Hi Murat,
The code doesn’t "make an exception for the bot". The bot already ALWAYS automatically decides whether to roll or move. The exceptions are for the human, to speed up our play when the decision is trivial. The human is allowed
to roll, EXCEPT if the auto roll is checked and he doesn’t have cube access; the human allowed to make their own move, EXCEPT if the auto play options are checked for forced moves or greedy bearoff, in which case it makes the move generated by the bot.
If the bot looked at the auto-roll or auto-play options, how would you expect behaviour to differ from what it does now?
- The bot will always roll if it can’t double. Auto-roll can't change this.
- The bot will always play a forced move. Play forced moves can't change this.
- The bot will always select the best bearoff play. It wouldn’t save any time by enforcing greedy bearoff, but it would lead to occasional errors.
As I understand it. what you are trying to do is make have Player 1 make decisions that are not the one the bot would make. If you want a Player to make moves that gnubg would not make, then gnubg can’t be the player. You can’t make
the gnubg engine only be itself for some moves, and ask you to make the choices for other moves.
That is I why I suggested setting Player 1 as Human, because you can make your own choices, but also ask gnubg for a Hint as to what it recommends. (What you seem to want to do is to make gnubg ask YOU for a hint and play that move instead
of its own!)
The Options, Players dialogue does have a 3rd option for an external player. But I’ve never heard of anybody using it and I don’t know whether it works.
You didn’t say what script language you are using. As far as I know, gnubg only supports Python. You can also run a simple file of gnubg commands, but there is no calculation of choice options there if I recall correctly (any scripts
I wrote were at least 2 computers ago and I’ve lost them
L).
The simplest way to ensure both sides are using the same evaluation parameters, set Player 0 to GNU Backgammon to World Class, as above. Set Settings, Analysis to World Class and the Hint Level to use the same as analysis.
The alternative is to look in gnubgautorc and ensure the settings are as you wish, or edit the settings you want in another file of commands that you load.
It’s interesting that you think the cube decision can be improved upon. Have you reviewed the changes that Philippe has just announced to 0-and 1-ply cubes? I haven’t, but I’m wondering what changes you have planned compared to the originals
or to his.
What’s your algorithm?
Regards,
Ian
-----Original Message-----
From: MK <playbg-rgb@yahoo.com>
Sent: Wednesday, January 17, 2024 5:34 PM
To: Ian Shaw <Ian.Shaw@riverauto.co.uk>; bug-gnubg@gnu.org
Subject: Re: Problems with auto-roll and auto-play functions
On 1/17/2024 2:52 AM, Ian Shaw wrote:
> Hi Murat,
Hi Ian,
Thanks for trying to help with such detailed suggestions, which I don't get too often. So, I appreciate it. I will respond to your ideas and also explain what I want to do, so that you or others may offer further/better ideas.
> I would expect the auto-roll and auto-play functions to only apply to
> the human player. The bot makes it's own decisions on whether to roll
> or move.
I understand. Bot can still make its own cube and checker decisions with or without auto-roll or auto-move. The code should not waste additional logic to make an exception for the bot. The flags should work the same for human and bot.
What's wrong with that? It will take five minute to take out the addition "if" in the code. Very simple with huge flexibility and benefits to experimenters.
> If you have a script that makes moves, perhaps you could approach it
> in another way. Set the player to human. Get a hint. Parse the first
> hint and make that move.
> In the CLI, “hint 1” shows the best move. But if you have a script,
> perhaps can extract the output of hint.
Parsing the bot's cube and checker hints and then making the human player execute these would take excessive and unnecessary effort and even then it may not work if the hints don't exactly match how the bot actually plays at the same
ply settings.
> What are you trying to achieve?
Okay, let me explain. I want to experiment with alternative double/take points, etc. using the CLI. Both players will be the bot set at, let's say, 3-ply cube and 3-ply checker.
When Player-1 is on roll, the script simply will say "Play"
and the bot will make its cube and/or checker move as usual.
When Player-2 is on roll, if it doesn't have access to the cube, the script simply will say "Play" and the bot will make its checker move as usual.
If it has access to the cube or has to respond to a cube action from Player-1, the script will get its winning chances from the hint and make a cube decision based on that to double, beaver, take or drop, according to a set of custom
cube points hard-coded within the script.
After that, again the script simply will say "Play" and the Player-2 bot will make its checker move as usual.
Not only that this is much easier and cleaner since it will only need to parse for one value from the hint, it also makes sure that both sides will make checker moves per the same player settings, without worrying if hints based on analysis
settings will exactly match those.
I hope I explained clearly. I will appreciate any further suggestions and even an existing similar script that I can modify instead of inventing the wheel from scratch.
MK