Mon 25 Nov 2013 11:33:06 PM UTC, comment #6:
Okay. There are a few things going on here. First the position that is posted in this bug report by the original poster is actually an obvious take (unless I am reading it wrong). Player O has all of their pieces off and player X is on move. So O definitely has a take. Unless I am missing something I think the position posted is not a good example.
Secondly, there was a bug in the external interface regarding cubing that was fixed in early 2012 that would make it appear like most cubes are takes. Your 2011 version pre-dates that fix. The bug in question we dealt with (and closed) here: https://savannah.gnu.org/bugs/?36485
So, what I'd like you to try is to get a newer version of GNUBG. There are a couple version you might consider. There is a newer version 1.02.000 from July of this year that uses the new neural network (It is stronger than the versions earlier than 2013). It is recommended. That can be downloaded here: http://files.gnubg.org/media/windows/gnubg-release-1_02_000-20130728-setup.exe
If you want to use the older neural network but have the bug fixes that relate to the socket (external) interface then you can download:
http://files.gnubg.org/media/windows/gnubg-MAIN-20121023-setup.exe
Can you try one of them out and see if you still have issues?
|
Mon 25 Nov 2013 07:08:30 PM UTC, comment #4:
What is generating your FibsBoard positions. Are you building a bot that plays on the Fibs Server and taking the output from Fibs and sending it to GNUBG (via the socket) or are you generating a FibsBoard position yourself and sending it to GNUBG?
Second what version of GNUGB are you using. The original author was using an older one, but don't know what you are using now.
|
Mon 25 Nov 2013 07:55:39 AM UTC, comment #3:
I tried the board string provided and my gnubg-cli gave back the response drop.
However, I understand what the original submission was talking about as I have the same feeling - that the cli almost always gives back a take response. I've even had a case where my winning percentage was 0.1 % and the chance of losing by a gammon was 48% and the cli reported a take decision on a double.
I'll try and post some examples when I next come across this issue.
|
Mon 21 Jan 2013 07:26:35 PM UTC, original submission:
Hi, gnubg-cli.exe seems to be working wrong when I am sending it a FIBS board which needs a take or pass cube decision evaluation. I build the FIBS board string so that the "was doubled" value is set to one, that's how gnubg-cli.exe knows that it has to make a take/pass decision. And it seems to recognize that, it's only that it seems that it is always returning the reply "take", even for obvious "pass" board states. I never saw it return one "pass", and it never seems to "think" on it, even if it uses the analisys settings and those are set to a high standard (at least 3 plies, pruning, etc.), it just always returns a "take" right away. So my guess is that it is either a bug, or this feature is just simply not implemented in gnubg-cli.exe, it just doesn't call GNU backgammon's appropriate cube evaluation functions. Any feedback would be very welcome, even if just a confirmation of my guess. Of course, if this could be also fixed, would be just great!
Here is an example of an obviously "pass" board state, you can test it with this, if you send this string to gnubg-cli.exe with some simple socket tester program:
board:dianepop:opponent:7:0:0:0:2:2:1:0:0:0:0:0:0:0:0:0:0:0:0:0:0:0:0:0:0:0:1:1:0:-1:0:0:0:0:1:1:1:1:1:-1:0:25:0:0:0:0:2:0:0:0
|