Feature builder

[iorch]

• Stone colour, 3. (Player stone / opponent stone / empty)
• Ones, 1.
• Turns since, 8. (How many turns since a move was played) (¿in adjacent points?)
• Liberties, 8. (Number of liberties (empty adjacent points))
• Capture, 8. (How many opponent stones would be captured) (¿in adjacent points?)
• Self-atari, 8. (How many of own stones would be captured)
• Liberties after move, 8. (Number of liberties after this move is played) (¿in adjacent points?)
• Ladder capture, 1. Whether a move at this point is a successful ladder capture
• Ladder escape, 1. Whether a move at this point is a successful ladder escape
• Sensibleness, 1. Whether a move is legal and does not fill its own eyes
• Zeros, 1.
• Current player color, 1. Whether current player is black

[iorch]

@magania creo que para varias de estas features tenemos que extender la clase board_layer, para que permita guardar un short y no sólo binario.

[magania]

@iorch Como para cuales features? En realidad yo me imagino que usan floats como input para las neural networks

[iorch]

En turns since, por ejemplo. por cada posición debe ir un entero,