def __init__(self, config):
print('--- create model ---')
config['entity_embeddings'] = config['entity_embeddings'] / \
np.maximum(np.linalg.norm(config['entity_embeddings'],
axis=1, keepdims=True), 1e-12)
config['entity_embeddings'][config['entity_voca'].unk_id] = 1e-10
config['word_embeddings'] = config['word_embeddings'] / \
np.maximum(np.linalg.norm(config['word_embeddings'],
axis=1, keepdims=True), 1e-12)
config['word_embeddings'][config['word_voca'].unk_id] = 1e-10
print('prerank model')
self.prerank_model = ntee.NTEE(config)
self.args = config['args']
print('main model')
if self.args.mode == 'eval':
print('try loading model from', self.args.model_path)
self.model = load_model(self.args.model_path, ModelClass)
else:
print('create new model')
if config['mulrel_type'] == 'rel-norm':
config['use_stargmax'] = False
if config['mulrel_type'] == 'ment-norm':
config['first_head_uniform'] = False
config['use_pad_ent'] = True
config['use_local'] = True
config['use_local_only'] = False
config['oracle'] = False
self.model = ModelClass(config)
self.prerank_model.cuda()
self.model.cuda()
def __init__(self, config):
print('---- create ed rank model ----')
# maximum 取最大值
# np.linalg.norm 求矩阵或者向量的范数，axis=1表示按行向量处理，求多个行向量的范数，keepding：是否保持矩阵的二维特性 True表示保持矩阵的二维特性
config['entity_embeddings'] = config['entity_embeddings'] / \
np.maximum(np.linalg.norm(config['entity_embeddings'],
axis=1, keepdims=True), 1e-12)
config['entity_embeddings'][config['entity_voca'].unk_id] = 1e-10
config['word_embeddings'] = config['word_embeddings'] / \
np.maximum(np.linalg.norm(config['word_embeddings'],
axis=1, keepdims=True), 1e-12)
config['word_embeddings'][config['word_voca'].unk_id] = 1e-10
self.args = config['args']
if self.args.mode == 'prerank':
self.ent_net = config['ent_net']
print('prerank model')
self.prerank_model = ntee.NTEE(config)
self.prerank_model.cuda()
print('main model')
if self.args.mode in {'eval', 'ed'}:
print('try loading model from', self.args.model_path)
self.model = load_model(self.args.model_path, ModelClass)
else:
try:
print('try loading model from', self.args.model_path)
self.model = load_model(self.args.model_path, ModelClass)
except:
print('create new model')
if config['mulrel_type'] == 'rel-norm':
config['use_stargmax'] = False
if config['mulrel_type'] == 'ment-norm':
config['first_head_uniform'] = False
config['use_pad_ent'] = True
config['use_local'] = True
config['use_local_only'] = False
config['oracle'] = False
self.model = ModelClass(config)
self.model.cuda()