XG paradox !

[dorbel]

Michael is correct in that bots don't take into account skill differential when making their cube decisions and of course knowledgeable humans will use this to their advantage. It isn't a bug, it just isn't part of the program.
However Michael's opinion on bot weaknesses in defending against backgames is only an assertion. There play may be and probably is less than optimal, but there is no way to show that without a human or a superior bot producing better results than XG.
The bot weakness of failing to find the best play where there were large numbers of options, usually with small doubles, was occasionally noticeable in gnu and snowie and may for all I know apply to XG. To save time, they narrowed the field for higher plies of evaluation by doing a faster scan at 0-ply (1-ply in snowie), which could and sometimes did fail to include the best play. Of course humans do this too. However, I don't believe that there is any evidence to show that this is "more frequent in backgames" as Michael asserts.

[dorbel]

It would be possible for a bot to continually analyse its opponent's play and adjust its own cube action accordingly. Would that actually be desirable though? At present, its actions constitute a reference point, which is a basis for our own actions. If that reference point is constantly shifting depending on who it is playing, how would we know what was right? The bot would win more, but that isn't its raison d'etre is it?

[ah_clem]

It would be possible for a bot to continually analyse its opponent's play and adjust its own cube action accordingly. Would that actually be desirable though? At present, its actions constitute a reference point, which is a basis for our own actions. If that reference point is constantly shifting depending on who it is playing, how would we know what was right? The bot would win more, but that isn't its raison d'etre is it?

There was an interesting thread on that a few year's back over at Stick's site:
http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=171858#171858

For analyzing my own matches to see where I made mistakes, the current bot behavior of assuming perfect play on both sides would still be desirable.  If you're designing a bot to win tournaments against human players, using an asymmetric MET and otherwise adjusting play based on the opponent might produce better results.  But my understanding is that the bots will beat the humans already so it would seem to be an unnecessary enhancement.  Plus, I don't think any TDs would allow a bot to enter.

[FancyPrince]

I actually agree with you, in what the bots do. A human would very reluctantly cube in a match Vs a weaker opponent in backgames for various reasons, while the bots simply base their cube decisions on the Match Equity table after estimating Wins (single), % gammons and % backgammons for each side . Hence they would  double or redouble with no restriction if they judge it worths it.

There is also another weakness in how Bots handle backgames imo. That has to do with the so called "timing". A human would avoid hitting too much because this offers  his opponent all the timing he needs to have an effective backgame. Bots prefer hitting for their + in equity without usually caring much about the timing factor. Probably because the effect of timing will show up in 8+ plies while bots are not designed to "think" that deep.   Of course they have some correction code running at the end that tries to correct their timing, some rolls before the bearing off phase, by recirculating some checkers. Sometimes it works sometimes it doesn't.

And thirdly you might want to know that the way bots evaluate each possible move, fails many times when the options for that roll are too many. Imagine a roll has 24 possible plays, it often occurs that the best move fails to make it trough their 3-4 ply evaluation and  it could even fail to make it through in rollouts. This is because at some point during this evaluation phase bots start scraping out moves. This is what some people call "confusion" of bots. Such cases where a roll has too many alternative plays are more frequent in backgames.

thanks for your comment. i learned new things . when i was 19 i saw my friends playing a game with two dices..blocking each other and bearing checkers and finally one of them wins. i told them this is a game just based on luck. they said come and examine .i played many games and lost all they made me 6-prime and won all the games and a lot of fun for them . after years i found this game much more exciting and tried to play better and better. i think i can play at good level now. but now, again i have came back to my first idea: this game is based on luck! plenty of matches i have analyzed since i have played at FIBS or NETGAMMON servers, maybe more than 1000 matches. but the result was same. any game i won my luck was much more than my opponent and any game i lost my luck was less. do you agree with me? thats why i have created my own style:-) . the point i noticed in new style is that i have gained more than 100 outplays! and its not reachable for who plays like XG .
in backgammon i think the main goal is winning not receiving better Pr. i prefer to have much luck and less Pr than losing with -1 wow!!Pr.

[MichaelP7]

snip..... this game is based on luck! plenty of matches i have analyzed since i have played at FIBS or NETGAMMON servers, maybe more than 1000 matches. but the result was same. any game i won my luck was much more than my opponent and any game i lost my luck was less.

a)I have studied the matter of luck to some depth and concluded that as long as the abs.Fibs Rating difference between the players is < 230 then the number of games won by pure luck are more than those won by skill. If skillA=skillB then all games are won because of pure luck. At abs.Fibs Rating difference =230 it breaks even and at >230 skill wins more games than luck.
So I don't agree with your over generalization that "this game is based on luck". Surely if you play an almost equal skill player e.g you play someone who is weaker by about 100 Fibs rating points then playing 100 7 pointers you would only win 14 matches because of skill and 43 matches because of luck. So in general it's a combination of Skill and luck.

b)Your observation that 99%+ of  matches or games won are by the player who "appears" [at bots' summary results]to have accumulated more nett luck is true,  and coincides with my observation as well. It is very very rare for the looser to ever have accumulated more nett luck than the winner. I am currently studying this issue but got no conclusions yet. I am concentrating on the possibility that it just "appears" to be so, but I could be wrong.
The bots results surely deviate a lot from the mathematical equation of Nett Luck+Net skill of the winner in a match=+50%...

NB. You will need a better PR to win more times. Surely you can continue playing those backgames to get some experience how they could possibly win, but I think you should revert to normal strategies and improve your PR.

[dorbel]

The great majority of short matches, ie less than 11 pts are won by the luckiest player. When you examine stats closely you will see that the player who plays best on the day is more likely to have positive luck. The greater the skill differential, the more pronounced this is. This is counter intuitive, as we all learn at our mother's knee that the luck will even out, so why doesn't it? Go to http://www.fibsboard.com/fibsboard-forum-matches/how-do-%27luck%27-calculations-work/ , read the whole thread, particularly the contributions of pck, boomslang and Zorba and all will become clear. The best thread ever on this site.

[MichaelP7]

When you examine stats closely you will see that the player who plays best on the day is more likely to have positive luck.

Hmm I don't think so. I just checked a random folder containing 13 matches Vs bots that by definition play better all the times and here are the luck results for the bots: +46, -22, -22, -63, +13, +14, +33, -13, +15, +24, -45, -22, -2.
If you are talking for nett luck for the winner that's almost always positive regardless of skill.

NB.Thanks for the link.  I 've actually read it a few weeks ago.

[dorbel]

You think a sample of 13 matches is statistically significant? Are you pulling my leg?

[MichaelP7]

You think a sample of 13 matches is statistically significant? Are you pulling my leg?

It surely is an indication. Searched and found another folder with 5 more matches Vs bots.
Again here are the results for the bot +14, +80, +38, -5, -43. The indication to me so far is that the odds for any side to get positive total luck is rather 50-50!
On the other hand if you have statistical data to back up your statement feel free to present it.