def dev_one_epoch(self, sess, dev):
"""
:param sess:
:param dev:
:return:
"""
batches = dev_batch_iter(dev, self.batch_size, shuffle=False)
true_label_list, pre_label_list = [], []
for step, dev in enumerate(batches):
dev_x, dev_p1, dev_p2, dev_y = zip(*dev)
dev_feed_dict = {
self.target_word_ids: dev_x,
self.target_position_ids1: dev_p1,
self.target_position_ids2: dev_p2,
self.relation_labels: dev_y
}
# [batch,]
label_pre = sess.run(self.labels_predict, feed_dict=dev_feed_dict)
# [batch,]
label_true = np.argmax(dev_y, axis=-1)
true_label_list.extend(label_true)
pre_label_list.extend(label_pre)
print("true label:" + str(true_label_list))
print("pre label:" + str(pre_label_list))
return true_label_list, pre_label_list
def dev_one_epoch(self, sess, dev, epoch):
"""
:param sess:
:param dev:
:return:
"""
batches = dev_batch_iter(dev, self.batch_size, shuffle=False)
true_label_list,pre_label_list = [],[]
for step, dev in enumerate(batches):
dev_x, dev_p1, dev_p2, dev_y = zip(*dev)
dev_feed_dict = {self.source_word_ids: dev_x,
self.source_position_ids1: dev_p1,
self.source_position_ids2: dev_p2,
self.relation_labels: dev_y,
self.dropout:1.0}
# [batch,]
label_pre = sess.run(self.labels_predict,feed_dict=dev_feed_dict)
#print(label_pre)
# [batch,]
label_true = np.argmax(dev_y, axis=-1)
true_label_list.extend(label_true)
#label_pre = label_pre.to_list()
pre_label_list.extend(label_pre)
print("true label:"+str(true_label_list))
print("pre label:"+str(pre_label_list))
macro_f1 = f1_score(true_label_list, pre_label_list, average="macro")
self.logger.info('epoch {}, macro_f1: {:.4}'
.format(epoch + 1,macro_f1))
def dev_one_epoch(self, sess, dev, epoch):
"""
:param sess:
:param dev:
:return:
"""
batches = dev_batch_iter(dev, self.batch_size, shuffle=False)
true_label_list, pre_label_list = [], []
dev_acc = 0
dev_los = 0
num = 0
num_batches = (len(dev) + self.batch_size - 1) // self.batch_size
for step, dev in enumerate(batches):
step_num = epoch * num_batches + step + 1
dev_x, dev_p1, dev_p2, dev_y = zip(*dev)
dev_feed_dict = {
self.target_word_ids: dev_x,
self.target_position_ids1: dev_p1,
self.target_position_ids2: dev_p2,
self.relation_labels: dev_y,
self.dropout_kp: 1.0
}
# [batch,]
label_pre, dev_loss, dev_accuracy, summary = sess.run(
[
self.labels_predict, self.dev_loss, self.dev_accuracy,
self.dev_merged
],
feed_dict=dev_feed_dict)
# [batch,]
label_true = np.argmax(dev_y, axis=-1)
self.file_writer.add_summary(summary, step_num)
print(label_pre)
true_label_list.extend(label_true)
pre_label_list.extend(label_pre)
dev_acc += dev_accuracy
dev_los += dev_loss
num += 1
dev_acc = dev_acc / (num + 1.0)
dev_los = dev_los / (num + 1.0)
print("true label:" + str(true_label_list))
print("pre label:" + str(pre_label_list))
macro_f1 = f1_score(true_label_list, pre_label_list, average="macro")
self.logger.info(
'epoch {}, dev_loss: {:.4}, dev_accuracy: {:.4},macro_f1: {:.4}'.
format(epoch + 1, dev_los, dev_acc, macro_f1))
return macro_f1