import argparse
parser = argparse.ArgumentParser(description='Compare trained model with random and perfect AI')
parser.add_argument('--path', metavar='N', type=str, nargs='?', default=None, help='working path')
parser.add_argument('--count', metavar='N', type=int, nargs='?', default=None, help='compare count')
args = parser.parse_args()
arg_dict = vars(args)
if arg_dict['path'] == None:
scan_dir = os.path.join('output',MY_NAME)
filename_list = os.listdir(scan_dir)
if len(filename_list) <= 0:
raise '{} is empty'.format(scan_dir)
filename_int_list = [util.to_int(filename,-1) for filename in filename_list]
arg_timestamp = str(max(filename_int_list))
arg_dict['path'] = os.path.join(scan_dir,str(arg_timestamp))
if arg_dict['count'] == None:
arg_dict['count'] = 0
while os.path.isfile(os.path.join(os.path.join(arg_dict['path'],'deeplearn','sess','{}.index'.format((arg_dict['count']+1)*1000)))):
arg_dict['count'] += 1
logging.info('QLCMICZK arg_dict {}'.format(json.dumps(arg_dict)))
with open(os.path.join(arg_dict['path'],'deeplearn','input_arg_dict.json'),'r') as deeplearn_arg_dict_file:
deeplearn_arg_dict = json.load(deeplearn_arg_dict_file)
vs_dict_meta_list = [
{
metavar='N',
type=int,
nargs='?',
default=None,
help='compare filename')
args = parser.parse_args()
arg_dict = vars(args)
if arg_dict['filename'] == None:
dlcompare_path = os.path.join('output', MY_NAME, 'dlcompare')
filename_list = os.listdir(dlcompare_path)
if len(filename_list) <= 0:
raise '{} is empty'.format(dlcompare_path)
filename_int_list = [
myutil.to_int(filename[:-5], -1) for filename in filename_list
if filename.endswith('.json')
]
arg_dict['filename'] = os.path.join(
dlcompare_path, '{}.json'.format(max(filename_int_list)))
data_dict = {}
with open(arg_dict['filename'], 'r') as in_file:
result_list = json.load(in_file)
win_key_list = [
{
'type': 'win',
'name': 'pd',
'win': 'X',
'style': 'ro'
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Show graph')
parser.add_argument('filename', metavar='N', type=int, nargs='?', default=None, help='compare filename')
args = parser.parse_args()
arg_dict = vars(args)
if arg_dict['filename'] == None:
dl_path = os.path.join('output',MY_NAME)
dl_filename_list = os.listdir(dl_path)
if len(dl_filename_list) <= 0:
raise '{} is empty'.format(dl_path)
dl_filename_int_list = [myutil.to_int(filename,-1) for filename in dl_filename_list]
dl_filename_int = max(dl_filename_int_list)
dc_path = os.path.join(dl_path,str(dl_filename_int),'dlcompare')
dc_filename_list = os.listdir(dc_path)
dc_filename_int_list = [myutil.to_int(filename[:-5],-1) for filename in dc_filename_list if filename.endswith('.json')]
dc_filename_int = max(dc_filename_int_list)
arg_dict['filename'] = os.path.join(dc_path,'{}.json'.format(dc_filename_int))
data_dict = {}
with open(arg_dict['filename'],'r') as in_file:
result_list = json.load(in_file)
itr_count = 0
for result in result_list:
if not 'itr' in result['input']:
def main(_):
argparser = argparse.ArgumentParser()
argparser.add_argument(
'--output_path',
type=str,
help='The path to which checkpoints and other outputs '
'should be saved. This can be either a local or GCS '
'path.',
default=None
)
argparser.add_argument(
'--random_stddev',
type=float,
help='random_stddev',
default=0.1
)
argparser.add_argument(
'--random_move_chance',
type=float,
help='RANDOM_MOVE_CHANCE',
default=0.05
)
argparser.add_argument(
'--train_beta',
type=float,
help='TRAIN_BETA',
default=0.99
)
argparser.add_argument(
'--turn_count',
type=int,
help='turn_count',
default=None
)
argparser.add_argument(
'--device',
type=str,
help='device',
default=None
)
argparser.add_argument(
'--continue',
action='store_true',
help='continue'
)
# argparser.add_argument(
# '--element_l2_factor',
# type=float,
# help='ELEMENT_L2_FACTOR',
# default=10.0
# )
# argparser.add_argument(
# '--l2_weight',
# type=float,
# help='L2_WEIGHT',
# default=0.1
# )
argparser.add_argument(
'--train_memory',
type=int,
help='TRAIN_MEMORY',
default=10000
)
argparser.add_argument(
'--dropout_keep_prob',
type=float,
help='dropout_keep_prob',
default=0.5
)
args, _ = argparser.parse_known_args()
arg_dict = vars(args)
logging.info('YGYMBFMN arg_dict {}'.format(json.dumps(arg_dict)))
if(arg_dict['output_path']==None):
if not arg_dict['continue']:
timestamp = int(time.time())
arg_dict['output_path'] = os.path.join('output',MY_NAME,str(timestamp),'deeplearn')
else:
filename_list = os.listdir(os.path.join('output',MY_NAME))
filename_int_list = [util.to_int(filename,-1) for filename in filename_list]
arg_timestamp = str(max(filename_int_list))
arg_dict['output_path'] = os.path.join('output',MY_NAME,str(arg_timestamp),'deeplearn')
if arg_dict['continue']:
with open(os.path.join(arg_dict['output_path'],'input_arg_dict.json'),'r') as deeplearn_arg_dict_file:
deeplearn_arg_dict = json.load(deeplearn_arg_dict_file)
arg_dict = deeplearn_arg_dict
arg_dict['continue'] = True
else:
py23.makedirs(arg_dict['output_path'],exist_ok=True)
with open(os.path.join(arg_dict['output_path'],'input_arg_dict.json'),'w') as out_file:
json.dump(arg_dict,out_file)
with tf.device(arg_dict['device']):
# with tf.device('/gpu:0'):
game = Game()
dl = DeepLearn(arg_dict)
po = DLPlayer(dl)
po.set_side(tttl.Pid.O)
px = DLPlayer(dl)
px.set_side(tttl.Pid.X)
game.setPlayer(tttl.Pid.O, po)
game.setPlayer(tttl.Pid.X, px)
game.run(turn_count=arg_dict['turn_count'],p=logging.debug)
def main(_):
argparser = argparse.ArgumentParser()
argparser.add_argument(
'--output_path',
type=str,
help='The path to which checkpoints and other outputs '
'should be saved. This can be either a local or GCS '
'path.',
default=None
)
argparser.add_argument(
'--random_stddev',
type=float,
help='random_stddev',
default=0.1
)
argparser.add_argument(
'--random_move_chance',
type=float,
help='RANDOM_MOVE_CHANCE',
default=0.05
)
argparser.add_argument(
'--train_beta',
type=float,
help='TRAIN_BETA',
default=0.99
)
argparser.add_argument(
'--turn_count',
type=int,
help='turn_count',
default=None
)
argparser.add_argument(
'--device',
type=str,
help='device',
default=None
)
argparser.add_argument(
'--continue',
action='store_true',
help='continue'
)
# argparser.add_argument(
# '--element_l2_factor',
# type=float,
# help='ELEMENT_L2_FACTOR',
# default=10.0
# )
# argparser.add_argument(
# '--l2_weight',
# type=float,
# help='L2_WEIGHT',
# default=0.1
# )
argparser.add_argument(
'--train_memory',
type=int,
help='TRAIN_MEMORY',
default=10000
)
args, _ = argparser.parse_known_args()
arg_dict = vars(args)
logging.info('YGYMBFMN arg_dict {}'.format(json.dumps(arg_dict)))
if(arg_dict['output_path']==None):
if not arg_dict['continue']:
timestamp = int(time.time())
arg_dict['output_path'] = os.path.join('output',MY_NAME,str(timestamp),'deeplearn')
else:
filename_list = os.listdir(os.path.join('output',MY_NAME))
filename_int_list = [util.to_int(filename,-1) for filename in filename_list]
arg_timestamp = str(max(filename_int_list))
arg_dict['output_path'] = os.path.join('output',MY_NAME,str(arg_timestamp),'deeplearn')
if arg_dict['continue']:
with open(os.path.join(arg_dict['output_path'],'input_arg_dict.json'),'r') as deeplearn_arg_dict_file:
deeplearn_arg_dict = json.load(deeplearn_arg_dict_file)
arg_dict = deeplearn_arg_dict
arg_dict['continue'] = True
else:
py23.makedirs(arg_dict['output_path'],exist_ok=True)
with open(os.path.join(arg_dict['output_path'],'input_arg_dict.json'),'w') as out_file:
json.dump(arg_dict,out_file)
with tf.device(arg_dict['device']):
# with tf.device('/gpu:0'):
game = Game()
dl = DeepLearn(arg_dict)
po = DLPlayer(dl)
po.set_side(tttl.Pid.O)
px = DLPlayer(dl)
px.set_side(tttl.Pid.X)
game.setPlayer(tttl.Pid.O, po)
game.setPlayer(tttl.Pid.X, px)
game.run(turn_count=arg_dict['turn_count'],p=logging.debug)
type=int,
nargs='?',
default=None,
help='compare filename')
args = parser.parse_args()
arg_dict = vars(args)
if arg_dict['filename'] == None:
dl_path = os.path.join('output', MY_NAME)
dl_filename_list = os.listdir(dl_path)
if len(dl_filename_list) <= 0:
raise '{} is empty'.format(dl_path)
dl_filename_int_list = [
myutil.to_int(filename, -1) for filename in dl_filename_list
]
dl_filename_int = max(dl_filename_int_list)
dc_path = os.path.join(dl_path, str(dl_filename_int), 'dlcompare')
dc_filename_list = os.listdir(dc_path)
dc_filename_int_list = [
myutil.to_int(filename[:-5], -1) for filename in dc_filename_list
if filename.endswith('.json')
]
dc_filename_int = max(dc_filename_int_list)
arg_dict['filename'] = os.path.join(dc_path,
'{}.json'.format(dc_filename_int))
data_dict = {}
with open(arg_dict['filename'], 'r') as in_file:
metavar='N',
type=int,
nargs='?',
default=None,
help='compare count')
args = parser.parse_args()
arg_dict = vars(args)
if arg_dict['path'] == None:
scan_dir = os.path.join('output', MY_NAME)
filename_list = os.listdir(scan_dir)
if len(filename_list) <= 0:
raise '{} is empty'.format(scan_dir)
filename_int_list = [
util.to_int(filename, -1) for filename in filename_list
]
arg_timestamp = str(max(filename_int_list))
arg_dict['path'] = os.path.join(scan_dir, str(arg_timestamp))
if arg_dict['count'] == None:
arg_dict['count'] = 0
while os.path.isfile(
os.path.join(
os.path.join(
arg_dict['path'], 'deeplearn', 'sess',
'{}.index'.format((arg_dict['count'] + 1) * 1000)))):
arg_dict['count'] += 1
logging.info('QLCMICZK arg_dict {}'.format(json.dumps(arg_dict)))