def extract_game_data(gcs_path, root_node):
props = root_node.properties
komi = float(sgf_wrapper.sgf_prop(props.get('KM')))
result = sgf_wrapper.sgf_prop(props.get('RE', ''))
board_size = int(sgf_wrapper.sgf_prop(props.get('SZ')))
value = utils.parse_game_result(result)
was_resign = '+R' in result
filename = os.path.basename(gcs_path)
filename_no_ext, _ = os.path.splitext(filename)
# BigQuery's TIMESTAMP() takes in unix millis.
completion_millis = 1000 * int(filename_no_ext.split('-')[0])
worker_id = filename_no_ext.split('-')[-1]
model_num = shipname.detect_model_num(props.get('PW')[0])
sgf_url = gcs_path
first_comment_node_lines = root_node.next.properties['C'][0].split('\n')
# in-place edit to comment node so that first move's comment looks
# the same as all the other moves.
root_node.next.properties['C'][0] = '\n'.join(first_comment_node_lines[1:])
resign_threshold = float(first_comment_node_lines[0].split()[-1])
return {
'worker_id': worker_id,
'completed_time': completion_millis,
'board_size': board_size,
'model_num': model_num,
'result_str': result,
'value': value,
'was_resign': was_resign,
'sgf_url': sgf_url,
'resign_threshold': resign_threshold,
}
def parse_sgf(sgf_path):
with open(sgf_path) as f:
sgf_contents = f.read()
collection = sgf.parse(sgf_contents)
game = collection.children[0]
props = game.root.properties
assert int(sgf_prop(props.get('GM', ['1']))) == 1, "Not a Go SGF!"
result = utils.parse_game_result(sgf_prop(props.get('RE')))
positions, moves = zip(*[(p.position, p.next_move) for p in sgf_wrapper.replay_sgf(sgf_contents)])
return positions, moves, result, props