python - 如何在 TensorFlow 中实现 Numpy where 索引？

Question

我有以下使用 numpy.where 的操作:

    mat = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.int32)
    index = np.array([[1,0,0],[0,1,0],[0,0,1]])
    mat[np.where(index>0)] = 100
    print(mat)

如何在 TensorFlow 中实现等价物？

mat = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.int32)
index = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]])
tf_mat = tf.constant(mat)
tf_index = tf.constant(index)
indi = tf.where(tf_index>0)
tf_mat[indi] = -1   <===== not allowed 

Answer 1

假设您想要创建一个带有一些替换元素的新张量，而不是更新变量，您可以这样做:

import numpy as np
import tensorflow as tf

mat = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.int32)
index = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]])
tf_mat = tf.constant(mat)
tf_index = tf.constant(index)
tf_mat = tf.where(tf_index > 0, -tf.ones_like(tf_mat), tf_mat)
with tf.Session() as sess:
    print(sess.run(tf_mat))

输出:

[[-1  2  3]
 [ 4 -1  6]
 [ 7  8 -1]]