def data_augmentation(x_label, y_label):
"""
Data augmentation.
:param x_label: x_label Input of the neural network.
:param y_label: x_label Output of the neural network.
:return: x_label, y_label. X_label, y_label after data augmentation.
"""
all_action_probs, values = y_label
extend_xlabel = []
extend_ylabel_action_probs = []
extend_ylabel_value = []
for board_input in x_label:
# The board_input 4-dimensional data is disassembled and expanded separately.
augmentation_board = np.array([
get_data_augmentation(one_board_input)
for one_board_input in board_input
])
board_augmentation = np.array(list(zip(*augmentation_board)))
extend_xlabel.extend(
np.array(
[one_augmentation for one_augmentation in board_augmentation]))
for action_probs in all_action_probs:
extend_action_probs = get_data_augmentation(
action_probs.reshape(BOARD.board_size, BOARD.board_size),
operation=lambda a: a.flatten())
extend_ylabel_action_probs.extend(extend_action_probs)
for value in values:
extend_value = get_data_augmentation(np.array(value))
extend_ylabel_value.extend(extend_value)
return extend_xlabel, (extend_ylabel_action_probs, extend_ylabel_value)
def data_augmentation(x_label, y_label):
"""
:param x_label:
:param y_label:
:return:
"""
all_action_probs, values = y_label
extend_xlabel = []
extend_ylabel_action_probs = []
extend_ylabel_value = []
for board_input in x_label:
augmentation_board = np.array([get_data_augmentation(one_board_input) for one_board_input in board_input])
board_augmentation = np.array(list(zip(*augmentation_board)))
extend_xlabel.extend(np.array([one_augmentation for one_augmentation in board_augmentation]))
for action_probs in all_action_probs:
extend_action_probs = get_data_augmentation(action_probs.reshape(BOARD.board_size, BOARD.board_size),
operation=lambda a: a.flatten())
extend_ylabel_action_probs.extend(extend_action_probs)
for value in values:
extend_value = get_data_augmentation(np.array(value))
extend_ylabel_value.extend(extend_value)
return extend_xlabel, (extend_ylabel_action_probs, extend_ylabel_value)