如何在TensorFlow中对标签进行编码？

我需要将字符串标签转换为向量，如[0，0，…，1，…，0]
据我所知，这就是所谓的热向量
我有10个类，所以有10个不同的字符串标签

有人能帮忙做正反变换吗

---

我是tensorflow的新手，所以请友好一点。

前进方向很简单，因为有

tf。one\u hot

op:

import tensorflow as tf

original_indices = tf.constant([1, 5, 3])
depth = tf.constant(10)
one_hot_encoded = tf.one_hot(indices=original_indices, depth=depth)

with tf.Session():
  print(one_hot_encoded.eval())

产出：

[[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  1.  0.  0.  0.  0.]
 [ 0.  0.  0.  1.  0.  0.  0.  0.  0.  0.]]

与此相反，使用

tf也不太糟糕。其中

查找非零索引：

def decode_one_hot(batch_of_vectors):
  """Computes indices for the non-zero entries in batched one-hot vectors.

  Args:
    batch_of_vectors: A Tensor with length-N vectors, having shape [..., N].
  Returns:
    An integer Tensor with shape [...] indicating the index of the non-zero
    value in each vector.
  """
  nonzero_indices = tf.where(tf.not_equal(
      batch_of_vectors, tf.zeros_like(batch_of_vectors)))
  reshaped_nonzero_indices = tf.reshape(
      nonzero_indices[:, -1], tf.shape(batch_of_vectors)[:-1])
  return reshaped_nonzero_indices

with tf.Session():
  print(decode_one_hot(one_hot_encoded).eval())

印刷品：

[1 5 3]