[computer-go] batch size and impact on strength

[Magnus Persson]

Quoting Carter Cheng <carter_cheng at yahoo.com>:

> Is UCT really that good at finding the best move in alot of  situations?

The search algorithms do not find the best move. They prune the bad  
ones. The quicker and more reliably bad moves can be pruned the less  
candidates remain to be chosen from. Thus increasing the likelyhood of  
playing the best one.

When you write UCT you also have to separate the actual knowledge  
about go given to the program in heavy playouts from the search  
process. The search can be blind or it can have eyes to see with,  
nethertheless a good search algorithm do well given the knowledge it  
has.

Yet, the separation of search and knowledge is not that simple. I  
believe the best parameters for the search is likely to depend on the  
quality of moves. One example is that Mogo and Valkyria for example  
simply follows a greedy tree search, becuase build in knowledge and  
online learned knowledge (AMAF) allows the search to search very  
selectively.


-Magnus