken as a percentage and <cite>Crafty</cite>
will compute the time limit it thinks is appropriate for the current
clock settings, then increase this limit by this percentage (50 would
make it take 1.5 times longer than normal.) -50 would make it take
1/2 the time it would normally take.
<cite>Crafty</cite> adjusts the usage internally based on time
left, opponent's time left, how quickly or slowly the opponent is
moving, etc. Further modifying things with this is dangerous, but
possible.
<a name="whisper/kibitz"></a>
<dt> whisper/kibitz <var><n></var>
<dd> These commands are used to control what <cite>Crafty</cite> will
whisper or kibitz on a chess server. The options are (1) only
whispers or kibitzes mate announcements; (2) adds time, score, depth
to the previous option, but no PV or moves. (3) adds the PV. (4)
adds book move information to the output. The remaining two options
generate a lot of output and should be used with caution. (5)
displays the PV after each iteration completes. I use this when using
my custom interface to let <cite>Crafty</cite> observe/comment on
games in progress on ICC. Noise can be used to prevent shallow
searches from generating output and keeping "noise" down on the games
being watched. (6) basically will whisper/kibitz nearly everything
you see on the console from a search, each PV when it changes, fail
highs and fail lows, etc. A significant amount of output that should
be carefully weighed before turning it "loose."
<a name="xboard"></a>
<dt> xboard
<dd> This command turns on <cite>Xboard</cite>/<cite>WinBoard</cite>
compatibility mode, and makes <cite>Crafty</cite> behave somewhat like
GnuChess. This is designed to be used *only* when <cite>Crafty</cite>
is interfacing with <cite>Xboard</cite>/<cite>WinBoard</cite>.
<cite>Crafty</cite> will not work with these two GUIs without this
option, and really won't work very well with this option if it is not
connected to one of them.
</dl>
There are other commands that are not documented. They are part of
the xboard protocol and really should not be used by the normal user.
You can find all the commands in option.c should you be interested.
<hr>
<a name="Opening Book"></a>
<h2>Opening Book Setup and Usage</h2>
<cite>Crafty</cite> uses two pre-compiled opening books, called
"<code>book.bin</code>" and "<code>books.bin</code>".
The file <code>book.bin</code> is usually build from a large text
file containing PGN games, often taken from collections of GM games.
Building <code>book.bin</code> is a simple exercise and requires
nothing other than the raw input file you are going to use. Generally
this will be either <code>medium.zip</code> or the set of four files
<code>large</code>[1-4].<code>zip</code>, all of which are stored on
the ftp machine <a
href="ftp://ftp.cis.uab.edu/pub/hyatt/">ftp.cis.uab.edu/pub/hyatt/</a>.
To create the file <code>book.bin</code>, you need a PGN game
collection that is a *real* PGN-compliant file. Supposing this file
is called "large.pgn" you would use the following command:
<center>book create large.pgn <var><m></var>
[<var>n</var>] [<var>wpct</var>]</center>
The only thing you have to supply is <var><m></var>,
a number indicating how many moves from each game are to be stored in
the book.bin database. I typically use 60, which stores the first 60
moves from each game. Increasing this number slightly increases the
probability that <cite>Crafty</cite> will stay in book longer, but it also
increases the probability that it will follow a game too far, so that
it begins to reach positions where the move actually played might not
be the best move, letting it fall into a bad hole. Increasing this
also increases the size of the database as well.
You can decrease the size of the book, and also reduce the number
of ugly moves by specifying [<var>n</var>], which says that unless a
move is played in at least <var>n</var> games, the move is discarded.
This will substantially decrease the size of the book.bin file, and
also eliminate single game moves that often have significant errors or
blunders.
You can increase the quality of book lines by specifying
<var>wpct</var> which is the "winning percentage". This is specified
as a percentage of lost games, and is used to discard moves that led
to mostly losses. A safe value is 50, which says that if a particular
opening move didn't win at least 50% as many games as it lost, the
move is culled. A value of 100 would mean that moves are culled if
they produced more losses than wins, and is really a strict criterion.
After creating <code>book.bin</code>, you need to create
<code>books.bin</code>. This is a small version of
<code>book.bin</code>, which is intended to give you more control over
the moves/openings <cite>Crafty</cite> will play. This is usually
built from the file <code>start.pgn</code> on the ftp machine, but you
can modify this file to suit your taste easily. To build
<code>books.bin</code>, you use the following command:
<center>books create start.pgn 60</center>
Again, 60 is what I use, but none of my <code>start.pgn</code>
lines go anywhere near that many moves. The main point here is that
in <code>start.pgn</code>, you can append a "!" to any move you want,
and when it is <cite>Crafty</cite>'s turn to move for that color, it
will play from the set of moves with "!" if there are any, ignoring
the rest of the book moves. If you only want it to play 1. e4 as
White, you would just enter the short game:
<center>[Crafty only plays 1. e4]
1. e4!</center>
and you are finished! You can enter as many as you want. If on
the other hand there is a move you don't want <cite>Crafty</cite> to
play, then follow that move with a "?" and it will never play it.
Moves in <code>books.bin</code> that are not flagged with ! or ? don't
have any influence on <cite>Crafty</cite>'s choice at all.
Here's how the files are used. When searching a position,
<cite>Crafty</cite> first enumerates the set of moves it can find in
the opening database. It then does the same for the
<code>books.bin</code> database, and performs a "merge" operation to
combine the ? and ! flags... The purpose of the
<code>books.bin</code> file is to give you a small database that you
can easily modify, rebuild, and repeat this process over and over.
Since it takes a fraction of a second to completely rebuild this file,
it is very easy to modify this to control what <cite>Crafty</cite>
will play, without having to rebuild the large database.
One important characteristic of the PGN input is the Result tag
must be specified in most of the lines, because <cite>Crafty</cite>
counts wins, losses and draws for each book move and uses these counts
with some of the book selection options given below.
<h3>How the flags are used.</h3>
The ! and ? flags should only appear in the small
<code>books.bin</code> file, although there is no reason why they can
not appear in the large file as well. For this discussion, it doesn't
matter since <cite>Crafty</cite> takes the moves from both files and
"merges" the flags, etc., into one entry for each move.
Quite simply, if any book legal book move has a ! flag, then
<cite>Crafty</cite> will only play moves from the set of moves which
all have the ! flag. If a move has a ? flag, then that move is
immediately removed from the set of possible book moves. If the only
legal book move has a ? flag, it will not be played as a book move and
<cite>Crafty</cite> will simply pretend that it found no book moves
and will execute a normal tree search. Pretty simple.
<h3>
How to control the frequency of opening move selection.</h3>
A new feature in version 15.15 and beyond allows you to append a PGN
comment to any move in a text file used to create
<code>books.bin</code>, of the form {play
<var>nn</var>%}. This will force the move it follows
to be played that percentage of the time, regardless of the normal
book-ordering values based on frequency and so forth.
Note that {play 0%} will not prevent a move from
being played at all, as this will look just like a move with no
percent specified. To avoid playing a move, use the ? flag.
<h3>
How does <cite>Crafty</cite> choose book moves?</h3>
<cite>Crafty</cite>'s book selection algorithm depends on two
user-defined values that can be set at any time during a game:
<center>book random <var><n></var></center>
<center>book width<var> <w></var></center>
The selection algorithm first finds the set of legal book moves
as above. This set will either be all ! moves, or will have no !
moves in it. This set is then sorted based on the setting of book
random. Here's the options:
book random 0. This is a special case for book
selection. <cite>Crafty</cite> simply takes the set of book moves,
and searches only these moves using a normal alpha/beta search, but
with a shorter than usual time limit. It then plays the move that
produces the best search value. This has one serious disadvantage in
that there is no randomness to this at all. It will always play the
same move in the same position, unless the evaluation is modified, or
the time per move is different enough to let the search find a
different move from the book move set.
book random 1. This enables a random form of
book move selection, but you have a lot of control over how moves are
randomly chosen. The moves are ordered, based on 4 parameters:
frequency of play, win/lose ratio, static evaluation and learned
results. Normally these are factored into the value used to sort the
moves, based on default settings that you can modify by using the
command bookw <var><option> <N></var>,
where <var><option></var> should be "freq", "ratio", "eval" or
"learn". <var><N></var> should be a number between 0 and 1.
<cite>Crafty</cite> finds the min and max values for each of the
4 parameters, and then maps this into the range 0-1000 for each
parameter. Each parameter is multiplied by the corresponding "weight"
you have assigned, and this is used as a value to sort the moves from
low to high. Note that the first sort value is always the "play
percent" to move them to the top of the list. For moves with equal
"play percent" values, the normal sort-value is used as the
second-level sort variable (if no moves have a play-percent, then this
second-level variable is the only one used, of course.)
Once <cite>Crafty</cite> has sorted the moves as given above, it
next excludes any book moves which have 0 wins. This culls the odd
lines where a player chose a bad line and lost quickly. With zero
wins, it will never be chosen, although <cite>Crafty</cite> will
happily follow it from the other side. :) This is not anywhere near
perfect, however, because an opening could have 1 win and 19 losses
and that still would stay in the list.
If a move has a learned value of > 100, this move is elevated
in priority to that of a ! move, since it appears to win material
instantly. If a value is < -100, it is given a ? since it appears
to be a lemon.
After this, the setting for book width
<var><w></var> is used to keep the first
<var><w></var> moves in the list, after the above culling
has been completed. The smaller you make <var><w></var>
the less randomness you get, but the chance of <cite>Crafty</cite>
popping out a really bizarre book move gets smaller as well.
After sorting, the final step is to fold in any mandatory "play
percent" values. What this step does is that it finds all the moves
in the "playable move set" just computed, which have no
percent-to-play value set. It sums the sort-values for these moves,
then adjusts the sort-values for the other moves so that their
percentages will be honored.
Once this has been done, <cite>Crafty</cite> simply uses the
"sort value" for each move to compute a total for all moves. It then
generates a random number between 1 and this limit, and chooses the
move that this probability distribution matches. This will certainly
evolve over time as new ideas are developed.
For my play on ICC, I use book random 1, and
book width 5 normally, although for titled players
this is reduced to book width 3. For computers, I
reduce this further to 2, to try to play reasonable
openings and cull the gambits and things that don't work out.
<h3> How does book learning work and how can I share learned
results?</h3>
1. *All* of <cite>Crafty</cite>'s "learned knowledge" is in the
<code>book.bin</code> file. It keeps the learned value and learned
count right in the book file for speed. You can't modify it, although
<a href="#show">show book</a> will make
<cite>Crafty</cite> display all the book stuff before it makes a move.
2. the <code>book.lrn</code> file has two purposes: (a) to
serve as a log for your prying eyes, so you can see what it's learned,
against whom, and what the score was that got its attention in the
first place. The values on the end of each book line, inside the {}
characters are as follows:
<center>{value, depth, rating_difference}</center>
<dl>
<dt> value is the value of the "key" search that
comes from the first 10 moves out of book. It's in centipawns, and +
is good - is bad.<dd>
<dt> depth is the depth the search reached at this "key"
position, the deeper the search, the more the value is "trusted."<dd>
<dt> rating_difference is <cite>Crafty</cite>'s
rating - opponent's rating a negative value means pay more attention
to the score since the opponent is better than <cite>Crafty</cite>, a
positive value means to take the score with a grain of salt, because
the opponent was weaker than <cite>Crafty</cite>.<dd></dl>
You can delete this file at any time, and it has no effect on
learning. As I mentioned, the learning takes place in
<code>book.bin</code>... this is mainly for you to peek at if you are
interested. However, this is the "portable learning data file" also,
and can be given to others to import into their <cite>Crafty</cite>,
where it will affect the opening book just like their
<cite>Crafty</cite> had played the openings and got the same scores.
There are two ways to use such "lrn" files:
1. <a href="#import">import</a>
<filename> will read <filename> and
import the knowledge therein into your book.bin. Since I use the same
learning code as is used when playing games, this information also
gets appended to *your* <code>book.lrn</code> file as well, so that
your <code>book.lrn</code> always reflects *everything* your program
has learned, so long as you don't ever remove this file. It would be
a mistake to use this command on your own book.lrn file, because the
things would get counted twice, which might or might not be a good
thing.
2. <a href="#import"import</a>
<filename> clear will read <filename) and
import the knowledge as above, but first clears *all* learned results
from book.bin. you will want to do this if you import my book.lrn,
*and*, you have contributed to my book.lrn data by sending me yours.
I'll take care of eliminating duplicates if you screw up in what you
send me, but once you send me something, you run the risk of getting
it "back again" later. This is going to be a problem until everyone
gets used to sharing something that is rather "vapid" like this
"learned info" is...
Other than that, we are now "open for business"... when the urge
strikes you, email me your .lrn file, I'll keep a large master here
and update it on occasion. Probably the best thing to do is to send
me your .lrn and at the same *instant* delete yours. This will
capture anything learned *after* you send me the file, but you'll get
it all right back with the next version of book.lrn that I
distribute.
After getting this new book.lrn back, here's what you should
do:
1. rename your old <code>book.lrn</code> to something else.
I'll call it "book.lrn.old" here.
2. copy my blearn.dat to your machine, but *do not* put it in
the directory with your <code>book.bin</code> and
<code>books.bin</code> file because it will get confusing very quickly
if you do. Put it somewhere else, because you are going to remove it
quickly anyway. I'll call it new.lrn for this example.
3.<pre>
import new.lrn clear
import book.lrn.old</pre>
and you are ready to rumble again. The first command clears the
learned values, sucks in my new learn file and updates everything.
the second command re-learns everything you've learned since you sent
me the last book.lrn file. After doing this your book.lrn will have
my .lrn stuff, plus your old.lrn stuff, just waiting to be sent to me
again...
If this is confusing, I can probably add an automatic command to
do all of this by renaming book.lrn, etc. Hopefully this is not too
error-prone for the time being anyway...
<h3>
What is this new Position Learning I've heard about?</h3>
<cite>Crafty</cite> now has a "permanent" hash table that is kept from
game to game. A position gets into this "hash file" when
<cite>Crafty</cite> executes a search and the search value is
significantly lower than the last search value.
When this happens, <cite>Crafty</cite> stores the current
information for this position in the permanent hash file, which can
hold up to 65536 positions. Once it fills up, the positions are
replaced on a FIFO basis, always keeping the most recent 64K entries.
Each time <cite>Crafty</cite> starts a search, the
positions/scores from this file are stuffed into the normal
transposition table, and used during the search just like any other
table entry. Here's how it helps: In a game that was played, the
following moves and scores were found by <cite>Crafty</cite> (playing
White):
1. Ng5 (+.277) h6 2. Nh7 (+.321) Kg8 3. Qh5 (+.133) Qg7 4. Ng5
(-2.122) hxg5
So, the knight got trapped at h7, and at move 4
<cite>Crafty</cite> discovered that this is gross and "learns" this
result/position.
We play the exact same game again: except that two things can
happen here.
1. It might be that g5 is the *only* square the knight can move
to here, which means this whole thing is forced. The first search
would find 1. Ng5 (-2.122) if the search can reach 8 plies deep, which
happens even in 5 second games. It's learned that Ng5 is bad. It
stores *this* position in the permanent hash file also, and the next
time you try this same trap, it will discover 4-5 moves earlier that
if the knight gets to g5 it is in trouble. Each game will diverge
from the first game 3-4 moves earlier. Simple and effective.
2. Ng5 might not be forced, and if not, it knows Ng5 loses a
piece for a pawn, so it will promptly play something else, which is
exactly what is desired.
This is implemented with two (count 'em, two) files. One file
"<code>position.bin</code>" is a binary file that contains the hash
table entries, and it right at one megabyte in size, *max*. (16 bytes
per hash entry X 65536 entries = exactly one meg, but I have 8 extra
bytes for the FIFO queue implementation and to see how many entries
are currently in the file if it is not full.
The second file is "<code>position.lrn</code>" and is, you
guessed it, a file that can be shared with others, just like book.lrn.
It contains all information needed to reconstruct the position, the
score, the depth, etc. and also included the pgn tags for who was
what color and when the game was played...
This data can be imported with the new <a
href="#import">import</a> command (the old book learn
<filename> is no longer around) which will import either
book.lrn type data or position.lrn type data and can tell them apart
without your having to do anything. The [clear]
option is still there, should you want to use it, and simply removes
the position.lrn and position.bin files before starting the import
process for position learning.
This can be turned off, if you like, by checking out the <a
href="#learn">learn</a> command, which gives you the
ability to turn off book learning (as it presently works), position
learning, and the next book learning stage which will add to the book
in addition to learning which book lines are good and bad.
<h3>
What is this new "result" learning?</h3>
Result learning works just like normal book learning, except
that if <cite>Crafty</cite> is checkmated or resigns, it will step back
through the book line to find the last point where it had
more than one move to choose from. It will flag the move it
chose as "never play again".
This handles the case where the first ten non-book moves produce
reasonable scores, but the position is one that <cite>Crafty</cite>
simply can't handle very well. If it loses such a game, it will still
vary the next time this opening is played, as otherwise it would
possibly repeat the same opening, and would certainly repeat the
remainder of the game.
All three learning modes are turned on by default, although any
of them can be disabled with the appropriate command option to <a
href="#learn">learn</a>.
<hr>
<address></address>
Last modified: Sat Dec 20 01:41:05 EST 2003
</body> </html>

Subversion Repositories Games.Chess Giants

Games.Chess Giants/engines/Crafty/readme.html @ 33 – Rev