一个基于keras实现seq2seq(Encoder-Decoder)的序列预测例子
NLP中常见任务的开源解决方案、数据集、工具、学习资料
- 输入序列为随机产生的整数序列,目标序列是对输入序列前三个元素进行反转后的序列,当然这只是我们自己定义的一种形式,可以自定义更复杂的场景。
输入序列 目标序列
[13, 28, 18, 7, 9, 5] [18, 28, 13]
[29, 44, 38, 15, 26, 22] [38, 44, 29]
[27, 40, 31, 29, 32, 1] [31, 40, 27]
...
from numpy import array
from numpy import argmax
from keras.utils import to_categorical
# 随机产生在(1,n_features)区间的整数序列,序列长度为n_steps_in
def generate_sequence(length, n_unique):
return [randint(1, n_unique-1) for _ in range(length)]
# 构造LSTM模型输入需要的训练数据
def get_dataset(n_in, n_out, cardinality, n_samples):
X1, X2, y = list(), list(), list()
for _ in range(n_samples):
# 生成输入序列
source = generate_sequence(n_in, cardinality)
# 定义目标序列,这里就是输入序列的前三个数据
target = source[:n_out]
target.reverse()
# 向前偏移一个时间步目标序列
target_in = [0] + target[:-1]
# 直接使用to_categorical函数进行on_hot编码
src_encoded = to_categorical(source, num_classes=cardinality)
tar_encoded = to_categorical(target, num_classes=cardinality)
tar2_encoded = to_categorical(target_in, num_classes=cardinality)
X1.append(src_encoded)
X2.append(tar2_encoded)
y.append(tar_encoded)
return array(X1), array(X2), array(y)
# one_hot解码
def one_hot_decode(encoded_seq):
return [argmax(vector) for vector in encoded_seq]
# 输入参数
n_features = 50 + 1
n_steps_in = 6
n_steps_out = 3
# 生成处理后的输入序列与目标序列,这里测试产生了一个序列样本
X1, X2, y = get_dataset(n_steps_in, n_steps_out, n_features, 1)
print(X1.shape, X2.shape, y.shape)
print('X1=%s, X2=%s, y=%s' % (one_hot_decode(X1[0]), one_hot_decode(X2[0]), one_hot_decode(y[0])))
# 结果如下:
(1, 6, 51) (1, 3, 51) (1, 3, 51)
X1=[32, 16, 12, 34, 25, 24], X2=[0, 12, 16], y=[12, 16, 32]