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]如何在TensorFlow中对标签进行编码?,tensorflow,Tensorflow,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 hotop: import tensorflow as tf original_indices = tf.constant([1, 5, 3]) depth = tf.constant(10) one_hot_encoded = tf.one_hot(
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]