design document.txt

Trabalho prático Reversi
- MinMax com alpha beta
	Decisão sobre avaliar um nó:
		Minimum Depth
		Maximum Depth
		Heuristic eval
			Uses MLP (?) to weigh the different heuristics
		Time elapsed	
			Experimentally verify the ratio between forward and backpropagation time
				Run many experiments, and establish a confidence interval
			Check how much time it takes to check the time every time a new state is expanded
				Compute by comparing total time for a fixed depth with and without verifier
		Compute parameters through experimentation
			Other programs found on the net
			Tulios wining program from last semester
			Itself, with more time
	Keep track of alpha beta
	Recursive implementation (if we want)
		Each recursion stores the best move from its children, and then returns it, allong with the evaluation function (standard alpha/beta)
- Initial plays
	Lookup table
		Optional: Add our own entries by precomputing large depths
		Binary file (if there are many)
		Hash input
			Board size
			Purely string
			If everything is computed up to a given depth, just use an array (maybe)	
- Kernel/representação
	structure
		char array[8][8],
	Enum class com char {-1 : black, 0 : empty, 1 : white }
	Print configuration
	load from string
	load from file
	expand
		Generates children for the player that currently has the go
		Returns tuples of movement, board	
- Heuristics and meta-heuristics
	Weights/Parameters
	Heuristics:
		Weights for each position
			Multiple boards
				Multiplex between them with metaheuristic, maybe
		Number of pieces on the board, by gameplay time
		Number of unflippable pieces
			Extra weight for corners
		Relative frontier size
		Number of pieces each player has
			Very important at end game
			Not so important at start	
		Check interwebs
		Check tulio
	Metaheuristics	
		MLP
			May be costly
		Perceptron
			Simple
		Agebraic
			Easily implement piece count x game time importance
		Check web for ways to weigh heuristics
			With NN
			With models
			etc
	Experimental parametrization
		As discussed above
- Interfaces com o mundo
	- Jogar humano/ia
	- Interface com competição
	- Makefile, main