Welcome to Averno
home page
by José
Carlos Martínez Galán
Latest news:
|
|
Information on this page
updated (20/03/2005). |
|
|
New version 0.81, that
played in CCT7, performing very well against vastly superior hardware. |
|
|
I’ve uploaded some pictures
of my recent wedding (20/09/2003): Roger Davis has created a beautiful
gallery for me here. And now, little José David is welcomed to the world!!!
Pictures are here. |
What's
new in v0.81
|
|
Recapture extension margin
modified to extend more when closer to the leaves. Fixed bug in Multi-Cut
pruning. Developement eval now
measures differences, not absolute numbers. Closed opening eval
improved. Root search: modification
on increments in case of fail high / fail low. Pondering: number of nodes
between controls reduced. Pondering: keep searching
if ponder hit (no need to restart and use hash info). Several internal
programming improvements. QSearch: general futility
pruning against alpha improved. Qsearch: use SEE more aggressively. Distance to mate used for
pruning. No need to save PV between
moves: the hash table has enough information. Mobility eval improved. Small speed optimizations. Old bug in move generation
(!!) fixed. Polling input improved to
work better under Winboard GUI. Other prunning heuristics
in regular search. Eval: knight vs bishop
balance improved. Book: learning bug fixed. Time management improved,
specially with ponder on. Eval: King safety
improved. New SEE, faster and
cleaner. |
|
|
Many bugs fixed. |
Download:
|
|
|
|
|
Averno_Book081.zip
(backwards compatible; built after Carlos Pesce’s pgns –thanks Carlos-) |
|
|
BookBuilder.zip
(this is obsolete, I’ll release the new one soon) |
How
to run Averno:
Averno is designed to run under Winboard. It supports protocols 2 only. It works
fine under Arena 1.02.
Averno070.zip includes two files:
|
|
Averno070.exe |
|
|
Averno.per |
Averno.per is a personality file. It's meant to contain configuration
information but it’s not needed anymore. I recommend to copy it to a new file
(let's call it MyPersonality.per) and modify it. Averno will use the
default settings for every parameter not found in the configuration file, so
you actually only need to include those you want to modify.
The
most important personality file parameters:
|
|
TABLA_HASH n
:
Sets the has table size to a maximum of n Megs. The actual amount of allocated
memory will be smaller than that because Averno uses certain fixed (3, 6,
12...). |
|
|
NALIMOV_PATH path : You can indicate Averno to use
Nalimov's endgame tablebases. Set here the path to them. |
|
|
EGTB_CACHE_SIZE n
: Use this parameter to indicate the number of Megs to use
as EGTB cache. |
|
|
ICS n
: n = 0 to disable and n = 1 to enable. This parameter is
used to kibitz information in chess servers. In this version, there's no
kibitz file (it's hard coded) so if enabled, Averno will only kibitz against
"known" opponents: Hossa, Tinker, Postmodernist, Sjeng... In future
versions, there will be a kibitz file where you can chose the opponents to
kibitz. |
|
|
LIBROB_FICH path
: Different personalities can use different books. Use this
parameter to tell Averno what file must be used as white book. If ommited,
Averno will still use Averno070B.libro. Note that if the files are in the
same directory as the program, the full path is not needed. |
|
|
LIBRON_FICH path
: Use this parameter to tell Averno what file must be used
as black book. If ommited, Averno will still use Averno070N.libro. Note that
you can share a file for different personalities. For exaple, you can use MyBook.white
for all your personalities and MyBook_1.black, MyBook_2.black,
etc. |
|
|
APRENDER n
: If you want Averno to learn from its own games, set this
parameter to n > 0. n means the number of moves after the last book move
that Averno will try to add the the book file. Note that the book files will
be modified so it is a good idea to save a copy of the original files. |
|
|
AP_SEGUNDOS_MIN n
: If APRENDER is > 0 (learning enabled) this parater defines
the minimum number of seconds of search Averno will consider enough to learn.
Apart from adding moves to the book, Averno will remember the number of wins
/ losses for the moves in the book, so that it will eventually try to follow
lines where it gets good results. For example, if n = 30, when the program is
to move (book or learned move, or unknown position), only when it allocates
30 or more seconds for that move will the learning algorithms be triggered. |
|
|
SELECTIVIDAD_LIBRO n
: This parameter tells the program how random the book move
selection should be. n is expressed in centipawns. The bigger, the more
random the move selection is. Set to 0 (zero) to force the program to always
select the book it considers best. |
|
|
LIMITE_ACEPTACION n
: If APRENDER is > 0 (learning enabled) this parater
defines the lower score (in centipawns) a searched move needs to be included
in book so that the next time the program reaches that position, it will play
the move without searching. For example, if after searching a move the score
is -0.25, n must be >= 25 so that, the next time the program plays that
position, the move will be played immediately. Note that there other factors
the program considers rearding learning, such as the total time spent in a
search (the program won't play immediately if the move was searched during 30
seconds and now it has 45 to think), the result of the game (wins encourage
and losses penalyze), the score after a search affects also to the previous
position, etc... |
When you run Averno, type the name of the personality file (without .per)
you want to use after Averno.exe. For example (in Winboard.ini):
“Averno
MyPersonality” /fd=c:\chess\averno
How
to build a book for Averno:
Averno uses
two book files, one for white and one for black (these names are only an
example, any name is acceptable):
|
|
AvernoB.libro |
|
|
AvernoN.libro |
Note that B and N stand for Blancas and Negras, which is the spanish for White
and Black.
To build a book for Averno you will use two pgn files (without comments or
variations, just the moves), which names will be:
|
|
Blancas.pgn |
|
|
Negras.pgn |
Blancas.pgn will contain games in which white plays good chess (I hope!)
and Negras.pgn the same for black. Of course, you can use the same file,
copying and renaming it.
Additionally, you can use the signs ! and ? in the games. Place a !
after a move you want the program to play more often, and a ? after a
move you don't want the program to play at all.
When you have your pgn files ready, unzip Av_BookBuilder.exe and copy it
to same folder where the pgn's are. The pgn must be without comments,
variations, etc.
Run Av_BookBuilder.exe and type:
ConstruirL n
where n is the number of full moves to use from every game. Typical is
20 or 25.
The program will parse the pgn files and build the book in memory. After the
process is finished, the program will automatically save it to disk. Two files
will be created: AvernoB.libro and AvernoN.libro. Rename them at your wish and
copy them to the directory where the program will run in.
Implemented features:
Averno supports all winboard features needed to play, plus: Analysis mode, edit position, EPD processing using EPD2WB by Bruce Moreland, pondering.
Email me at smyslov@hotmail.com