def load_pretrained_model(serialization_dir: str) -> models.Model:
"""
Given serialization directory, returns: model loaded with the pretrained weights.
"""
# Load Config
config_path = os.path.join(serialization_dir, "config.json")
model_path = os.path.join(serialization_dir, "model.ckpt.index")
model_files_present = all(
[os.path.exists(path) for path in [config_path, model_path]])
if not model_files_present:
raise Exception(
f"Model files in serialization_dir ({serialization_dir}) "
f" are missing. Cannot load_the_model.")
model_path = model_path.replace(".index", "")
with open(config_path, "r") as file:
config = json.load(file)
# Load Model
model_name = config.pop("type")
if model_name == "basic":
from model import MyBasicAttentiveBiGRU # To prevent circular imports
model = MyBasicAttentiveBiGRU(**config)
elif model_name == "advanced":
from model import MyAdvancedModel # To prevent circular imports
model = MyAdvancedModel(**config)
else:
raise Exception(f"model_name: {model_name} is not supported.")
model.load_weights(model_path)
return model
train_instances = read_instances(args.data_file, MAX_TOKENS)
print(f"\nReading Val Instances")
val_instances = read_instances(args.val_file, MAX_TOKENS)
with open(GLOVE_COMMON_WORDS_PATH) as file:
glove_common_words = [line.strip() for line in file.readlines() if line.strip()]
vocab_token_to_id, vocab_id_to_token = build_vocabulary(train_instances, VOCAB_SIZE,
glove_common_words)
train_instances = index_instances(train_instances, vocab_token_to_id)
val_instances = index_instances(val_instances, vocab_token_to_id)
vocab_size = len(vocab_token_to_id)
config = {'vocab_size': vocab_size, 'embed_dim': args.embed_dim, 'training': True, 'hidden_size': args.hidden_size}
model = MyBasicAttentiveBiGRU(**config)
config['type'] = 'basic'
optimizer = optimizers.Adam()
embeddings = load_glove_embeddings(args.embed_file, args.embed_dim, vocab_id_to_token)
model.embeddings.assign(tf.convert_to_tensor(embeddings))
save_serialization_dir = os.path.join('serialization_dirs', 'basic')
if not os.path.exists(save_serialization_dir):
os.makedirs(save_serialization_dir)
train_output = train(model, optimizer, train_instances, val_instances,
args.epochs, args.batch_size, save_serialization_dir)
config_path = os.path.join(save_serialization_dir, "config.json")