def main(_):
vocab_word2index, _ = create_or_load_vocabulary(
FLAGS.data_path,
FLAGS.mask_lm_source_file,
FLAGS.vocab_size,
test_mode=FLAGS.test_mode,
tokenize_style=FLAGS.tokenize_style)
vocab_size = len(vocab_word2index)
print("bert_pertrain_lm.vocab_size:", vocab_size)
index2word = {v: k for k, v in vocab_word2index.items()}
#train,valid,test=mask_language_model(FLAGS.mask_lm_source_file,FLAGS.data_path,index2word,max_allow_sentence_length=FLAGS.max_allow_sentence_length,test_mode=FLAGS.test_mode)
train, valid, test = mask_language_model(
FLAGS.mask_lm_source_file,
FLAGS.data_path,
index2word,
max_allow_sentence_length=FLAGS.max_allow_sentence_length,
test_mode=FLAGS.test_mode,
process_num=FLAGS.process_num)
train_X, train_y, train_p = train
valid_X, valid_y, valid_p = valid
test_X, test_y, test_p = test
print("length of training data:", train_X.shape, ";train_Y:",
train_y.shape, ";train_p:", train_p.shape, ";valid data:",
valid_X.shape, ";test data:", test_X.shape)
# 1.create session.
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
#Instantiate Model
config = set_config(FLAGS, vocab_size, vocab_size)
model = BertModel(config)
#Initialize Save
saver = tf.train.Saver()
if os.path.exists(FLAGS.ckpt_dir + "checkpoint"):
print("Restoring Variables from Checkpoint.")
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.ckpt_dir))
for i in range(2): #decay learning rate if necessary.
print(i, "Going to decay learning rate by half.")
sess.run(model.learning_rate_decay_half_op)
else:
print('Initializing Variables')
sess.run(tf.global_variables_initializer())
if FLAGS.use_pretrained_embedding:
vocabulary_index2word = {
index: word
for word, index in vocab_word2index.items()
}
assign_pretrained_word_embedding(
sess, vocabulary_index2word, vocab_size,
FLAGS.word2vec_model_path, model.embedding,
config.d_model) # assign pretrained word embeddings
curr_epoch = sess.run(model.epoch_step)
# 2.feed data & training
number_of_training_data = len(train_X)
print("number_of_training_data:", number_of_training_data)
batch_size = FLAGS.batch_size
iteration = 0
score_best = -100
for epoch in range(curr_epoch, FLAGS.num_epochs):
loss_total_lm, counter = 0.0, 0
for start, end in zip(
range(0, number_of_training_data, batch_size),
range(batch_size, number_of_training_data, batch_size)):
iteration = iteration + 1
if epoch == 0 and counter == 0:
print("trainX[start:end]:", train_X[start:end],
"train_X.shape:", train_X.shape)
feed_dict = {
model.x_mask_lm: train_X[start:end],
model.y_mask_lm: train_y[start:end],
model.p_mask_lm: train_p[start:end],
model.dropout_keep_prob: FLAGS.dropout_keep_prob
}
current_loss_lm, lr, l2_loss, _ = sess.run([
model.loss_val_lm, model.learning_rate, model.l2_loss_lm,
model.train_op_lm
], feed_dict)
loss_total_lm, counter = loss_total_lm + current_loss_lm, counter + 1
if counter % 30 == 0:
print(
"%d\t%d\tLearning rate:%.5f\tLoss_lm:%.3f\tCurrent_loss_lm:%.3f\tL2_loss:%.3f\t"
% (epoch, counter, lr, float(loss_total_lm) /
float(counter), current_loss_lm, l2_loss))
if start != 0 and start % (800 *
FLAGS.batch_size) == 0: # epoch!=0
loss_valid, acc_valid = do_eval(sess, model, valid,
batch_size)
print(
"%d\tValid.Epoch %d ValidLoss:%.3f\tAcc_valid:%.3f\t" %
(counter, epoch, loss_valid, acc_valid * 100))
# save model to checkpoint
if acc_valid > score_best:
save_path = FLAGS.ckpt_dir + "model.ckpt"
print("going to save check point.")
saver.save(sess, save_path, global_step=epoch)
score_best = acc_valid
sess.run(model.epoch_increment)
def main(_):
# 1.load vocabulary of token from cache file save from pre-trained stage; load label dict from training file; print some message.
vocab_word2index, _ = create_or_load_vocabulary(
FLAGS.data_path,
FLAGS.training_data_file,
FLAGS.vocab_size,
test_mode=FLAGS.test_mode,
tokenize_style=FLAGS.tokenize_style)
# label2index=get_lable2index(FLAGS.data_path,FLAGS.training_data_file, tokenize_style=FLAGS.tokenize_style)
label2index = {'0': 0, '1': 1, '2': 2, '3': 3}
vocab_size = len(vocab_word2index)
print("cnn_model.vocab_size:", vocab_size)
num_classes = len(label2index)
print("num_classes:", num_classes)
# load training data.
train, valid, test = load_data_multilabel(
FLAGS.data_path,
FLAGS.training_data_file,
FLAGS.valid_data_file,
FLAGS.test_data_file,
vocab_word2index,
label2index,
FLAGS.sequence_length,
process_num=FLAGS.process_num,
test_mode=FLAGS.test_mode,
tokenize_style=FLAGS.tokenize_style)
train_X, train_Y = train
valid_X, valid_Y = valid
test_X, test_Y = test
print("test_model:", FLAGS.test_mode, ";length of training data:",
train_X.shape, ";valid data:", valid_X.shape, ";test data:",
test_X.shape, ";train_Y:", train_Y.shape)
# 2.create session.
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
#Instantiate Model
config = set_config(FLAGS, num_classes, vocab_size)
model = BertModel(config)
#Initialize Save
saver = tf.train.Saver()
#if os.path.exists(FLAGS.ckpt_dir+"checkpoint"):
if True:
print("Restoring Variables from Checkpoint.")
sess.run(tf.global_variables_initializer())
for i in range(6): #decay learning rate if necessary.
print(
i, "Going to decay learning rate by a factor of " +
str(FLAGS.decay_rate))
sess.run(model.learning_rate_decay_half_op)
# restore those variables that names and shapes exists in your model from checkpoint. for detail check: https://gist.github.com/iganichev/d2d8a0b1abc6b15d4a07de83171163d4
optimistic_restore(
sess, tf.train.latest_checkpoint(FLAGS.ckpt_dir)
) #saver.restore(sess,tf.train.latest_checkpoint(FLAGS.ckpt_dir))
else:
print('Initializing Variables as model instance is not exist.')
sess.run(tf.global_variables_initializer())
if FLAGS.use_pretrained_embedding:
vocabulary_index2word = {
index: word
for word, index in vocab_word2index.items()
}
assign_pretrained_word_embedding(
sess, vocabulary_index2word, vocab_size,
FLAGS.word2vec_model_path, model.embedding,
config.d_model) # assign pretrained word embeddings
curr_epoch = sess.run(model.epoch_step)
# 3.feed data & training
number_of_training_data = len(train_X)
batch_size = FLAGS.batch_size
iteration = 0
score_best = -100
f1_score = 0
epoch = 0
for epoch in range(curr_epoch, FLAGS.num_epochs):
loss_total, counter = 0.0, 0
for start, end in zip(
range(0, number_of_training_data, batch_size),
range(batch_size, number_of_training_data, batch_size)):
iteration = iteration + 1
if epoch == 0 and counter == 0:
print("trainX[start:end]:", train_X[start:end],
"train_X.shape:", train_X.shape)
feed_dict = {
model.input_x: train_X[start:end],
model.input_y: train_Y[start:end],
model.dropout_keep_prob: FLAGS.dropout_keep_prob
}
current_loss, lr, l2_loss, _ = sess.run([
model.loss_val, model.learning_rate, model.l2_loss,
model.train_op
], feed_dict)
loss_total, counter = loss_total + current_loss, counter + 1
if counter % 30 == 0:
print(
"Learning rate:%.7f\tLoss:%.3f\tCurrent_loss:%.3f\tL2_loss%.3f\t"
% (lr, float(loss_total) / float(counter),
current_loss, l2_loss))
#if start!=0 and start%(1000*FLAGS.batch_size)==0:
# loss_valid, f1_macro_valid, f1_micro_valid= do_eval(sess, model, valid,num_classes,label2index)
# f1_score_valid=((f1_macro_valid+f1_micro_valid)/2.0) #*100.0
# print("Valid.Epoch %d ValidLoss:%.3f\tF1_score_valid:%.3f\tMacro_f1:%.3f\tMicro_f1:%.3f\t" % (epoch, loss_valid, f1_score_valid, f1_macro_valid, f1_micro_valid))
# save model to checkpoint
# if f1_score_valid>score_best:
# save_path = FLAGS.ckpt_dir_save + "model.ckpt"
# print("going to save check point.")
# saver.save(sess, save_path, global_step=epoch)
# score_best=f1_score_valid
#epoch increment
print("going to increment epoch counter....")
sess.run(model.epoch_increment)
# 4.validation
print(epoch, FLAGS.validate_every,
(epoch % FLAGS.validate_every == 0))
if epoch % FLAGS.validate_every == 0:
# loss_valid,f1_macro_valid2,f1_micro_valid2=do_eval(sess,model,valid,num_classes,label2index)
# f1_score_valid2 = ((f1_macro_valid2 + f1_micro_valid2) / 2.0) #* 100.0
# print("Valid.Epoch %d ValidLoss:%.3f\tF1 score:%.3f\tMacro_f1:%.3f\tMicro_f1:%.3f\t"% (epoch,loss_valid,f1_score_valid2,f1_macro_valid2,f1_micro_valid2))
print("epoch %d", epoch)
do_eval_acc(sess, model, valid, num_classes, label2index)
#save model to checkpoint
# if f1_score_valid2 > score_best:
# save_path=FLAGS.ckpt_dir_save+"model.ckpt"
# print("going to save check point.")
# saver.save(sess,save_path,global_step=epoch)
# score_best = f1_score_valid2
if (epoch == 2 or epoch == 4 or epoch == 6 or epoch == 9
or epoch == 13):
for i in range(1):
print(i, "Going to decay learning rate by half.")
sess.run(model.learning_rate_decay_half_op)
# 5.report on test set
#loss_test, f1_macro_test, f1_micro_test=do_eval(sess, model, test,num_classes, label2index)
#f1_score_test=((f1_macro_test + f1_micro_test) / 2.0) * 100.0
#print("Test.Epoch %d TestLoss:%.3f\tF1_score:%.3f\tMacro_f1:%.3f\tMicro_f1:%.3f\t" % (epoch, loss_test, f1_score_test,f1_macro_test, f1_micro_test))
print("test data ")
do_eval_acc(sess, model, test, num_classes, label2index)
print("training completed...")