Re: Murat's first mutant cube script experiment and preliminary results

I think one problem with your current test is that it does way too much (random) doubling. So you get huge total scores coming out, which adds loads of noise to the results and makes it hard to be confident what it means. A simpler test might be to compare gnubg's best strategy against a "dumber"

Re: # of hidden nodes

current hidden layer with more nodes. > > ... But you might have different plans... > > -Ø > > On Thu, 28 Dec 2023, 01:26 Mark Higgins, wrote: > >> Thx! >> >> On Wed, Dec 27, 2023 at 7:20 PM Øystein Schønning-Johansen < >> oyste...@gmail.com>

Re: # of hidden nodes

Thx! On Wed, Dec 27, 2023 at 7:20 PM Øystein Schønning-Johansen < oyste...@gmail.com> wrote: > The normal contact position neural network had 250 inputs, 128 hidden > nodes and 5 outputs. > > -Ø > > On Wed, 27 Dec 2023, 23:23 Mark Higgins, wrote: > >> Can someon

# of hidden nodes

Can someone remind me pls - how many hidden nodes does the gnubg 0-ply bot use in its neural networks? Thx!

updated gnubg benchmark files

I know Philippe Michel at some point created new benchmark databases; the list archive has the message here: https://lists.gnu.org/archive/html/bug-gnubg/2012-08/msg00010.html Those links don't work anymore - can anyone point me to the most recent versions of these files? Similarly - are there

gnubg-nn position ID definition

I've been looking at the gnubg benchmark databases, which have positions represented as gnubg-nn position strings (20 characters, like JIGHPAABDAOAHDPAABDA), rather than regular gnubg position IDs. I found some code in gnubg.c that converts a gnubg-nn position string to a list of checker counts;

Re: human level opponents in GNUbg?

I’d be interested in that new backgammon engine initiative!On Aug 4, 2023, at 1:59 PM, Øystein Schønning-Johansen wrote:(not harsh at all, Joseph!)Yes! I think everyone agree to that. The engine and the gui should be separated in two processes with a protocol between them.BTW. There is a new

Re: [Bug-gnubg] GNU Backgammon

I think Backgammon NJ uses their own net, but other iPhone apps do seem to be based on GnuBG: https://www.reddit.com/r/backgammon/comments/2urzom/are_there_good_backgammon_apps_for_the_iphone/ On Wed, Jun 15, 2016 at 11:06 AM, Ian Shaw wrote: > Hi Sharon, > > > >

Re: [Bug-gnubg] Is GNUBG actively developed?

What training approach have you been using, if you don't mind elaborating? On Jan 14, 2013, at 5:26 PM, Philippe Michel philippe.mich...@sfr.fr wrote: 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

Re: [Bug-gnubg] Problem with the crashed benchmark database

Thanks gents! On Aug 19, 2012, at 1:49 AM, Michael Petch mpe...@capp-sysware.com wrote: On 2012-08-18 16:25, Philippe Michel wrote: I have uploaded my updated benchmark files there : crashed:http://dl.free.fr/kTXAotsaa contact:http://dl.free.fr/kKwX4WzhM race:

Re: [Bug-gnubg] Problem with the crashed benchmark database

, Ian or Mark Higgins. Note that there currently are no weights files present. I can ask to have an additional directory made where weights can be uploaded. There is already nn-training/nets/. Anyway, the current weights file is in CVS and experimental ones are small enough to be easily

Re: [Bug-gnubg] Problem with the crashed benchmark database

Thx - can you calculate the gnubg 0-ply score against the new crashed benchmarks? On Jun 18, 2012, at 5:36 PM, Philippe Michel philippe.mich...@sfr.fr wrote: On Tue, 5 Jun 2012, Philippe Michel wrote: The benchmark database for the crashed positions seems seriously corrupted. I have

[Bug-gnubg] cube decisions

I think the standard for cube decisions is to follow a Tesauro (2002) approach of approximating cube decisions with a function that takes cubeless probabilities and some measure of volatility (eg std dev of equity over the next roll or two). I assume this is what GNUbg does? Has anyone looked

Re: [Bug-gnubg] Benchmark error to ppg loss

I did some regressions of benchmark scores vs cubeless money play scores: http://compgammon.blogspot.com/2012/02/bug-in-benchmark-calculation-fixed.html Short story: almost all the money play score variation comes from Contact benchmark score. There's a strong dependence of money score against

Re: [Bug-gnubg] Joseph-ID in benchmark db

Can you clarify what rollout values of the outcome probability mean pls? eg how do I turn those into a cubeful equity, or a cube decision? Is there any benchmark for cube decisions directly, like whether to double/redouble at the start of a turn, and whether the opponent should take/pass?

Re: [Bug-gnubg] Joseph-ID in benchmark db

to discuss your extra inputs? n Ian From: bug-gnubg-bounces+ian.shaw=riverauto.co...@gnu.org [mailto:bug-gnubg-bounces+ian.shaw=riverauto.co...@gnu.org] On Behalf Of Øystein Schønning-Johansen Sent: 12 February 2012 16:39 To: Mark Higgins Cc: bug-gnubg@gnu.org Subject: Re: [Bug

Re: [Bug-gnubg] Joseph-ID in benchmark db

Does anyone have the average error stats for 0-ply gnubg on the contact benchmarks? I see race crashed results at Joseph's page here: http://homepages.ihug.co.nz/~peps/ngb/index-top.html but can't find the contact result anywhere. (Though I'd guess it's pretty close to the crashed error, ie

Re: [Bug-gnubg] Joseph-ID in benchmark db

it. -Øystein 2012/2/12 Mark Higgins migg...@gmail.com Does anyone have the average error stats for 0-ply gnubg on the contact benchmarks? I see race crashed results at Joseph's page here: http://homepages.ihug.co.nz/~peps/ngb/index-top.html but can't find the contact result anywhere

Re: [Bug-gnubg] crashed position

the motivation for the current definition? On Dec 17, 2011, at 12:22 PM, Øystein Schønning-Johansen wrote: On Sat, Dec 17, 2011 at 4:05 PM, Mark Higgins migg...@gmail.com wrote: I'm trying to find the exact definition gnubg uses for a crashed position. The one reference I've found

[Bug-gnubg] Joseph-ID in benchmark db

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. ___ Bug-gnubg mailing

Re: [Bug-gnubg] pubeval benchmark

comparable? gr boomslang From: Mark Higgins migg...@gmail.com To: bug-gnubg@gnu.org Sent: Tuesday, 17 January 2012, 6:28 Subject: [Bug-gnubg] pubeval benchmark How does gnubg perform against the pubeval benchmark in cubeless play? I ask because I'm playing around with a backgammon

[Bug-gnubg] 1-ply vs 0-ply equity improvement

Does anyone know roughly how much of a cubeless money equity improvement 1-ply enjoys over 0-ply? (And 2-ply vs 1-ply.) I googled around for a while but couldn't find anything. ___ Bug-gnubg mailing list Bug-gnubg@gnu.org

[Bug-gnubg] Berliner

Hi all - when I see references to Berliner for contact inputs etc - am I correct in assuming that refers to the 1980 Sci Am article Computer Backgammon about BKG? ___ Bug-gnubg mailing list Bug-gnubg@gnu.org

Re: [Bug-gnubg] Berliner

Of Mark Higgins Sent: 26 January 2012 13:55 To: bug-gnubg@gnu.org Subject: [Bug-gnubg] Berliner Hi all - when I see references to Berliner for contact inputs etc - am I correct in assuming that refers to the 1980 Sci Am article Computer Backgammon about BKG

Re: [Bug-gnubg] Berliner

Of Mark Higgins Sent: 26 January 2012 13:55 To: bug-gnubg@gnu.org Subject: [Bug-gnubg] Berliner Hi all - when I see references to Berliner for contact inputs etc - am I correct in assuming that refers to the 1980 Sci Am article Computer Backgammon about BKG

Re: [Bug-gnubg] playing a bot on FIBS

Wow - very handy - thanks very much for the detailed answer. I should be able to run from here. On Jan 23, 2012, at 5:37 AM, Øystein Schønning-Johansen oyste...@gmail.com wrote: I did this with glib and the GSocket classes. I guess you can connect to fibs with whatever system you want.

Re: [Bug-gnubg] playing a bot on FIBS

Thanks Michael! Those are some great starting points. On Jan 21, 2012, at 10:04 PM, Michael Petch wrote: On 21/01/2012 7:41 PM, Mark Higgins wrote: Can anyone point me to some docs on how I can connect my bot to play on FIBS? I know gnubg did this, but I can't find the code. Howdy Mark

Re: [Bug-gnubg] pubeval benchmark

I used the following post to define pubeval; is this the proper definition still: http://www.bkgm.com/rgb/rgb.cgi?view+610 ? On Jan 19, 2012, at 8:31 AM, Joseph Heled wrote: Hi, I just run 1 1 point match games between gnubg and pubeval. I think this is the most fair, since there

Re: [Bug-gnubg] pubeval benchmark

:) I'll give it a try and check the couple games you sent. On Jan 19, 2012, at 9:39 AM, Joseph Heled wrote: I was asking for help, not for more work. For example, anyone who implemented play against pubeval can check a few of the moves I sent. -Joseph On 20 January 2012 03:35, Mark

Re: [Bug-gnubg] pubeval benchmark

, 0, -2, 0) On 20 January 2012 03:45, Mark Higgins migg...@gmail.com wrote: :) I'll give it a try and check the couple games you sent. On Jan 19, 2012, at 9:39 AM, Joseph Heled wrote: I was asking for help, not for more work. For example, anyone who implemented play against pubeval

Re: [Bug-gnubg] pubeval benchmark

doesn't agree with the moves selection in the games you sent. In the first game, the first roll of pubeval (4-5) is played as 24/15, and in the second game the first roll (52) is played as 13/8, 13/11. Nikos On 19/1/2012 17:14 μμ, Mark Higgins wrote: I get something different for a 3-1

[Bug-gnubg] Benchmark database

As I (imperfectly!) understand it, the benchmark database was created by randomly picking off 100k positions from FIBS games. Why use FIBS games instead of generating them from eg an intermediate-strength neural net player? Presumably we don't care about the exact positions, but rather just

[Bug-gnubg] pubeval benchmark

How does gnubg perform against the pubeval benchmark in cubeless play? I ask because I'm playing around with a backgammon network and have got one that wins 83% of games and +0.945ppg against pubeval (10k cubeless games). This is a single 80-hidden-node network with outputs for prob of win,

[Bug-gnubg] crashed position

I'm trying to find the exact definition gnubg uses for a crashed position. The one reference I've found (Thomas Haug's thesis) says it's contact, plus the restrictions that the player has fewer than 7 pieces remaining with none in the opponent's 1 or 2 position. Is that correct? If so, can

Re: [Bug-gnubg] Neural network symmetry question

computing power it should be easy to do at least a couple of millions. -Joseph On 12 December 2011 11:22, Mark Higgins migg...@gmail.com wrote: I tried a little experiment on this: a 10-hidden-node network with a single probability-of-win output, but two setups. The first doesn't have

Re: [Bug-gnubg] Gammon output setup

with the usual gammon win output: do you train this any longer in mid-game if the opponent has borne in a checker, or do you just stop training? On Dec 11, 2011, at 7:56 AM, Øystein Schønning-Johansen wrote: Hi Mark! How's your rally driving going. ;-) On Sun, Dec 11, 2011 at 4:45 AM, Mark Higgins

Re: [Bug-gnubg] Neural network symmetry question

here: http://compgammon.blogspot.com/2011/12/testing-value-of-symmetry-constraint.html Of course not conclusive with such a simple setup, but kind of suggestive anyways. On Dec 10, 2011, at 2:22 PM, Mark Higgins wrote: Thx! Makes sense. Though I wonder if adding back in the whose move

Re: [Bug-gnubg] Neural network symmetry question

of millions. -Joseph On 12 December 2011 11:22, Mark Higgins migg...@gmail.com wrote: I tried a little experiment on this: a 10-hidden-node network with a single probability-of-win output, but two setups. The first doesn't have a whose turn is it input and doesn't add any symmetry constraints

Re: [Bug-gnubg] Neural network symmetry question

you flip the position, since it matters who is on the move. -Joseph On 10 December 2011 16:17, Mark Higgins migg...@gmail.com wrote: I've been playing around a bit with neural networks for backgammon and found something interesting, and want to see whether this is already part of gnubg

Re: [Bug-gnubg] Neural network symmetry question

Thx! Makes sense. Though I wonder if adding back in the whose move is it input and reducing the hidden-output weights by half ends up as a net benefit for training. Maybe I'll test it out. On Dec 10, 2011, at 2:06 PM, Frank Berger fr...@bgblitz.com wrote: Hi Mark, If I take a given

[Bug-gnubg] Gammon output setup

I notice in gnubg and other neural networks the probability of gammon gets its own output node, alongside the probability of (any kind of) win. Doesn't this sometimes mean that the estimated probability of gammon could be larger than the probability of win, since both sigmoid outputs run from 0

[Bug-gnubg] Neural network symmetry question

I've been playing around a bit with neural networks for backgammon and found something interesting, and want to see whether this is already part of gnubg. Assume a Tesauro-style network with the usual inputs, and some number of hidden nodes. And for simplicity, just one output representing the

Re: [Bug-gnubg] python module available?

time. :) On Jan 1, 2010, at 7:09 AM, Øystein Johansen wrote: On Thu, Dec 31, 2009 at 10:33 PM, Mark Higgins migg...@me.com wrote: I'm coding up my own little backgammon simulator in python and want to benchmark it against gnubg. I was hoping there's a gnubg python module I can install

[Bug-gnubg] python module available?

I'm coding up my own little backgammon simulator in python and want to benchmark it against gnubg. I was hoping there's a gnubg python module I can install and import that would expose the underlying engine so I can pass my generate boards and dice rolls into it, and have it return the set of