[Design] Redesign for stats

[KYLChiu]

          I'll clean these up - mainly the design is so convoluted because I need them for SMP.

_Originally posted by @KYLChiu in https://github.com/KYLChiu/sporkfish/pull/97#discussion_r1473485325_

Current situation:

• Stats are not propagated to engine/tests, making it difficult to do any analysis.

• Stats are not well designed, it only contains nodes visited and no other information. Moreover it's convoluted - we have to pass in a dictionary and mutate it due to lazy SMP - we can't spawn a process agnostic object from within the class.

Requirements or points of discussion:

• Add visited type of nodes, ie differentiate between quiescence and negamax nodes, plus node properties like cutoff or returned by transposition table.

• Make this retrievable from engine and/or tests, which can form the basis of branching factor analysis in future.

• Let's have the Statistics class owns data structure(s) for now? Lazy SMP issue for later.

[ccjeremylo]

          I'll clean these up - mainly the design is so convoluted because I need them for SMP.

_Originally posted by @KYLChiu in #97 (comment)_

Current situation:

• Stats are not propagated to engine/tests, making it difficult to do any analysis.
• Stats are not well designed, it only contains nodes visited and no other information. Moreover it's convoluted - we have to pass in a dictionary and mutate it due to lazy SMP - we can't spawn a process agnostic object from within the class.

Requirements or points of discussion:

• Add visited type of nodes, ie differentiate between quiescence and negamax nodes, plus node properties like cutoff or returned by transposition table.
• Make this retrievable from engine and/or tests, which can form the basis of branching factor analysis in future.
• Let's have the Statistics class stats data structure(s) for now? Lazy SMP issue for later.

sorry, what is SMP? @KYLChiu

[KYLChiu]

Lazy Symmetric Multiprocessing

[ccjeremylo]

Discussed offline:

• To add visited type of nodes: naive plan is to implement a dict to carry the count of different nodes
• make self.nodes_visited protected, and add getter methods
• the self._statistics.increment() should be aware of what types of nodes visited

General comments - the statistics class would become very specific to negamax after our change, for now we just keep it as a negamax specific statistics class

• Decided we can just expose statistics at searcher, not the engine -> essentially just need a @property for getting statistics

Tests:

• we should add a test to check searcher can retrieve statistics
• This is useful for branching factor analysis in the future