#71 Don't use playout policy in prior knowledge

[Anonymous]

Originally created by: enz
Originally owned by: somebody

Currently, GoUctDefaultPriorKnowlede uses GoUctPlayoutPolicy and GoUctPatterns for positively initializing moves that the playout policy would have played and global instances of the playout patterns.

This makes it very difficult to improve the playout policy, because ever change to the policy and patterns affects both the playout policy and the prior knowledge and what is good for one is not necessarily good for the other.

I propose thst the prior knowledge does not use the playout policy anymore, but computes the features like capture and atari defense moves by itself. There should also be a new class GoUctKnowledgePatterns that can be changed independently from the playout patterns. Because the prior knowledge is not as time-critical as the playout, it could even use a pattern matcher that supports patterns larger than 3x3.