with tf.Session() as sess:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.bind((socket.gethostname(), 6666))
s.listen(1)
print("server ready")
while True:
c, _ = s.accept()
sock_func(c, sess, module)
# t = threading.Thread(target=sock_func, args=[c, sess, module])
# t.start()
def sock_func(sock, sess, module):
try:
while True:
data = sock.recv(4096).decode("utf-8")
print(data, len(data))
result = json.dumps(module.predict(data, sess),
ensure_ascii=False).encode("utf-8")
sock.send(result)
except Exception:
print("EXCEPTION")
sock.close()
if __name__ == '__main__':
args = argparse.Namespace()
args.save_path = "saves/wikipedia_150/"
args.load_from_file = True
ner_module = NER.NER(args)
open_socket(ner_module)
embeddings_file_path=None,
stacked_embeddings=stacked_embeddings)
model_params = {
"filter_width": 3,
"embeddings_dropout": True,
"n_filters": [256],
"dense_dropout": True,
"token_embeddings_dim": 300,
"char_embeddings_dim": 50,
"cell_type": 'lstm',
"use_batch_norm": True,
"concat_embeddings": True,
"use_crf": True,
"use_char_embeddins": True,
"net_type": 'rnn',
"use_capitalization": False,
}
net = NER.NER(corp, stacked_embeddings, **model_params)
learning_params = {
'dropout_rate': 0.5,
'epochs': 200,
'learning_rate': 0.001, # 0.0003
'batch_size': 20,
'learning_rate_decay': 0.94
}
results = net.fit(**learning_params)
