                      |
| ||||
| XGID=-aBBa-aa---c---A-bbbbA----:1:-1:1:21:0:0:3:0:10 | |||||
| Blue to play 21 | |||||
1. | 7-ply | 15/13 2/1* | eq: -0.0746 | |||
| ||||||
| 2. | 7-ply | 15/14 3/1* | eq: -0.2630 (-0.1883) | |||
| ||||||
| 3. | 7-ply | 3/2 3/1* | eq: -0.3509 (-0.2763) | |||
| ||||||
| 4. | 7-ply | 3/1* 2/1 | eq: -0.3628 (-0.2882) | |||
| ||||||
| 5. | 7-ply | 15/12 | eq: -0.3861 (-0.3115) | |||
| ||||||
eXtreme Gammon Version: 2.19.208.pre-release
Late for the party again, baby sitting duties , then lost time playing with my new SSD. On cloning the hard drive Mr Xavier threw out my registration. No problem a couple of clicks and I was back. I liked that bit of programming from Mr Xavier, but have come to expect it. I am one of those clever guys who never reads manuals until the need arises, stupid boy. The need has arisen I have to justify my views on XG’s analysing abilities. I post my rollout, mmmm the result looks very similar to MR Symborg's post, but there is a big difference its 7 ply and 1296 games to 100% . My roll out is obviously way over the top but these are learning times. For whatever reason it never displays additional details below but now I realise whilst writing ,why would it there aint any lol..
The default analysis options on XG have been tinkered with and refined to get quicker results on slow machines. In that more often than not the right result is obtained is a credit to the programming. I guess this will not always be the case and a deep roll out will be required on occasions. I don’t pretend to understand what goes on in XG’s brain but was always of the opinion that a rollout went and got the T shirt and indeed it does according to the following excerpt from the bible with no help from the neural network . Enlightened with my reading I set a new deep roll out setting which excludes all the short cuts. I have struggled to get any information on this subject from anyone least of all the internet. Lets hope I’m on the right track. Check the word IMPRECISION.
"Rollouts are an interesting way to evaluate a position. The interest is that it doesn't rely too much on the neural network so it can pick on the imprecision of the neural network. The draw back is that it relies on making a sample game so the results are not very accurate until a large number of games are played. As a result Rollouts are slow.
Here is a quick list of pros and cons of rollouts:
Pros There is no horizon effect as the games are played to the end The result is somehow dissociated with the neural network output It takes better into account cube vitality
Cons It can take quite long to get accurate results When using 1-ply evaluation for the cube, some decisions can be wrong and if such positions happen rather quickly from the initial position the result my be not accurate."
So what is going on in the truncated default analysis options in that they are getting some difficult positions correct. It is worth noting that the play options have been small in these disputed cases. We know the best play as per XG is true in this case, but I need a human interpretation of the position to silence the doubters. I got it wrong I looked for a play without understanding the position. Back to XG and dice distribution the roll odds mans companion. I think I have the answer but do not have the OTB knowledge to be confident as I don’t play regular furthermore not very well when I do.. Falafel could probably explain the logic. Those people who fret because the computer is better than them need to come into the real world.
The time has come for all those doubters to don the sack cloth. If you give XG the legs it will give you the answer the mistake is human error in not recognising the input XG requires.
Bottoms the SSD will not load my monster game.