overrides = {}
if args.device is not None:
overrides["trainer"] = {"cuda_device": args.device}
if args.lazy is not None:
overrides["dataset_reader"] = {"lazy": args.lazy}
configs.append(Params(overrides))
for config_file in args.config:
configs.append(Params.from_file(config_file))
configs.append(Params.from_file(args.base_config))
else:
serialization_dir = args.resume
configs.append(
Params.from_file(os.path.join(serialization_dir, "config.json")))
train_params = util.merge_configs(configs)
if "vocabulary" in train_params:
# Remove this key to make AllenNLP happy
train_params["vocabulary"].pop("non_padded_namespaces", None)
predict_params = train_params.duplicate()
import_submodules("udify")
try:
util.cache_vocab(train_params)
train_model(train_params, serialization_dir, recover=bool(args.resume))
except KeyboardInterrupt:
logger.warning("KeyboardInterrupt, skipping training")
dev_file = predict_params["validation_data_path"]
log_dir_name = args.name
if not log_dir_name:
file_name = args.dataset_config if args.dataset_config else args.parameters_config
log_dir_name = os.path.basename(file_name).split(".")[0]
if not args.resume:
serialization_dir = os.path.join(
"logs", log_dir_name,
datetime.datetime.now().strftime("%Y.%m.%d_%H.%M.%S"))
overrides = {}
if args.device is not None:
overrides["trainer"] = {"cuda_device": args.device}
if args.lazy is not None:
overrides["dataset_reader"] = {"lazy": args.lazy}
train_params = util.merge_configs(args.parameters_config,
args.dataset_config, overrides)
#else:
# serialization_dir = args.resume
# train_params = Params.from_file(os.path.join(serialization_dir, "config.json"))
#TODO override stuff!
if "vocabulary" in train_params:
# Remove this key to make AllenNLP happy
train_params["vocabulary"].pop("non_padded_namespaces", None)
predict_params = train_params.duplicate()
import_submodules("udify")
try:
util.cache_vocab(train_params)