def convert_to_cheating_data(data):
"""
:param data: format is SelfPlayWorker.buffer
:return:
"""
state_list = []
policy_list = []
value_list = []
for state_fen, policy, value in data:
state_planes = canon_input_planes(state_fen)
if is_black_turn(state_fen):
policy = Config.flip_policy(policy)
move_number = int(state_fen.split(' ')[5])
value_certainty = min(
5, move_number
) / 5 # reduces the noise of the opening... plz train faster
sl_value = value * value_certainty + testeval(
state_fen, False) * (1 - value_certainty)
state_list.append(state_planes)
policy_list.append(policy)
value_list.append(sl_value)
return np.asarray(state_list, dtype=np.float32), np.asarray(
policy_list, dtype=np.float32), np.asarray(value_list,
dtype=np.float32)
def convert_to_cheating_data(data):
"""
:param data: format is SelfPlayWorker.buffer
:return:
"""
state_list = []
policy_list = []
value_list = []
for state_fen, policy, value in data:
state_planes = canon_input_planes(state_fen)
if is_black_turn(state_fen):
policy = Config.flip_policy(policy)
move_number = int(state_fen.split(' ')[5])
value_certainty = min(5, move_number)/5 # reduces the noise of the opening... plz train faster
sl_value = value*value_certainty + testeval(state_fen, False)*(1-value_certainty)
state_list.append(state_planes)
policy_list.append(policy)
value_list.append(sl_value)
return np.asarray(state_list, dtype=np.float32), np.asarray(policy_list, dtype=np.float32), np.asarray(value_list, dtype=np.float32)