def dev_by_ckpt():
model = MyModel(embedding_dim=768,
hidden_dim=300,
vocab_size_char=len(char2id),
vocab_size_bio=len(tag2id),
use_crf=True,
embeddings=char_embeddings,
)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state('D:/Expriment/model_output/ner_tool/lstm_crf/single_task/clue_ner/runs/1620802544/checkpoints')
if ckpt is None:
print ('Model not found, please train your model first')
else:
path = ckpt.model_checkpoint_path
print ('loading pre-trained model from %s.....' % path)
saver.restore(sess, path)
return sess, model
"./data/train/output.seq.bioattr",
w2i_char,
w2i_bio,
shuffling=True)
data_processor_valid = DataProcessor("./data/test/input.seq.char",
"./data/test/output.seq.bioattr",
w2i_char,
w2i_bio,
shuffling=True)
logger.info("building model...")
model = MyModel(embedding_dim=300,
hidden_dim=300,
vocab_size_char=len(w2i_char),
vocab_size_bio=len(w2i_bio),
use_crf=True)
logger.info("model params:")
params_num_all = 0
for variable in tf.trainable_variables():
params_num = 1
for dim in variable.shape:
params_num *= dim
params_num_all += params_num
logger.info("\t {} {} {}".format(variable.name, variable.shape,
params_num))
logger.info("all params num: " + str(params_num_all))
logger.info("start training...")
logger.info("Writing to {}\n".format(checkpoint_out_dir))
# Training
# ==================================================
with tf.Graph().as_default():
session_conf = tf.compat.v1.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
logger.info("building model...")
model = MyModel(
embedding_dim=768,
hidden_dim=300,
vocab_size_char=len(char2id),
vocab_size_bio=len(tag2id),
use_crf=True,
embeddings=char_embeddings,
)
# Define Training procedure
global_step = tf.Variable(0, name="global_step", trainable=False)
optimizer = tf.train.AdamOptimizer(FLAGS.learning_rate)
grads_and_vars = optimizer.compute_gradients(model.loss)
if FLAGS.use_clip_grad:
grads_and_vars = [[
tf.clip_by_value(g, FLAGS.clip_grad, FLAGS.clip_grad), v
] for g, v in grads_and_vars]
train_op = optimizer.apply_gradients(grads_and_vars,
global_step=global_step)