def train_epoch(self, train_set, valid_set, epoch):
num_batches = len(train_set)
prog = Progbar(target=num_batches)
for i, batch_data in enumerate(train_set):
feed_dict = self._get_feed_dict(batch_data,
is_train=True,
keep_prob=self.cfg["keep_prob"],
lr=self.cfg["lr"])
_, train_loss, summary = self.sess.run(
[self.train_op, self.loss, self.summary], feed_dict=feed_dict)
cur_step = (epoch - 1) * num_batches + (i + 1)
prog.update(i + 1, [("Global Step", int(cur_step)),
("Train Loss", train_loss)])
if i % 100 == 0:
self.train_writer.add_summary(summary, cur_step)
step = cur_step
for j, batch_data in enumerate(valid_set):
feed_dict = self._get_feed_dict(batch_data)
val_summary = self.sess.run(self.summary, feed_dict=feed_dict)
self.test_writer.add_summary(val_summary, step)
micro_f_val, out_str, micro = self.evaluate_punct(valid_set, "val")
return micro_f_val, train_loss
if not os.path.exists(config.ckpt):
os.makedirs(config.ckpt)
# training the model
print("start training...")
sess_config = tf.ConfigProto(allow_soft_placement=True)
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
# build model
print("build model...")
model = Model(config, verb_vocab_count)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver(max_to_keep=1)
for epoch in range(config.epochs):
prog = Progbar(target=int(2.663e8) / config.batch_size)
for i, data in enumerate(
dataset_iterator(config.dataset, word_dict, verb_dict,
config.batch_size)):
feed_dict = model.get_feed_dict(data, is_train=True, lr=config.lr)
_, losses = sess.run([model.train_op, model.loss],
feed_dict=feed_dict)
prog.update(i + 1, [("train loss", losses)])
# save the model
saver.save(sess,
config.ckpt + config.model_name,
global_step=config.epochs)
# save the trained target embedding
target_emb = sess.run(model.verb_embeddings)
np.savez_compressed("data/trained_target_emb.npz", embeddings=target_emb)