def evaluate(self, label_list, seq_len_list, data, epoch=None):
def get_f1(pred, truth):
p_hit = 0
p_not = 0
n_hit = 0
n_not = 0
acc1 = 0
acc2 = 0
for it in zip(pred, truth):
if it[0] == it[1] and sum(it[1]):
p_hit += 1
if it[0] != it[1] and sum(it[1]):
p_not += 1
if it[0] != it[1] and not sum(it[1]):
n_not += 1
if it[0] == it[1] and not sum(it[1]):
n_hit += 1
if it[0] == it[1]:
acc1 += 1
cnt = 0
for i in range(len(it[0])):
if it[0][i] == it[1][i]:
cnt +=1
acc2 += (cnt/len(it[0]))
P = p_hit / (p_hit + p_not)
N = n_hit / (n_hit + n_not)
F1 = P * N * 2 / (P + N +0.00001)
ACC1 = acc1 / len(pred)
ACC2 = acc2 / len(pred)
return P, N, F1, ACC1, ACC2
true_label = []
pred_label = []
sent_list = []
for label_, (sent, tag) in zip(label_list, data):
p_ind = []
t_ind = []
if sum(label_) == 0 and sum(tag) == 0:
pass
else:
for i, p in enumerate(label_):
if p:
p_ind.append(i)
for i, p in enumerate(tag):
if p:
t_ind.append(i)
sent_list.append(id2sentence(sent, self.vocab)+'\n'+str(p_ind)+'VS'+str(t_ind))
true_label.append(tag)
pred_label.append(label_)
import codecs
with codecs.open('result.txt','w',encoding='utf-8') as f:
f.write('\n'.join(sent_list))
print(get_f1(pred_label, true_label))
def split_sent(self, label_list,seq_len_list, data):
sent_list = []
for label_, (sent, tag) in zip(label_list, data):
split_ind = []
for i, p in enumerate(label_):
if p:
split_ind.append(i)
resent = id2sentence(sent, self.vocab).split()
out_sent = copy.deepcopy(resent)
for ind in range(len(resent)):
if ind in split_ind:
out_sent[ind] = resent[ind] + '。'
else:
out_sent[ind] = resent[ind]
sent_list.append(''.join(out_sent))
return ' '.join(sent_list)
for target_sentence in target_batch:
target_sentence = data_helper.sentence2id(
target_sentence,
vocab=decoder_vocab,
max_sentence_length=decoder_sentence_length,
is_target=True)
target_token_ids.append(target_sentence)
batch_preds, batch_loss, _ = sess.run(
[predictions, loss, train_op],
feed_dict={
encoder_inputs: input_token_ids,
encoder_sequence_length: input_sentence_lengths,
decoder_inputs: target_token_ids,
})
epoch_loss += batch_loss
all_preds.append(batch_preds)
if epoch % 400 == 0:
print 'Epoch={}'.format(epoch)
print 'Epoch loss={}'.format(epoch_loss)
for input_batch, target_batch, batch_preds in zip(
input_batches, target_batches, all_preds):
for input_sentence, target_sentence, pred in zip(
input_batch, target_batch, batch_preds):
print '\t' + input_sentence
print '\t => ' + data_helper.id2sentence(
pred, reverse_vocab=decoder_reverse_vocab)
print '\tCorrect answer:' + target_sentence
