| |
BGonline.org Forums
Is Snowie still relevant?
Posted By: Jack Mack In Response To: Is Snowie still relevant? (Casper Van der Tak)
Date: Monday, 13 March 2017, at 7:20 a.m.
Regarding your point #1: You can also see that Xavier was responding to Kees van den Doel's comments "I learned bg on gamesite2000, playing many games against their XG bot. I remember reading it was then based on gnubg. Is this new commercial stand-alone version based on gnubg?"
It's clear to me that Xavier was giving a "yes + but" answer. If your understanding of the same words is different than mine, so be it.
Regarding your point #2: Since both programs do erratically resign, they must already be checking for "gin-positions" (i.e. no chance of winning), otherwise they would never resign (just like Snowie that never resigned, for example).
Thus, obviously it's not a case of some "look-ahead horizon effect" regardless of whatever it means, which I don't understand and would appreciate if you explained what it is.
Regarding your point #3: I understand and I agree on seemingly random moves being the result of "no equity difference between the moves" but I don't think you understand that here we are taking about "all moves having an equity of zero or less" (which is what we refer to as "gin-positions")!?
So, the question is: why do both bots fail to resign even though they know all possible moves have zero or less equity? (Which in some other such positions they "do" resign.)
Before I move on to your next point, let me underline that you conveniently ignored the odd behavior of both bots resigning a gammon when their opponents only need to win a single game to win the match!
Equally oddly, at least until now, there seems to not have been observed that either bot resigned a backgammon when a gammon (or a single game for the proper argument) was enough for their opponents to win the match.
This "shared" bug is just as important and as telling as the "shared" failing to resign bug!
Regarding your point #4: Yours and other's argument that 5/4 1/off(3) is correct (vs 5/4(4)) because it avoids blotting in X2 followed by X2 sequences seems very simplistic and based on XG/GNUBG opting for it.
The only X2 rolls that matter are 62, 52, 42 and maybe 32 but not necessarily 32, 22, 12. And if that happens then Blue needs to roll a 1 (unless he also happens to dance 3 rolls in a row). I think the odds of this is about 0.01155
However, a 65 immediately after 5/4 1/off(3) would leave a blot and Blue's hitting chances would be 0.009258 (almost equal to above) if Blue had entered; but with 5/4 1/off(3) still keeping the 5-point, Blue would be more likely to dance and his chances of hitting then would increase to 0.02324!
So, I really can't understand where you guys are coming from on this??
Surely Snowie wasn't such a terrible "neural-net" bot and surely not worse than "stick" at it...
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.