def main():
wandb.init(project="msc_thesis_hendrig")
ap = argparse.ArgumentParser()
ap.add_argument("data_path", help="Path to data root")
ap.add_argument("vocabulary_path", help="Path to vocabulary file")
ap.add_argument("config", help="Path to config file")
ap.add_argument("-m",
"--models",
help="Directory to store trained models (optional)")
ap.add_argument("-l",
"--log",
help="Path to store training log (optional)")
ap.add_argument("-e",
"--eval_only",
help="Whether to run just the final model evaluation")
args = ap.parse_args()
config = yaml.safe_load(open(args.config))
wandb.config.update(args)
wandb.config.update(config)
lowest_memory_gpu = gpu_selector.GPUSelector().pick_gpu_lowest_memory()
if lowest_memory_gpu is not None:
os.environ["CUDA_VISIBLE_DEVICES"] = str(lowest_memory_gpu)
print("Training with configuration:", config)
data = data_loader.DataLoader(args.data_path, config["data"],
vocabulary.Vocabulary(args.vocabulary_path))
if args.eval_only:
if args.models is None or args.log is None:
raise ValueError(
"Must provide a path to pre-trained models when running final evaluation"
)
test(data, config, args.models, args.log)
else:
train(data, config, args.models, args.log)
def main():
ap = argparse.ArgumentParser()
ap.add_argument("data_path", help="Path to data root")
ap.add_argument("vocabulary_path", help="Path to vocabulary file")
ap.add_argument("config", help="Path to config file")
ap.add_argument("-m",
"--models",
help="Directory to store trained models (optional)")
ap.add_argument("-l",
"--log",
help="Path to store training log (optional)")
ap.add_argument("-e",
"--eval_only",
help="Whether to run just the final model evaluation")
args = ap.parse_args()
config = yaml.safe_load(open(args.config))
print("Training with configuration:", config)
data = data_loader.DataLoader(args.data_path, config["data"],
vocabulary.Vocabulary(args.vocabulary_path))
if args.eval_only:
if args.models is None or args.log is None:
raise ValueError(
"Must provide a path to pre-trained models when running final evaluation"
)
test(data, config, args.models, args.log)
else:
train(data, config, args.models, args.log)
def vocabulary(self) -> data_vocab.Vocabulary:
if not hasattr(self, '_vocabulary'):
_vocabulary = data_vocab.Vocabulary(
vocab_fname=None,
vocab_dict=self._dictionary.vocab,
num_oov_buckets=1,
unk_token=data_vocab._BERT_UNK)
self._vocabulary = _vocabulary
return self._vocabulary