def __init__(self, prefix=''):
self.prefix = prefix
self.save_dir = op.join(SAVE_DIR, prefix)
self.config_dir = op.join(CONFIG_DIR, prefix)
self.config_dict = {}
# list all directories
c_dirs = os.listdir(self.config_dir)
regex = r'config([0-9]+)'
config_list = [
re.search(regex, p)[1] for p in c_dirs if re.search(regex, p)
]
s_dirs = os.listdir(self.save_dir)
for c_idx in config_list:
c_idx = int(c_idx)
self.config_dict[c_idx] = {}
# read all config params and store them in a dict
self.config_dict[c_idx]['path'] = op.join(self.save_dir,
f'config{c_idx}')
config = ConfigReader(op.join(self.config_dir, f'config{c_idx}'))
for name, setting in config.settings.items():
value = setting.get_value()
self.config_dict[c_idx][name] = value
# add the special OCCLUDER param
self.config_dict[c_idx]['OCCLUDER'] = \
('occluder' in self.config_dict[c_idx]['EXPE'])
# did the run complete without error ?
self.config_dict[c_idx]['completed'] = 'yes'
if prefix:
# this means the results come from clusters and were computed
# with slurm, so we can read the error logs
err_log_path = op.join(self.save_dir, f'config{c_idx}_log.err')
with open(err_log_path, 'r') as errf:
error_message = errf.readlines()
if error_message:
self.config_dict[c_idx]['completed'] = 'no'
# check if model file and train data are present
files = os.listdir(self.config_dict[c_idx]['path'])
if 'model.pt' not in files:
self.config_dict[c_idx]['completed'] = 'no'
if 'train_data.hdf5' not in files:
self.config_dict[c_idx]['completed'] = 'no'
train_data = utl.load_dict_h5py(
op.join(self.config_dict[c_idx]['path'], 'train_data.hdf5'))
if len(train_data['energy']) != config.val('NUM_EPOCHS'):
print(f'config {c_idx}')
print(
f'length of train data ({len(train_data["energy"])}) does'
f'not match number of epochs ({config.val("NUM_EPOCHS")})')
self.config_dict[c_idx]['completed'] = 'partial'
args = parser.parse_args()
# num_args = len(sys.argv) - 1
# if num_args != 1:
# print('run.py accepts a single argument specifying the config file.')
# exit(1)
# Read the config file
# config = ConfigReader(sys.argv[1])
config_id = args.config
config = ConfigReader(f"configs/config{config_id}")
RELATIONAL = config.val("RELATIONAL")
RELATION_TYPE = config.val("RELATION_TYPE")
RECURRENT_TRANSITION = config.val("RECURRENT_TRANSITION")
TRAINING = config.val("TRAINING")
G_FUNC = config.val("G_FUNC")
HINGE = torch.tensor(config.val("HINGE"))
NUM_SLOTS = config.val("NUM_SLOTS")
SLOT_DIM = config.val("SLOT_DIM")
HIDDEN_DIM = config.val("HIDDEN_DIM")
NUM_HEADS = config.val("NUM_HEADS")
EXPE = config.val("EXPE")
NUM_EPOCHS = config.val("NUM_EPOCHS")
LEARNING_RATE = config.val("LEARNING_RATE")
BATCH_SIZE = config.val("BATCH_SIZE")