def train(self, sess, save_file, X_train, y_train, X_val, y_val):
count = 0
max_f = -1
writer = tf.summary.FileWriter('./graphs_ar', sess.graph)
trainloss = tf.summary.scalar("training_loss", self.ner_model.loss)
for epoch in range(self.num_epochs):
print "current epoch: %d" % (epoch)
for iteration in range(self.num_iterations):
print "this iteration is %d"%(iteration)
X_train_batch, y_train_batch = helper.nextRandomBatch(X_train, y_train, batch_size=self.batch_size)
transition_batch = helper.getTransition(y_train_batch)
_, train_loss, loss_train, max_scores, max_scores_pre, length =\
sess.run([
self.ner_model.optimizer,
trainloss,
self.ner_model.loss,
self.ner_model.max_scores,
self.ner_model.max_scores_pre,
self.ner_model.length,
],
feed_dict={
self.ner_model.input_data:X_train_batch,
self.ner_model.targets:y_train_batch,
self.ner_model.targets_transition:transition_batch
})
print "the loss : %f"%loss_train#, X_train_batch
writer.add_summary(train_loss, epoch*self.num_iterations+iteration)
if iteration % 100 == 0:
presicion_loc, recall_loc, f_loc, presicion_org, recall_org, f_org, presicion_per, recall_per, f_per = \
self.validationBatch(sess, X_val, y_val)
print "iteration: %5d, valid , valid precision: LOC %.5f, ORG %.5f, PER %.5f, valid recall: LOC %.5f, ORG %.5f, PER %.5f, valid f1: LOC %.5f, ORG %.5f, PER %.5f" %\
(iteration, presicion_loc, presicion_org, presicion_per, recall_loc, recall_org, recall_per, f_loc, f_org, f_per)
if f_loc + f_org + f_per >= max_f:
max_f = f_loc + f_org + f_per
saver = tf.train.Saver()
save_path = saver.save(sess, save_file)
print "saved the best model with f1: %.5f" % (max_f / 3.0)
self.last_f = f_loc + f_org + f_per
def train(self, sess, save_file, X_train, y_train, X_val, y_val):
saver = tf.train.Saver()
char2id, id2char = helper.loadMap("char2id")
label2id, id2label = helper.loadMap("label2id")
merged = tf.contrib.deprecated.merge_all_summaries()
summary_writer_train = tf.contrib.summary.SummaryWriter(
'loss_log/train_loss', sess.graph)
summary_writer_val = tf.contrib.summary.SummaryWriter(
'loss_log/val_loss', sess.graph)
num_iterations = int(math.ceil(1.0 * len(X_train) / self.batch_size))
cnt = 0
for epoch in range(self.num_epochs):
# shuffle train in each epoch
sh_index = np.arange(len(X_train))
np.random.shuffle(sh_index)
X_train = X_train[sh_index]
y_train = y_train[sh_index]
print("current epoch: %d" % (epoch))
for iteration in range(num_iterations):
# train
X_train_batch, y_train_batch = helper.nextBatch(
X_train,
y_train,
start_index=iteration * self.batch_size,
batch_size=self.batch_size)
y_train_weight_batch = 1 + np.array(
(y_train_batch == label2id['B']) |
(y_train_batch == label2id['E']), float)
transition_batch = helper.getTransition(y_train_batch)
_, loss_train, max_scores, max_scores_pre, length, train_summary = \
sess.run([
self.optimizer,
self.loss,
self.max_scores,
self.max_scores_pre,
self.length,
self.train_summary
],
feed_dict={
self.targets_transition: transition_batch,
self.inputs: X_train_batch,
self.targets: y_train_batch,
self.targets_weight: y_train_weight_batch
})
predicts_train = self.viterbi(max_scores,
max_scores_pre,
length,
predict_size=self.batch_size)
if iteration % 10 == 0:
cnt += 1
precision_train, recall_train, f1_train = self.evaluate(
X_train_batch, y_train_batch, predicts_train, id2char,
id2label)
summary_writer_train.add_summary(train_summary, cnt)
print(
"iteration: %5d, train loss: %5d, train precision: %.5f, train recall: %.5f, train f1: %.5f"
% (iteration, loss_train, precision_train,
recall_train, f1_train))
# validation
if iteration % 100 == 0:
X_val_batch, y_val_batch = helper.nextRandomBatch(
X_val, y_val, batch_size=self.batch_size)
y_val_weight_batch = 1 + np.array(
(y_val_batch == label2id['B']) |
(y_val_batch == label2id['E']), float)
transition_batch = helper.getTransition(y_val_batch)
loss_val, max_scores, max_scores_pre, length, val_summary = \
sess.run([
self.loss,
self.max_scores,
self.max_scores_pre,
self.length,
self.val_summary
],
feed_dict={
self.targets_transition: transition_batch,
self.inputs: X_val_batch,
self.targets: y_val_batch,
self.targets_weight: y_val_weight_batch
})
predicts_val = self.viterbi(max_scores,
max_scores_pre,
length,
predict_size=self.batch_size)
precision_val, recall_val, f1_val = self.evaluate(
X_val_batch, y_val_batch, predicts_val, id2char,
id2label)
summary_writer_val.add_summary(val_summary, cnt)
print(
"iteration: %5d, valid loss: %5d, valid precision: %.5f, valid recall: %.5f, valid f1: %.5f"
% (iteration, loss_val, precision_val, recall_val,
f1_val))
if f1_val > self.max_f1:
self.max_f1 = f1_val
save_path = saver.save(sess, save_file)
print("saved the best model with f1: %.5f" %
(self.max_f1))
def train(self, sess, save_file, train_data, val_data):
saver = tf.train.Saver(max_to_keep=3)
#train data
X_train = train_data['char']
X_left_train = train_data['left']
X_right_train = train_data['right']
X_pos_train = train_data['pos']
X_lpos_train = train_data['lpos']
X_rpos_train = train_data['rpos']
X_rel_train = train_data['rel']
X_dis_train = train_data['dis']
y_train = train_data['label']
#dev data
X_val = val_data['char']
X_left_val = val_data['left']
X_right_val = val_data['right']
X_pos_val = val_data['pos']
X_lpos_val = val_data['lpos']
X_rpos_val = val_data['rpos']
X_rel_val = val_data['rel']
X_dis_val = val_data['dis']
y_val = val_data['label']
#dictionary
char2id, id2char = helper.loadMap("char2id")
pos2id, id2pos = helper.loadMap("pos2id")
label2id, id2label = helper.loadMap("label2id")
merged = tf.summary.merge_all()
summary_writer_train = tf.summary.FileWriter('loss_log/train_loss', sess.graph)
summary_writer_val = tf.summary.FileWriter('loss_log/val_loss', sess.graph)
num_iterations = int(math.ceil(1.0 * len(X_train) / self.batch_size))
cnt = 0
for epoch in range(self.num_epochs):
# shuffle train in each epoch
sh_index = np.arange(len(X_train))
np.random.shuffle(sh_index)
X_train = X_train[sh_index]
X_left_train = X_left_train[sh_index]
X_right_train = X_right_train[sh_index]
X_pos_train = X_pos_train[sh_index]
X_lpos_train = X_lpos_train[sh_index]
X_rpos_train = X_rpos_train[sh_index]
X_rel_train = X_rel_train[sh_index]
X_dis_train = X_dis_train[sh_index]
y_train = y_train[sh_index]
train_data['char'] = X_train
train_data['left'] = X_left_train
train_data['right'] = X_right_train
train_data['pos'] = X_pos_train
train_data['lpos'] = X_lpos_train
train_data['rpos'] = X_rpos_train
train_data['rel'] = X_rel_train
train_data['dis'] = X_dis_train
train_data['label'] = y_train
print "current epoch: %d" % (epoch)
for iteration in range(num_iterations):
# train
#get batch
train_batches = helper.nextBatch(train_data, start_index=iteration * self.batch_size, batch_size=self.batch_size)
X_train_batch = train_batches['char']
X_left_train_batch = train_batches['left']
X_right_train_batch = train_batches['right']
X_pos_train_batch = train_batches['pos']
X_lpos_train_batch = train_batches['lpos']
X_rpos_train_batch = train_batches['rpos']
X_rel_train_batch = train_batches['rel']
X_dis_train_batch = train_batches['dis']
y_train_batch = train_batches['label']
# feed batch to model and run
_, loss_train, length, train_summary, logits, trans_params =\
sess.run([
self.optimizer,
self.loss,
self.length,
self.train_summary,
self.logits,
self.trans_params,
],
feed_dict={
self.inputs:X_train_batch,
self.lefts:X_left_train_batch,
self.rights:X_right_train_batch,
self.poses:X_pos_train_batch,
self.lposes:X_lpos_train_batch,
self.rposes:X_rpos_train_batch,
self.rels:X_rel_train_batch,
self.dises:X_dis_train_batch,
self.targets:y_train_batch
# self.targets_weight:y_train_weight_batch
})
# print (len(length))
#get predict f1
predicts_train = self.viterbi(logits, trans_params, length, predict_size=self.batch_size)
if iteration > 0 and iteration % 10 == 0:
cnt += 1
hit_num, pred_num, true_num = self.evaluate(y_train_batch, predicts_train, id2char, id2label)
precision_train, recall_train, f1_train = self.caculate(hit_num, pred_num, true_num)
summary_writer_train.add_summary(train_summary, cnt)
print "iteration: %5d/%5d, train loss: %5d, train precision: %.5f, train recall: %.5f, train f1: %.5f" % (iteration, num_iterations, loss_train, precision_train, recall_train, f1_train)
# a batch in validation
if iteration > 0 and iteration % 100 == 0:
val_batches = helper.nextRandomBatch(val_data, batch_size=self.batch_size)
X_val_batch = val_batches['char']
X_left_val_batch = val_batches['left']
X_right_val_batch = val_batches['right']
X_pos_val_batch = val_batches['pos']
X_lpos_val_batch = val_batches['lpos']
X_rpos_val_batch = val_batches['rpos']
X_rel_val_batch = val_batches['rel']
X_dis_val_batch = val_batches['dis']
y_val_batch = val_batches['label']
loss_val, length, val_summary, logits, trans_params =\
sess.run([
self.loss,
self.length,
self.val_summary,
self.logits,
self.trans_params,
],
feed_dict={
self.inputs:X_val_batch,
self.lefts:X_left_val_batch,
self.rights:X_right_val_batch,
self.poses:X_pos_val_batch,
self.lposes:X_lpos_val_batch,
self.rposes:X_rpos_val_batch,
self.rels:X_rel_val_batch,
self.dises:X_dis_val_batch,
self.targets:y_val_batch
# self.targets_weight:y_val_weight_batch
})
predicts_val = self.viterbi(logits, trans_params, length, predict_size=self.batch_size)
hit_num, pred_num, true_num = self.evaluate(y_val_batch, predicts_val, id2char, id2label)
precision_val, recall_val, f1_val = self.caculate(hit_num, pred_num, true_num)
summary_writer_val.add_summary(val_summary, cnt)
print "iteration: %5d, valid loss: %5d, valid precision: %.5f, valid recall: %.5f, valid f1: %.5f" % (iteration, loss_val, precision_val, recall_val, f1_val)
# calc f1 for the whole dev set
if epoch > 0 and iteration == num_iterations -1:
num_val_iterations = int(math.ceil(1.0 * len(X_val) / self.batch_size))
preds_lines = []
for val_iteration in range(num_val_iterations):
val_batches = helper.nextBatch(val_data, start_index=val_iteration * self.batch_size, batch_size=self.batch_size)
X_val_batch = val_batches['char']
X_left_val_batch = val_batches['left']
X_right_val_batch = val_batches['right']
X_pos_val_batch = val_batches['pos']
X_lpos_val_batch = val_batches['lpos']
X_rpos_val_batch = val_batches['rpos']
X_rel_val_batch = val_batches['rel']
X_dis_val_batch = val_batches['dis']
y_val_batch = val_batches['label']
loss_val, length, val_summary, logits, trans_params =\
sess.run([
self.loss,
self.length,
self.val_summary,
self.logits,
self.trans_params,
],
feed_dict={
self.inputs:X_val_batch,
self.lefts:X_left_val_batch,
self.rights:X_right_val_batch,
self.poses:X_pos_val_batch,
self.lposes:X_lpos_val_batch,
self.rposes:X_rpos_val_batch,
self.rels:X_rel_val_batch,
self.dises:X_dis_val_batch,
self.targets:y_val_batch
# self.targets_weight:y_val_weight_batch
})
predicts_val = self.viterbi(logits, trans_params, length, predict_size=self.batch_size)
preds_lines.extend(predicts_val)
preds_lines = preds_lines[:len(y_val)]
recall_val, precision_val, f1_val, errors = helper.calc_f1(preds_lines, id2label, 'cpbdev.txt', 'validation.out')
if f1_val > self.max_f1:
self.max_f1 = f1_val
save_path = saver.save(sess, save_file)
helper.calc_f1(preds_lines, id2label, 'cpbdev.txt', 'validation.out.best')
print "saved the best model with f1: %.5f" % (self.max_f1)
print "valid precision: %.5f, valid recall: %.5f, valid f1: %.5f, errors: %5d" % (precision_val, recall_val, f1_val, errors)
def train(self, sess, save_file, X_train, y_train, X_val, y_val):
saver = tf.train.Saver()
char2id, id2char = helper.loadMap("char2id")
label2id, id2label = helper.loadMap("label2id")
merged = tf.merge_all_summaries()
summary_writer_train = tf.train.SummaryWriter('loss_log/train_loss', sess.graph)
summary_writer_val = tf.train.SummaryWriter('loss_log/val_loss', sess.graph)
num_iterations = int(math.ceil(1.0 * len(X_train) / self.batch_size))
cnt = 0
for epoch in range(self.num_epochs):
# shuffle train in each epoch
sh_index = np.arange(len(X_train))
np.random.shuffle(sh_index)
X_train = X_train[sh_index]
y_train = y_train[sh_index]
print "current epoch: %d" % (epoch)
for iteration in range(num_iterations):
# train
X_train_batch, y_train_batch = helper.nextBatch(X_train, y_train, start_index=iteration * self.batch_size, batch_size=self.batch_size)
y_train_weight_batch = 1 + np.array((y_train_batch == label2id['B']) | (y_train_batch == label2id['E']), float)
transition_batch = helper.getTransition(y_train_batch)
_, loss_train, max_scores, max_scores_pre, length, train_summary =\
sess.run([
self.optimizer,
self.loss,
self.max_scores,
self.max_scores_pre,
self.length,
self.train_summary
],
feed_dict={
self.targets_transition:transition_batch,
self.inputs:X_train_batch,
self.targets:y_train_batch,
self.targets_weight:y_train_weight_batch
})
predicts_train = self.viterbi(max_scores, max_scores_pre, length, predict_size=self.batch_size)
if iteration % 10 == 0:
cnt += 1
precision_train, recall_train, f1_train = self.evaluate(X_train_batch, y_train_batch, predicts_train, id2char, id2label)
summary_writer_train.add_summary(train_summary, cnt)
print "iteration: %5d, train loss: %5d, train precision: %.5f, train recall: %.5f, train f1: %.5f" % (iteration, loss_train, precision_train, recall_train, f1_train)
# validation
if iteration % 100 == 0:
X_val_batch, y_val_batch = helper.nextRandomBatch(X_val, y_val, batch_size=self.batch_size)
y_val_weight_batch = 1 + np.array((y_val_batch == label2id['B']) | (y_val_batch == label2id['E']), float)
transition_batch = helper.getTransition(y_val_batch)
loss_val, max_scores, max_scores_pre, length, val_summary =\
sess.run([
self.loss,
self.max_scores,
self.max_scores_pre,
self.length,
self.val_summary
],
feed_dict={
self.targets_transition:transition_batch,
self.inputs:X_val_batch,
self.targets:y_val_batch,
self.targets_weight:y_val_weight_batch
})
predicts_val = self.viterbi(max_scores, max_scores_pre, length, predict_size=self.batch_size)
precision_val, recall_val, f1_val = self.evaluate(X_val_batch, y_val_batch, predicts_val, id2char, id2label)
summary_writer_val.add_summary(val_summary, cnt)
print "iteration: %5d, valid loss: %5d, valid precision: %.5f, valid recall: %.5f, valid f1: %.5f" % (iteration, loss_val, precision_val, recall_val, f1_val)
if f1_val > self.max_f1:
self.max_f1 = f1_val
save_path = saver.save(sess, save_file)
print "saved the best model with f1: %.5f" % (self.max_f1)
correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
# Initializing the variables
init = tf.global_variables_initializer()
# Launch the graph
with tf.Session() as sess:
sess.run(init)
step = 0
# Keep training until reach max iterations
while step < training_iters:
# batch_x, batch_y = mnist.train.next_batch(batch_size)
batch_x, batch_y = helper.nextRandomBatch(X_train,
y_train,
batch_size=125)
batch_y = batch_y[:, 2]
tmp_y = []
for i in range(0, batch_y.shape[0]):
if batch_y[i] == 2:
tmp_y.append([0, 1])
else:
tmp_y.append([1, 0])
tmp_y = np.array(tmp_y)
batch_y = tmp_y
# Run optimization op (backprop)
sess.run(optimizer, feed_dict={x1: batch_x, y: batch_y})
if step % display_step == 0:
# Calculate batch accuracy
def train(self, sess, save_file, X_train, y_train, X_val, y_val):
saver = tf.train.Saver()
summary_writer_train = tf.summary.FileWriter('loss_log/train_loss',
sess.graph)
summary_writer_val = tf.summary.FileWriter('loss_log/val_loss',
sess.graph)
num_iterations = int(math.ceil(1.0 * len(X_train) / self.batch_size))
for epoch in range(self.num_epochs):
# shuffle train in each epoch
sh_index = np.arange(len(X_train))
np.random.shuffle(sh_index)
X_train = X_train[sh_index]
y_train = y_train[sh_index]
print("current epoch: %d" % (epoch))
for iteration in range(num_iterations):
# train
X_train_batch1, X_train_batch2, y_train_batch = helper.nextBatch(
X_train, y_train, iteration * self.batch_size,
self.batch_size)
_, train_loss, train_acc, train_summary = sess.run(
[
self.optimizer,
self.loss,
# self.predictions,
self.accuracy,
self.summary_op
],
feed_dict={
self.input1: X_train_batch1,
self.input2: X_train_batch2,
self.targets: y_train_batch
})
if iteration % 20 == 0:
# train_acc = helper.extractSense(y_train_batch, train_y_hat)
summary_writer_train.add_summary(train_summary, iteration)
print(
"iteration: %5d, train loss: %5d, train precision: %.5f"
% (iteration, train_loss, train_acc))
# validation
if iteration % 20 == 0:
X_val_batch1, X_val_batch2, y_val_batch = helper.nextRandomBatch(
X_val, y_val, self.batch_size)
dev_loss, dev_acc, val_summary = sess.run(
[
self.loss,
# self.predictions,
self.accuracy,
self.summary_op
],
feed_dict={
self.input1: X_val_batch1,
self.input2: X_val_batch2,
self.targets: y_val_batch
})
# test_acc = helper.extractSense(y_val_batch, dev_y_hat)
summary_writer_val.add_summary(val_summary, iteration)
print(
"iteration: %5d, dev loss: %5d, dev precision: %.5f" %
(iteration, dev_loss, dev_acc))
if dev_acc > self.max_acc:
self.max_acc = dev_acc
saver.save(sess, save_file)
print("saved the best model with accuracy: %.5f" %
(self.max_acc))
