On Tue, 8 Sep 2009, Michael Petch wrote:
I use 8 cores on Linux with no real issues.
I have run a few 4-ply match analyses (no rollout yet) on a 16 cores
machine (CentOS 5, x86_64) with no obvious problems.
Time will tell on how well Gnubg scales on a large number of processors.
On the o
On Wed, 9 Sep 2009, Michael Petch wrote:
I find setting threads to 2 times or 3 times the core cpu count makes a
difference to help the starving issue. Curious if you see any difference. My
choice to up the thread count to 48 was not arbitrary (3 times 16). Can you
let me know if 24 threads on 1
On Thu, 10 Sep 2009, Michael Petch wrote:
On 10/09/09 4:33 PM, "Ingo Macherius" wrote:
My benchmarks with 4 cores showed that the cache brings about 10% improvement
And got this results :
8 thread, 0mb cache, 10 trials : 44.83 (seconds) mean, Std Dev of .18
8 thread, 21mb cache, 10 trials
On Tue, 22 Sep 2009, Massimiliano Maini wrote:
Other comments:
labels on the toolbar are too long [...]
Yes! Keep it keep working on a 1024 pixels screen width, please. This is
no longer the case with the recent changes
___
Bug-gnubg mailing li
On Sun, 4 Oct 2009, Timothy Y. Chow wrote:
If I type "save game" after the new game has started, it saves the new
game and not the one that just finished.
"save match" will save a money session.
There is a "analyse session" synonym to "analyse match" but no
such thing for save or load...
On Tue, 6 Oct 2009, Timothy Y. Chow wrote:
Now suppose I want to review the session. I tried
load match savedmoneysession.sgf
which seemed to work, but now I don't know how to go back and look at the
games I played. If I type "first game" it says "No game in progress."
Reviewing it from
On Sat, 31 Oct 2009, Yoshito Takeuchi wrote:
I rollout with 'Stop when result is accurate'.
But, Standard error enough under set value, non stop GnuBG rollout.
I use latest version GnuBG ( Today I compile gnubg from source code).
My setting.
Settings-rollouts
set 'Trials:' 12960
Check 'Stop w
On Sat, 16 Jan 2010, Christian Anthon wrote:
As usual an example makes the problem easier to track...
On Sat, Jan 16, 2010 at 1:47 PM, Michael Depreli
wrote:
Since updating to build 20091230 I'm having problems with cmarked rollouts
stopping
short of the set trials for some moves.
I'm using
On Sun, 17 Jan 2010, Philippe Michel wrote:
[...] I get the same kind of behaviour all the time [...] It doesn't
particularly shock me [...]
On the other hand, there are other problems in the stop on JSD code :
This :
/* it was stopped, catch it up to the other moves and r
On Wed, 10 Mar 2010, Timothy Y. Chow wrote:
What I want to do is to set the seed
for the random number generator. If I add the command "set seed 123" to the
above text file, e.g., just before "analyse rollout move", then gnubg will
report "Seed set to 123". However, this is apparently a lie,
In positionid.c, there is this comment :
* An implementation of the position key/IDs at:
*
* http://www.cs.arizona.edu/~gary/backgammon/positionid.html
*
* Please see that page for more information.
That URL is no longer available. Is it what is now in :
http://www.gnu.org/manual/gnubg/h
On Fri, 25 Jun 2010, Christian Anthon wrote:
I've added support for pasting XG ids in GNU backgammon. I didn't have
too many test cases, so please test the code and report back any
problems. Beavers are not supported.
There seems to be a problem regarding the Jacoby rule. Using the xgid from
h
There is a problem with www.gnubg.org's home page. It displays an error
message :
--!>
Page headers already sent
The page headers have already been sent out in
/home/www/20/27/gnubgorg/index.php line 1. This could cause Nucleus not to
work in the expected way.
[...]
Inner pages like the wik
On Sat, 13 Nov 2010, Adi Kadmon wrote:
Hello all
I'd like very much to have the following information:
(1) What is GNUbg's formula for calculating "EPC"/"Trice count" of bearoff
positions?
(2) Is that formula, or any approximation thereof, practically usable by a
human player during a real ga
On Sat, 27 Nov 2010, pierre zakia wrote:
What are the optimal settings for gnubg installed on a MacBook Pro 15" Intel
Core i7 2.66 GHz (April 2010 model) ?
The build is Version 0.9.0, running on Snow Leopard (10.6.5) without any
problem;
I have played changing figures in Settings/options/othe
On Sun, 29 Aug 2010, dankwart.plattner wrote:
I have some old sgf files created with gnubg 0.14 mingw. I recently opened
both with GNUBG 0.90 (my version: 0.90-mingw 20100220) and with a text
editor (see attached zip file).
The first line (for the first game) reads:
(;FF[4]GM[6]CA[UTF-8]AP[G
On Mon, 29 Nov 2010, pierre zakia wrote:
I did the test again with only X11 running (no Airport, no Ethernet connection).
The results are:
1 thread: 52 000 000
2 thread: 90
3 thread: 104
4 thread: 40
5 thread: 40
6 thread: 39
7 thread: 39
8 thread: 39
3 thread is still the winner. Why ?
On Mon, 29 Nov 2010, Louis Zulli wrote:
I have my own SL compiled binary on this 8-core Nehalem Mac Pro. As soon
as I'm able, I do some testing and report back.
If if works satisfactorily, would it be possible for you to build an
installable package like the one for 10.5 on www.gnubg.org and
On Tue, 14 Dec 2010, Peter Nikolov wrote:
Generally and theoretically speaking- I don?t think that it?s so
difficult artificially to create/construct an example with particular
position on the desk, when is not correct to resign before to rotate the
dices, but after turning the cubes the resig
On Tue, 27 Apr 2010, Øystein Johansen wrote:
I've uploaded a quick and dirty patch to neuralnetsse.c which should do the
trick.
http://www.gnubg.org/media/nnsse.patch
A binary weights file with the better net is found here:
http://www.gnubg.org/media/alternative-gnubg.wd
It should be quite sim
On Mon, 27 Dec 2010, Wilson Amaral Jorge wrote:
I need understand why there is a divergence so great between the numbers
showed in the (Analyse)Market Window and Evaluation/Hint. Better
explaining: in a particular game -2/-4, cube 2 with the trailer, the
Market Window showed Cash Point 50% and
On Sun, 2 Jan 2011, Thomas A. Moulton wrote:
set gnubgid 3wYAANjsGwEIAA:MAFpACAA
hint 1
in my local build gives a crash
(gdb) where
#0 MakeInt (pbc=, nPosID=3272738968,
arProb=, arGammonProb=0xbfeb9bd0, ar=0xbfeb9e10,
ausProb=0x0, ausGammonProb=0x0) at bearoff.c:834
#1 GetDistCompr
On Fri, 7 Jan 2011, Ian Shaw wrote:
To the developers,
This is a very common mistake, caused because most Windows programs
retain the settings when OK is pressed. I don't know the default
behaviour on Linux or Macs, but I think gnubg would be more user
friendly if it saved settings by defaul
Is there any reason why getLuckRating(), in analysis.c, uses fixed
constants rather than what is set in Settings|Analysis ?
___
Bug-gnubg mailing list
Bug-gnubg@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-gnubg
If you use gnubg on a Windows 7 machine, could you take a look at this :
http://savannah.gnu.org/bugs/?32052
and see if you can reproduce the problem or offer suggestions on where to
look for details about this kind of error ?
___
Bug-gnubg maili
I have tried to use the patch to SSE code that Oystein posted here about
one year ago to compute the pruning nets' outputs (using the current
weights file padded with zeroes at the right places).
It works and is a little faster than the current way, so I suppose it
could be used as is. The cos
On Fri, 4 Mar 2011, Michael Petch wrote:
I've committed a change that should fix it.
You don't say haw you diagnosed it, so maybe you already know this, but
FWIW on linux (rhel6) the original code looped repeating these messages :
(gnubg:24628): Gtk-CRITICAL **: gtk_statusbar_pop: assertion
Regarding Macs, are the PowerPC ones still relevant ?
Would replicating the SSE code and transcribing it to Altivec useful or
just a waste of time ? (assuming it's relatively straightforward, if not I
suppose the answer is the latter :-).
___
Bug-
On Sun, 20 Mar 2011, Michael Petch wrote:
I had never thoroughly dived into the contents of a GNUBGID but I was a
bit surprised when I didn't see either beaver information or Jacoby. I
was wondering if there is a reason this information is not included? If
there is a reason would someone be kind
It looks like there is something wrong in the recent GUI changes. The
Crafword toggle in the GUI is apparently always off (except just after
editing it to on).
For instance, load a match with a Crafword game and go to this game ; at
the top of the move list panel, there is, as expected, "Game
In the form for submitting bugs at savannah.gnu.org, the operating systems
list is seriously dated, with entries for various Windows 9X but neither
Vista nor Seven.
The Unixes do not look very relevant either (I guess what would be needed
is rather something like "Linux, binary install from a
On Mon, 4 Apr 2011, Michael Petch wrote:
For the moment I have added Windows Vista and Windows 7, as well as the
OSX releases that we have mad binaries for.
Did you add the OS type and processor type fields too ? Shouldn't they
have a "any", "N/A" or something like that choice ? (and ideally
On Wed, 23 Mar 2011, Guido Flohr wrote:
I write a client for FIBS, and I want to use SGF as the internal file
format.
I need a place where I can store some state information. Is it okay to
use the MI property for that?
And what about namespacing? Should I prefix my private tags or rather
writ
On Fri, 13 May 2011, Michael Petch wrote:
Posted on BGO by Leo:
What is the effect of stopping a 0-ply rollout, increasing the
number of plies, say to 2-ply, and then restarting the rollout, for
example, just for the two top candiate plays?
The rollout carries on for these candidates
I don't know if this is important (I don't use python and usually compile
with --without-python), but I get the warnings below on gnubgmodule.c :
gnubgmodule.c:54:3: warning: array index of '1' indexes past the end of an
array (that contains 1 elements)
[-Warray-bounds]
PyTuple_SET_IT
On Fri, 13 May 2011, Michael Petch wrote:
On 13/05/2011 12:40 PM, Michael Petch wrote:
That's not good. I'll look at this. I use Python(and have done work in
gnubgmodule.c) but don't think I build with -Warray-bounds . I assume
you have forced -Warray-bounds on for your builds? I hadn't seen t
On Sat, 14 May 2011, Michael Petch wrote:
Okay Guys,
This is rather comical. First the directory name was a red herring - it
was not the cause (that part works fine). Batch analysis does not work
properly if you turn off "Luck" analysis (Settings menu/Analysis/"Luck"
checkbox)! Leo sent me hi
On Thu, 26 May 2011, Michael Petch wrote:
I believe Philippe Michel mentioned a 30-40% gain. I'm seeing 36%-40%
depending on the variables but 38% would be about the average. I believe
most of Philippe's changes provide the bulk of the performance increases.
When I wrote 30-4
On Tue, 31 May 2011, Ian Shaw wrote:
Gnubg has a larger problem in it's rollout reporting.
The original is on the left. Look how messy it is.
On the right I have tweaked it to reduce redundant information, and
include the suggested JSD information. I think the primary data - the
rollout re
I recently had the opportunity to try to run gnubg on a 48-cores machine
(running RHEL6).
This didn't work well when the number of threads became high : at around
25-30 it hanged, usually almost immediately.
I replaced the locking code (for the evaluation cache only, not the
multithreading i
Until recently, loading the neural net weights from the text file didn't
work in locales where the decimal separator is a comma rather than a dot.
This should be now fixed. Considering this, is there any reason to keep
using a binary weights file ? It loads faster but in practice the
differenc
With modern multicore machines, it has become practical to use 4 ply for
match analysis. I think it would be useful to create a predefined setting
for it.
Moreover, the current settings are :
world class : 2 ply, normal move filter
supremo : 2 ply, large move filter
grandmaster : 3 ply, normal
On Mon, 1 Aug 2011, Michael Petch wrote:
I have discovered a serious problem with "Swap Player". Swap Players
works if you are dealing with games, and moves within games.If on the
other hand you paste a GNUBGID (Where a player is on roll) and Do "Swap
Players" with a position, it doesn't work pr
On Wed, 3 Aug 2011, Michael Petch wrote:
On 02/08/2011 12:44 AM, Philippe Michel wrote:
Handling of these single-position games should be better now, although
there is still a display problem (in 2d mode only) where the dice
colour changes but they stay on the now wrong side of the board
On Wed, 3 Aug 2011, Michael Petch wrote:
[...] the proper display of the board when importing XGID's (reported by
Leo Bueno on BGO).
What the proper display is may not be that obvious. If the imported
position has the top player on roll (as displayed), your new code will
swap players anyway.
On Thu, 4 Aug 2011, Michael Petch wrote:
Secondly what we don't know is whether the output of the board is
showing the player on the bottom as being clockwise or counter
clockwise. If we have a bit for this in an XGID/GNUBGID we'd be able to
render that in either mode.
That would risk the same
On Tue, 27 Sep 2011, Guido Flohr wrote:
I did. But then I have to start a match, and I don't get a chance for
starting the rollout before the first roll. Until then
Analyze->Rollout->* is insensitive.
After you start a Human vs. Human match, you can set one of the players on
roll (by clicki
On Fri, 14 Oct 2011, Michael Petch wrote:
On 14/10/2011 4:07 AM, Ian Shaw wrote:
2) Why is "Cmark" called "Cmark" and not simply "Mark"? What
information does the "C" convey?
I was curious about this naming convention as well. It has confused some
people in the past. They thought CMARK wa
Currently, dice generation code for the Mersenne twister random number
generator looks like this :
dice = 1 + (int)(6.0 * genrand_int32(&rngctx->mti, rngctx->mt) /
(0x + 1.0));
that is, the rng provides a random integer between 0 and 2^32-1 and it is
mapped to a value between 1 and 6
On Fri, 9 Dec 2011, Mark Higgins wrote:
I took a look through eval.c but it's a bit daunting. :)
The attached graph may be useful when trying to understand gnubg's
evaluation code.
pprof1742.0.pdf
Description: Adobe PDF document
___
Bug-gnubg mail
In lib/sigmoid.h, there is some code to calculate quickly exp(x) including
the statement :
i = (int)x1;
In lib/neuralnetsse.c, the corresponding SSE code is :
i.i = _mm_cvtps_epi32( x1 );
Shouldn't that be _mm_cvttps_epi32( x1 ) ? The cast-to-int truncates, it
doesn't round, does it ?
Amazi
On Wed, 21 Dec 2011, Joseph Heled wrote:
I find it hard to believe.
There is a simple test - if the two implementations give the same vale for
the same position, ply 0.
Analyzing and exporting to text a sample match at 0-ply, I get this as a
rough measure of number of discrepancies between t
On Thu, 22 Dec 2011, Joseph Heled wrote:
I am not sure which nets and which sigmoid implementation is used by
gnubg now - but note that my nets were trained against an
approximation. It does not matter that much if you use the exact
definition - what is important is using a "valid pair", that is
I've commited the fix, including the fast-math related ifdef below. As a
result of the latter, the code is now sligtly slower by default but one
can always use -ffast-math everywhere and recoup that and some more.
I didn't run serious speed or non-regression tests, but it looks like this
optio
I have just tried to use the gnubg-nn tools to train nets and I have some
questions for Joseph or other who may have some experience with these.
I started from the existing weights file (nngnubg.weights, 0.17-c5-123)
and with the race net :
% ./train.py -v $DATA/training_data/race-train-data
On Thu, 5 Jan 2012, Joseph Heled wrote:
Seems like you got a very very slightly better race net, but I would be
surprised if it makes a difference in real life.
Ok, so a worthwhile improvement would be a decrease of at least 10% in
benchmark error, something like that ?
Would be much more
On Sat, 21 Jan 2012, boomslang wrote:
Is it still possible to run scripts in GNUbg? I looked for gnubg-no-gui.exe
but to no avail.
gnubg-cli.exe -t
On Wed, 25 Jan 2012, Michael Petch wrote:
you need to navigate to the end of the match to see
the final score:
You can do something like:
next game 100
next 100
show score
If the match is loaded or imported, something like
set gotofirstgame off
load match / import ...
show score
sho
On Tue, 7 Feb 2012, Michael Petch wrote:
The bug is very reproducible. At any point if you click on a position
that has been previously analyzed in the move list, this will cause the
mouse to stop responding.
Running gnubg with the GDK_NATIVE_WINDOWS environment variable set to true
seems to
On Thu, 9 Feb 2012, Mark Higgins wrote:
Can anyone point me to a function to convert the keys in the benchmark dbs to a
board layout pls?
They don't look like 14-char position IDs, and checking this list archive I see
them occasionally referred to as Joseph-ID.
The underlying binary key is
On Tue, 14 Feb 2012, Ian Shaw wrote:
1 point of contact-benchmark error is about 156.5 micropoints/game
This is for checkers play errors only apparently. Do you have a similar
ratio for cube errors and for the crashed net benchmark ?
___
Bug-gnubg
It looks like there are some bugs in the evaluation of backgammon chances
in races, in raceBGprob() in eval.c . For instance, this position :
+24-23-22-21-20-19--18-17-16-15-14-13-+ O: gnubg (Cube: 2)
| O O O O O | |O | 0 points
|O | |
On Mon, 27 Feb 2012, Joseph Heled wrote:
clearing checkers from the back is not the same as bearoff - you *can*
move any checker you want.
Embarrassing. You're right of course and in my example 3s and 4s can
certainly be played in a different way than in a bearoff.
_
On Mon, 27 Feb 2012, Joseph Heled wrote:
Using some python trickery it is possible to get the true backgammon
figure. "true" here means that you always make the optimal move which
minimises the probability of backgammon. The number is 40615026 /
(36^5) = 0.6716982730973429
Now, nngnu gives tho
On Sun, 22 Apr 2012, Øystein Schønning-Johansen wrote:
Hi,
The board type in GNU Backgammon is typedef'ed as a 2d array
anBoard[2][25], but which player is on roll? Is the player with index 0 or
the player with index 1 ? I'm getting confused.
Player 1 is on roll, at least when it comes down t
I think there is something wrong (but harmless with the compilers we've
used) in these functions from positionid.c.
In PositionBearoff() for instance :
for( j = nPoints - 1, i = 0; i < nPoints; i++ )
j += anBoard[ i ];
fBits = 1u << j;
for( i = 0; i < nPoints; i++ ) {
In HeuristicBearoff(), around line 160-180 of bearoff.c.
Why is a n value computed in the first loops then discarded by the "for (
n = -1; ... )" without ever having been used ?
___
Bug-gnubg mailing list
Bug-gnubg@gnu.org
https://lists.gnu.org/mailm
The benchmark database for the crashed positions seems seriously
corrupted.
Offhand, it seems that many positions with significant backgammons in play
are very innacurate.
For instance :
#R HLDHABAADEAE 0.58 0.696718 0.000878771 0 0
#R HHDHACAADEAE 0.94 0.695403 0.002
On Tue, 5 Jun 2012, Philippe Michel wrote:
The benchmark database for the crashed positions seems seriously corrupted.
I have rerolled it. How should I proceed to have it uploaded to
ftp.demon.nl ?
The change for checker plays is quite large.
Original database :
% perr.py -W $DATA/nets
On Mon, 18 Jun 2012, Mark Higgins wrote:
Thx - can you calculate the gnubg 0-ply score against the new crashed
benchmarks?
I'm not sure what you mean by "score". The checker and cube errors I
quoted were the result of benchmarking the current net.
The averages are admittedly somewhat diffic
On Tue, 19 Jun 2012, Øystein Schønning-Johansen wrote:
Which tool did you use to roll out?
I used sagnubg, from gnubg-nn/sa. It didn't work at first, I had to fix a
small error (nothing involving the quality of results).
___
Bug-gnubg mailing list
B
On Wed, 20 Jun 2012, Jim Segrave wrote:
[access to the ftp server is complicated]
Wouldn't it make sense to put everything at the same place (like at
www.gnubg.org). The training/bechmarking data is about 100MB. Maybe ten
years ago it was serious online space, but now ?
The large files are
On Sun, 24 Jun 2012, Joseph Heled wrote:
On 24 June 2012 13:38, Joseph Heled wrote:
Hi,
It is heart warming to see that the race net can be improved still :)
The benchmark rates the new crashed net as stronger, which is great too,
but the new contact net is rated weaker in moves (0.01044531
On Sun, 24 Jun 2012, Joseph Heled wrote:
I am very interested to know how those nets were generated?
They were trained with your gnubg-nn tools, but from improved training
data. This is basically how it went :
I first tried to train the crashed net. Since it seemed one of its
problems was
On Sun, 24 Jun 2012, Jim Segrave wrote:
Can someone supply me with a list of the files to be updated and where I
can get the new versions?
I had sent you a mail with an url where to get the the new crashed.bm file
at Wed, 20 Jun 2012 23:59:02 +0200 (CEST). Is there a problem with it ?
Here i
On Mon, 25 Jun 2012, Joseph Heled wrote:
Hi again Philippe,
Did you find a way to show that the new net indeed is indeed more balanced
than the old with regard to the odd-even ply syndrome?
I can't prove it or show statistically credible evidence. What's clear is
that 1ply is much better tha
On Mon, 25 Jun 2012, Joseph Heled wrote:
By missing positions you mean the "n-out ..." line from the benchmark
ouput ?
I suppose there is an easy way to find and add them from a verbose
benchmark output since you apparently did this quickly although there must
be hundreds of them.
Yes, perr
On Mon, 2 Jul 2012, Joseph Heled wrote:
While "importing" the sse code to gnubg-nn I discovered an error in the
sigmoid implementation. The calculation made by the see code did not match
the one from the straight C code. (It was relatively easy for me to catch
since sse can be toggled during run
On Mon, 16 Jul 2012, Guido Flohr wrote:
On Mon, 2012-07-16 at 13:44 +0300, Guido Flohr wrote:
If I see it correctly, I could solve my problem by extending the series
of standard deviations arStddevTable[]. But where do the numbers come
from?
If I extend the series by just one value I have to
On Sun, 5 Aug 2012, Mark Higgins wrote:
Did you guys manage to upload this somewhere? (Or if not, can someone
email me the updated file pls?)
I have uploaded my updated benchmark files there :
crashed:http://dl.free.fr/kTXAotsaa
contact:http://dl.free.fr/kKwX4WzhM
race:
On Wed, 12 Sep 2012, M. J. Mannon wrote:
All the references I can find suggest setting "Eval Threads" to the
number of *cores* on your system. Most of these sources, I fear, refer
to CPUs that do not have hyperthreading.
I have a 4-core CPU. With hyperthreading (each core runs 2 threads), it
I have uploaded new weights at
http://www.gnubg.org/media/nn-training/pmichel/nets/20120907/
The difference since what I posted in June is mostly further training of
the contact net, and this time the result is statistically significantly
better than the current nets.
new @ 0ply won 50.19% o
On Tue, 18 Sep 2012, Joseph Heled wrote:
What did you use to play those 1M games?
This was with your gnubg-nn/scripts/play/matchplay.py script.
___
Bug-gnubg mailing list
Bug-gnubg@gnu.org
https://lists.gnu.org/mailman/listinfo/bug-gnubg
On Tue, 18 Sep 2012, Joseph Heled wrote:
Just in case you are unaware of it, my code does not reflect gnubg
playing ability per-se, since it uses its own cube decision code.
Move selection is similar, but not identical either. Nobody ever
really compared the two approaches.
Regarding cube deci
On Wed, 19 Sep 2012, Joseph Heled wrote:
when running the matches, did you use either -q or --q1 on the command line ??
I didn't. Looking at the whole script now, it looks like it may have been
useful.
I don't understand python code very well. Does -q use duplicate dice at
match granularit
On Wed, 19 Sep 2012, Joseph Heled wrote:
You should use -q1, where matches are played in pairs, and in each
game the same dice are fed (reversed) to the two players.
That way, if the two players make the same choices, you eliminate luck
completely :). If not, you reduce its effect.
But now you
On Wed, 19 Sep 2012, Joseph Heled wrote:
your test statistic is 117*-1 + 262*0 + 121*1 = 4
the standard deviation is (500 * (2/3))^0.5 = 18.25
This doesn't seem right (why 2/3 ?) and this is more than the
(1000 * 0.5 * (1 - 0.5))^0.5 = 15.8
from the dumb rollout.
On a second look I think it i
On Sat, 3 Nov 2012, Guido Flohr wrote:
I'm trying to save an edited position in GNUBG. When I reload that
file, the only thing that gets preserved seems to be the dice. Am I
missing something or is that simply not implemented?
An edited position in really a game consisting only of special "s
On Fri, 26 Oct 2012, Michael Petch wrote:
How do people feel about reformatting the code?
I agree that it would be useful
What style is preferred?
I mostly agree with yours and Joseph's suggestions except when it comes
to avoiding blank lines.
Should we consider reformatting all the co
On Thu, 13 Dec 2012, Domenico Martella wrote:
After the release, we'd like to help you to reintegrate our work in gnubg
source tree, to create a real AI linkable library (as it is for GNU
chess)...
Besides separating the playing engine from the user interface, did you add
some things that cou
On Fri, 11 Jan 2013, Stelios Togias wrote:
I was just wondering if is GnuBG being actively developed. Not so much as
in the GUI/program part as in the neural networks part. Is it considered
mature or there's still training and maybe improvement taking place?
I have been training new neural net
On Mon, 14 Jan 2013, Mark Higgins wrote:
What training approach have you been using, if you don't mind elaborating?
Supervised training. I used the same training tools that were used years
ago to create the current nets.
The main difference is that I rolled out the training database while i
I am willing to believe the new nets are better, but I have not seen the
results of a long-enough/statistically-significant run of matches between
the old and new.
I have uploaded my latest nets at
http://www.gnubg.org/media/nn-training/pmichel/nets/20130121/
Access doesn't seem to work well
On Fri, 8 Feb 2013, Michael Petch wrote:
I'm thinking we should upgrade our Wiki (if we intend to keep it)
Is the wiki really useful ? There weren't many updates there, were they ?
Moreover, the documentation is in a sorry state. There is :
- something in CVS, incomplete and not up to date
-
Currently, tutor's default analysis level is 0 ply. It plays quite well in
general but occasionally makes embarrassing errors that 2 plys doesn't.
It looks like most relatively recent machines (less that 5 years old, for
instance) should be fast enough for a 2 ply analysis to be instantaneous
On Wed, 6 Mar 2013, Russ Allbery wrote:
Jonathan Steel writes:
Is there any plan to release a "stable" version more frequently? I
maintain the package for Arch Linux. I'm just using the snapshots but
ideally I would use a version that doesn't change almost every day, and
that is relatively up
On Thu, 7 Mar 2013, Russ Allbery wrote:
For example, if you slapped a version number on each version for which
someone felt inspired to provide new "official" Windows binaries, that
would probably be about right.
That seems reasonable. The version numbers have been like 0.n.0 with the
third p
On Sun, 10 Mar 2013, Rasmus Althoff wrote:
Bug #2: The German locale setting doesn't translate all of the menu items and
shortcut graphical items. Is there an ASCII file with the locale traslations?
The files under locales seem to bee binary content. I would like to
contribute a translation if
On Sat, 16 Mar 2013, Neural Gnat wrote:
I've just re-analysed a 1000-game money session that I did about a week ago
with 2012's World Class versus Casual. This new test version has found 1839
doubtful moves, 304 bad moves and 247 very bad moves, knocking the mainstream
version down from Supern
On Sat, 16 Mar 2013, Michael Petch wrote:
It might be an interesting experiment to run the Depreli positions
through the new version to see how they compare with XG2 and the
previous version of GNUBG.
I did this at some point with the first 300 positions, but by hand it is
an extremely tediou
1 - 100 of 412 matches
Mail list logo
