def predict(self, seqs, demo=True):
if demo:
input_X = dev2vec(seqs,
word_dict=self.vocab,
max_seq_len=self.sequence_length)
else:
input_X, _ = pad_sequences(seqs)
inputs = torch.LongTensor(input_X)
outputs = self(inputs)
return torch.argmax(outputs, dim=1).numpy()
def predict_prob(self, seqs, demo=True):
if demo:
input_X = dev2vec(seqs,
word_dict=self.vocab,
max_seq_len=self.sequence_length)
else:
input_X, _ = pad_sequences(seqs)
inputs = torch.LongTensor(input_X)
outputs = torch.exp(self(inputs))
# softmax
outputs_softmax = outputs / torch.Tensor.reshape(
torch.sum(outputs, dim=1), [-1, 1])
return outputs_softmax.detach().numpy()
def predict_prob(self, sess, seqs, demo=True):
"""预测概率"""
if demo:
input_X = dev2vec(seqs,
word_dict=self.vocab,
max_seq_len=self.sequence_length)
else:
input_X, _ = pad_sequences(seqs)
possibility = sess.run(self.possibility,
feed_dict={
self.input_x: input_X,
self.dropout_kp: 1.0
})
return possibility
def predict(self, sess, seqs, demo=True):
"""预测标签"""
if demo:
input_X = dev2vec(seqs,
word_dict=self.vocab,
max_seq_len=self.sequence_length)
else:
input_X, _ = pad_sequences(seqs)
predictions = sess.run(self.predictions,
feed_dict={
self.input_x: input_X,
self.keep_prob: 1.0,
self.tst: True
})
return predictions
def train():
sents, tags = read_data()
x = dev2vec(sents, word2int, max_seq_len=FLAGS.sequence_length)
y = [tag2label[t] for t in tags]
model_path = os.path.join(MODEL_PATH, FLAGS.DEMO)
if not os.path.exists(model_path):
os.makedirs(model_path)
model_path = os.path.join(model_path, 'm.h5')
x = np.array(x)
y = np_utils.to_categorical(y)
print(x.shape, y.shape)
train_x, test_x, train_y, test_y = train_test_split(x, y, test_size=0.2, random_state=421)
lstm = Lstm(model_path, word2int, tag2label, input_length=FLAGS.sequence_length)
lstm.train(train_x, train_y, test_x, test_y)