| |
BGonline.org Forums
XGR++ RO and 1 ply RO..etc
Posted By: Henrik Bukkjaer In Response To: XGR++ RO and 1 ply RO..etc (Stick)
Date: Wednesday, 14 December 2011, at 1:02 p.m.
So, if I get this right, you work with a very short list of people, who's play can be realistic compared to the rollouts and evaluations you perform and rely on. Wouldn't that be leaving your analysis on the grey area most of the time, for all practical purposes then?
>> Using GNU to rollout something like this esp. adding 'noise' is nothing close to realistic.
This comes up very often I suppose, but:
1) Maybe gnubg 2-ply misplay a containment game like this, compared to XGR++. But not in the rough neighborhood as 99% of all live tournament players misplay it (according to your own statement regarding your list). So you pick at a possible error at the third decimal in gnubg numbers, and discard the idea of even using gnubg to analyze the position, even though you acknowledge that the players you are analyzing is far more off...
2) For theoretical purposes of pursuing optimal play, the current XG MET must be accepted as the reference. But for this particular analysis, wouldn't you think that skewed Jacobs table is more suitable? (I know, XG can't handle such tables yet).
3) Why is adding a little noise not correct, when analyzing players of very different ability (remember, it was Kaz who in his post presumed players of unequal ability, thus opting for a move which might be theoretically inferior, so the proposition was a skewed one)? OK, I know the answer to that question myself: bot noise and bad humans don't err in the same patterns. BUT, then let me ask: Why do you think going XGR++ is better than a lower setting, for analyzing a player who is clearly NOT playing at that level?? Just because something isn't 100% conceptually correct, it can still be an improvement on status quo.Basically speaking, if you close your eyes to this kind of analysis, and only opt for the pinnacle, you'll probably end up playing the lowest PR around, but not winning as much as you should... I know that you don't do this, having seen a lot of your posts taking into consideration player ability, but why then not try to get as close to that idea when using the bots to analyze as well?
All I reacted to, was Neil posting a comment, where he assumed a revised ME table for the remainder of the match due to a player having less skills, but at the same time assumed XGR++ checkerplay for the rest of the ongoing game. And THAT is "nothing close to realistic" I imagine :-)
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.