I hope you realize you will need hundreds of thousands of games, millions maybe, to get statistical significance.
-Joseph On Mon, 29 Jan 2024 at 10:03, MK <playbg-...@yahoo.com> wrote: > Firstly, let me be forthright that my purpose with these > experiments is to debunk the so-called "cube skill theory" > or at least raise serious doubts about it. So, I'd fully > understand you guys' unwillingness to contribute to it by > helping me with my scripts, etc. > > With that said, since the crude script I had posted here, > I decided that instead of duplicating Axel's experiment, > (which I was the one who had suggested), I would start > with an even more radical "mutant cube experiment" where > the mutant doubles and takes randomly but never resigns. > > I can't imagine any worse cube strategy than doubling or > taking even when one's winning chances are near zero. > > To make up for that "no cube skill at all" the mutant > never drops, this in effect, turning the cubeful game > into a cubeless game as much as possible by forcing it > to be played out until the end or until the bot drops. > > I have two other experiments planned and may come up with > more later on. I will share my scripts so that, instead > of taking my word for things, you can use them to run your > own experiments to see the results for yourselves. > > I have documented what they do using lots of comments. It > asks for a starting game number for a session and for the > number of games to run. On my, probably below average CPU, > it takes about 15 minutes for 100 games. That works out to > about 10,000 games in 24 hours. I suggest you run it in > sessions of 100-500 games and from the Python interactive > interpreter, so that, if it fails you still will have a > chance to view and save things. > > It saves each game with a long filename containing lots > of stats and finally saves the match with a similar name. > All this is documented in the script. If you want neatly > consecutive game numbers, use the last game number as the > start of session, i.e. after session=0, games=100, going > from 1 to 100, you can enter session=101 and games = 100 > to go from 101 to 200, etc... > > You can download the script from: > > https://montanaonline.net/backgammon/py/MuratMutantCubeRandom.py > > The directory listing of filenames for my first run of 100 > are saved in a text file that you can download from: > > https://montanaonline.net/backgammon/py/MuratMutantCubeRandom.txt > > I also imported the text file into a spreadsheet, (with sorted > O's and X's wins, including totals), that you can download from: > > https://montanaonline.net/backgammon/py/MuratMutantCubeRandom.ods > > One outstanding feature of my script is that it can go beyond > the maximum 4,096 cube value limit of GnuBG by recycling the > cube value without altering the cube ownership, as many times > as it may be needed. > > In my first session of 100 games, I was lucky enough to see the > cube go as high as 2^34 = 17,179,869,184 > > I ran another 100 and two 500 game sessions with the following > points won Gnubg World-Class checker and cube, Mutant World-Class > checker and random cube as described above): > > Gnubg = 543,696,648 > Mutant= 17,180,933,757 > > Gnubg = 6,995,268 > Mutant= 9,344 > > Gnubg = 4,441,646,726 > Mutant= 1,076,936,738 > > Gnubg = 6,764,733,622 > Mutant= 2,152,163,991 > > Granted this much is hardly significant sampling but I think it > gives a hint of what you may expect in the long run. Don't take > my word for anything; run your own sessions as long as you want. > I predict that the results will be quite unsettling to most of > you. Please share any results that you may get. > > MK > >
